Pith. sign in

REVIEW 3 major objections 5 minor 300 references

Learning in Deep Networks under Dale's Constraint

T0 review · 3 major / 5 minor · reviewed 2026-08-10 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗

Pith's one-line read A network whose neurons and synapses never change sign can still learn exactly like backpropagation.

desk verdict Central theorem is unproven as written, but the architecture and experiments are solid enough to warrant a careful review. read the letter →

arxiv 2608.06963 v1 pith:VITRPYWO submitted 2026-08-07 cs.AI

Roy Abel , Shimon Ullman This is my paper · ORCID
classification cs.AI
keywords Dale'slawon-offneuralcircuitbiologicallyplausiblelearningbackpropagationHebbiancreditassignmentnon-negativeactivityTinyImageNet
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

This paper asks whether a deep network can learn by gradient descent while respecting three biological constraints: neurons never fire negative values, each synapse keeps a fixed excitatory or inhibitory sign (Dale's law), and every weight update is local. It answers with an on-off circuit motif repeated through both bottom-up and top-down pathways, where every signed quantity is carried by two non-negative channels and the sign is recovered as their difference. The paper proves that when bottom-up and top-down connectivity are aligned, the local Hebbian update equals the exact backpropagation update, even though all propagated error signals are non-negative. It then shows empirically that the constrained architecture matches or beats comparable vanilla ReLU networks on MNIST, Fashion-MNIST, CIFAR-10, and Tiny ImageNet. If correct, this supplies a concrete circuit-level account of how a cortex-like network could implement gradient credit assignment without violating Dale's constraint.

What carries the argument

The on-off motif: a small excitatory-inhibitory circuit that takes two non-negative inputs $x_1,x_2$ and outputs two non-negative channels $y^+ = \mathrm{ReLU}_+(x_1 - x_2 - \theta)$ and $y^- = \mathrm{ReLU}_-(x_2 - x_1 - \theta)$, so at most one channel is active and the signed value is the difference $y^+ - y^-$. The motif is the repeating unit of both the bottom-up stream, which computes activations, and the top-down stream, which propagates feedback; in the top-down stream each pair of channels acts as an effective signed error, gated by the active bottom-up channel, and paired channels drive Hebbian increases versus decreases. The local Hebbian rule updates corresponding bottom-up and top-down synapses identically, which preserves alignment and makes the effective top-down difference follow the backpropagation recursion.

What would settle it

Train the symmetric on-off model on a small fully connected ReLU network, record the effective top-down signal (the difference of the two non-negative channels) at every layer, and compare it with the exact backpropagation error computed from the same loss; if the mean absolute difference does not shrink to the optimizer's numerical tolerance, the claimed exact recovery fails.

Watch

Extended reading notes

Core claim

The central claim is that signed credit assignment does not require signed activity. A small excitatory-inhibitory motif computes the rectified difference of two non-negative inputs, producing an On channel and an Off channel with at most one active; the difference of the channels is the represented signed value. The same motif is used top-down, with feedback carried by two non-negative populations whose difference is the effective error signal, one driving Hebbian increases and the other decreases. Under symmetric bottom-up and top-down weights, the propagated difference satisfies the backpropagation recursion, so the local update $\Delta W_l = \eta(\bar\partial_l - \partial_l) h_{l-1}^T$ equals the gradient descent step $-\eta\,\partial L/\partial W_l$. The paper further reports that the on-off architecture reaches 42.31% top-1 accuracy on Tiny ImageNet, ahead of matched vanilla convolutional baselines and of the forward-forward baseline it compares against.

Load-bearing premise

The load-bearing premise is that the top-down on-off wiring can send each non-negative error channel backwards through the same connection strengths used in the forward pass, with the two channels kept separate and every neuron staying non-negative; the paper assumes such a circuit can be built, and for a general signed weight matrix that requirement is in tension with fixed-sign synapses.

Editorial extensions

If this is right

  • If the theorem holds, a cortical circuit obeying Dale's law could implement exact gradient-based credit assignment without any neuron representing a negative value.
  • The matched Hebbian updates give a weight-transport-free scheme: bottom-up and top-down weights stay aligned automatically once initialized close, so explicit weight transport is not needed.
  • The Tiny ImageNet results imply that paired ReLU+/ReLU- channels are not just a cost of biological realism; comparable-width vanilla networks trained with backpropagation can be out-performed by the on-off Hebbian model.
  • Weak symmetry and noisy local updates degrade performance only slightly, suggesting approximate biological wiring and stochastic plasticity suffice for most of the benefit.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • A consequence the paper leaves implicit: the separation into potentiation-driving and depression-driving top-down populations maps naturally onto dendritic compartments with opposing plasticity signals, so the model offers concrete predictions for where to look for opponent error signals in cortex.
  • The empirical advantage over vanilla networks suggests a testable extension: on-off coding should help most on contrast- or edge-rich image tasks and least on tasks where positive and negative features are roughly symmetric; measuring that gradient could separate the representation benefit from the biological constraint.
  • The exact-recovery theorem hinges on the realizability of the transposed signed-weight propagation in non-negative channels; if that realizability fails for general signed matrices, the architecture would likely still function as feedback alignment in disguise, which the weak-symmetry experiments already anticipate.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

3 major / 5 minor

Summary. The paper proposes an on-off neural circuit motif that represents signed quantities as differences of two non-negative channels, and uses it in both bottom-up and top-down pathways with a local Hebbian learning rule. The authors claim that under symmetric bottom-up/top-down weights, the learning rule exactly recovers the backpropagation update despite honoring Dale's constraint, and they report experiments on MNIST, Fashion-MNIST, CIFAR-10, and Tiny ImageNet showing competitive or superior accuracy. The theoretical proof underlying the exact-recovery claim is given in Appendix A.4.

Significance. If the theoretical claim were established, the paper would make a notable contribution to biologically plausible credit assignment by demonstrating that Dale-constrained, non-negative circuits can implement backpropagation-like learning. The empirical results, especially the Tiny ImageNet gains over vanilla baselines, are interesting and suggest that the paired-channel representation may have computational benefits. However, the central theoretical contribution is not supported by the proof as written, so the paper's significance currently rests largely on its empirical findings.

major comments (3)
  1. [Appendix A.4, Eqs. (12)–(13); Theorem 1] The proof assumes the top-down propagation equations (12)–(13) rather than deriving them from the Dale-constrained circuit described in Section 3.3 and Appendix A.1.3. For a general signed effective weight matrix W_l = A_l − B_l with A_l, B_l ≥ 0, exact propagation of a signed error δ = δ⁺ − δ⁻ through W_lᵀ requires cross-coupling: the positive TD channel must receive A_lᵀδ⁺ + B_lᵀδ⁻ and the negative TD channel must receive B_lᵀδ⁺ + A_lᵀδ⁻. Equations (12)–(13) contain only same-sign routes (W_l⁺ᵀδ⁺ and W_l⁻ᵀδ⁻), so their difference is not W_lᵀδ in general. The subsequent identification W_l = W_l⁺ᵀ = W_l⁻ᵀ forces the TD matrices to be signed, contradicting Dale's constraint stated in Section 3. This invalidates Theorem 1 and the abstract's claim that the learning scheme exactly recovers the backpropagation update.
  2. [Section 4.2 and Appendix A.4] The theorem is circular with respect to the architecture: its assumption that the top-down pathway uses the transposed on-off channel structure in Equations (12)–(13) is precisely the property that needs to be established from the circuit. The induction proof then only shows that if the top-down difference matches backpropagation at layer l and the TD channels propagate via the signed matrix transpose, the difference matches at layer l−1. It does not show that the on-off circuit implements that propagation. Thus the proof reduces to assuming the conclusion that the paper claims to derive.
  3. [Section 3.2 and Appendix A.4] The relationship between the effective signed weight matrix W_l and the actual non-negative channel-level weights is never defined. The Hebbian update in Eqs. (2)–(3) is stated at the level of effective weights, but the Daleian architecture has separate channel-level synaptic weights whose update rules are not specified in matrix form. Without this mapping, the equivalence ΔW_l = η δ_l h_{l−1}ᵀ in Eq. (17) is not connected to the physical sign-constrained weights that the paper claims to update.
minor comments (5)
  1. [Appendix A.4] The proof of Theorem 1 alternates between W_l⁺ᵀ and W_lᵀ without explanation; please harmonize the notation consistently throughout the proof.
  2. [Section 4.2] The line "W_lᵀ = W_l⁺ = W_l⁻" is not well-defined and implies W_l is symmetric; please clarify the intended channel-level connectivity and whether the top-down weights are transposes of the bottom-up weights.
  3. [Appendix A.4 and Section 4.2] The gate is denoted D_{l−1} = diag(σ′(z_{l−1})) in the appendix but Gate h_{l−1} in the main text; please use consistent notation and clarify whether the gate uses the pre-activation or the post-activation, since for ReLU the derivative is applied to the pre-activation.
  4. [NeurIPS Checklist, items 5 and 13] Item 5 states that code is provided in the supplemental material, while item 13 says a detailed git page will be published; please make the availability statement consistent.
  5. [Table 2] The entry "Hebb-based [Lagani et al., 2021] –37.0" is ambiguous; use an explicit dash or placeholder for the missing top-1 value.

Circularity Check

1 steps flagged · score 7.0 of 10

The exact-recovery theorem assumes the backpropagation recursion in the TD channels (Eqs. 12–13) rather than deriving it from the Dale-constrained circuit, so the central theoretical claim reduces by construction.

  1. self definitional [Appendix A.4, Theorem 1 (Eqs. (12)–(13))]
    "Assume that the top-down pathway uses the transposed on-off channel structure in Equations(12)–(13), and that the bottom-up activity gates the top-down propagation according to the ReLU derivative. Then the effective top-down signal δl = ¯∂l −∂ l (15) satisfies the same recursion as backpropagation."

    The theorem's stated assumption is already the desired recursion. Equations (12)–(13) define the non-negative TD channels as D_{l-1}(W_l^{+T} ∂̄_l) and D_{l-1}(W_l^{-T} ∂_l). With the symmetric-connectivity assumption W_l = W_l^{+T} = W_l^{-T}, subtracting gives D_{l-1} W_l^T(∂̄_l − ∂_l) = D_{l-1} W_l^T δ_l, exactly the backpropagation recursion (5). The proof thus shows only that if the TD stream already implements transpose-weight, ReLU-gated propagation of the effective error, then it follows backpropagation. The Dale-constrained realizability of Eqs. (12)–(13) is not derived; for general signed W_l = A − B, non-negative channel propagation requires cross terms (A^T δ+ + B^T δ− and B^T δ+ + A^T δ−), which Eqs. (12)–(13) omit. The recovery is an assumption restated as a conclusion.

full rationale

The central theoretical claim is circular in the narrow sense: Theorem 1 in Appendix A.4 assumes exactly the transposed, ReLU-gated top-down recursion that backpropagation uses, and the symmetric weight condition W_l = W_l^{+T} = W_l^{-T} makes the difference of Eqs. (12)–(13) equal to the backpropagation recursion (5). The proof therefore reduces to 'if the TD stream computes D W^T δ, then the TD stream computes D W^T δ.' The paper does not derive Eqs. (12)–(13) from the on-off circuit; Appendix A.1.3 describes cross-channel connectivity but supplies no equations connecting that circuit to (12)–(13). Moreover, for a general signed effective weight matrix W_l = A − B with A,B ≥ 0, the non-negative-channel propagation of W_l^T δ requires cross terms A^T δ^+ + B^T δ^- in the positive channel and B^T δ^+ + A^T δ^- in the negative channel; Eqs. (12)–(13) contain only same-sign routes, so the assumed structure is not realizable by non-negative channels unless the effective weights are non-negative. The empirical Tiny ImageNet comparisons are self-contained and not circular, and the self-citations (Abel and Ullman 2024) are not load-bearing for the theorem. Score 7 reflects that the central 'prediction' of backpropagation recovery is built into the definition of the top-down propagation, while the architectural and experimental content remains independent.

Assumptions & free parameters 1 free parameters · 4 assumptions · 2 invented entities

The central claim depends on the ad hoc assumption that the top-down channels propagate independently through W_l^T (Eqs. 12-13), which is not derived from the circuit. It also relies on the assumption that BU/TD alignment is preserved. These are the main unproved inputs.

free parameters (1)
  • Threshold theta in ReLU+- activation = 0
    Set to zero in all experiments; a design choice, not fitted to data, but it controls the motif's operation.
assumptions (4)
  • ad hoc to paper The top-down propagation satisfies equations (12)-(13), where each non-negative channel propagates through the transpose of the effective weight matrix independently.
    This is the key assumption that makes the backpropagation-recovery proof work; it is stated without derivation from the on-off circuit and is not realizable with non-negative weights for a general signed effective W_l.
  • domain assumption Bottom-up and top-down weights remain exactly aligned throughout learning if initialized aligned.
    The paper argues that matched Hebbian updates preserve alignment; this is plausible but not proven in detail.
  • standard math The gating operation in top-down propagation equals the derivative of ReLU.
    This is a standard fact about ReLU networks, used in many bio-plausible learning analyses.
  • domain assumption The on-off motif's difference computation implements signed linear transformations with non-negative synaptic weights.
    The paper assumes the motif can represent arbitrary signed weight matrices through channel differences; this is plausible for the bottom-up direction but not reconciled with the top-down proof.
invented entities (2)
  • On-off neural circuit motif
    purpose: Represents signed values as pairs of non-negative channels (ReLU+ and ReLU-) using excitatory and inhibitory neurons.
    The motif is inspired by retinal ON/OFF pathways but is a new computational construct; the paper does not provide independent falsifiable predictions about biology.
  • Paired top-down populations (positive-update and negative-update TD channels)
    purpose: Carry signed error signals as differences of two non-negative populations for Hebbian updates.
    This is a mechanism invented for the model; the cited biological evidence (e.g., Francioni et al.) supports vectorized instructive signals, but not this exact two-population structure.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Learning in Deep Networks under Dale's Constraint." pith.science (2026). https://pith.science/paper/VITRPYWO

@misc{pith2026260806963,
  author       = {Pith},
  title        = {Pith review of: Learning in Deep Networks under Dale's Constraint},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/VITRPYWO}},
  note         = {Machine review of arXiv:2608.06963}
}
read the original abstract

Biologically plausible learning models aim to explain how neural circuits can implement effective learning under the constraints of real neurons. Although significant progress has been made, a major remaining challenge is that existing models often allow neurons or synapses to represent mixed-sign values, both positive and negative, in violation of a basic aspect of cortical circuitry -- Dale's constraint: biological neurons are either excitatory or inhibitory, but not both, and synapses cannot change sign. In this work, we address this discrepancy by introducing a biologically motivated neural architecture in which both neural activations and learning signals are represented by non-negative activity, and synapses have fixed sign, while still supporting backpropagation-like learning. Our approach uses two complementary interacting non-negative channels to represent positive and negative contributions, inspired by evidence of on-off representations in the brain. These channels are implemented through a simple neural circuit motif, which is repeated throughout the network in both bottom-up and top-down pathways. Combined with a local Hebbian learning rule, the resulting model propagates learning signals and updates weights using only local interactions between neurons. We show theoretically that our learning scheme can exactly recover the backpropagation update despite relying solely on non-negative error signals. Empirically, beyond satisfying stronger biological constraints, the on-off architecture learns efficient representations, yielding substantial gains over comparable vanilla networks on the Tiny ImageNet benchmark. These results demonstrate that effective learning can emerge from biologically plausible mechanisms without requiring mixed-sign signals, providing a step toward more realistic models of neural computation.

Figures

Figures reproduced from arXiv: 2608.06963 by the authors.

Figure 1
Figure 1. On-off neural circuit motif. Left: The basic motif receives two non-negative inputs, x1 and x2, and produces two non-negative output channels. Circles denote excitatory neurons and rectangles denote inhibitory neurons; solid and dashed arrows denote excitatory and inhibitory synapses, respectively. When all synaptic strengths are equal to one, the circuit encodes the signed difference between the inputs: the On chan… view at source ↗
Figure 2
Figure 2. Top-down streams and Hebbian learning. The figure illustrates the update of inter￾layer bottom-up weights connecting consecutive on-off layers. A bottom-up on-off motif receives feedback signals from two corresponding top-down motifs, associated with its On and Off channels. The top-down activities are non-negative, but are separated into channels corresponding to positive and negative synaptic updates (blue and ora… view at source ↗
Figure 3
Figure 3. shows the full organization of two consecutive on-off layers. Each layer contains a single bottom-up (BU) on-off motif, through which feedforward activity propagates upward. For every BU motif, the model includes two corresponding top-down (TD) motifs: one associated with the BU On channel and one associated with the BU Off channel. These TD motifs propagate feedback signals downward. Each TD motif is itself compose… view at source ↗
Figures from the paper (3 more)
Figure 4
Figure 4. Figure 4: Bottom-up connectivity and flow. The BU stream propagates upward through consecutive on-off motifs. Each motif applies ReLU+ and ReLU− nonlinearities, so that at most one output channel is active. The two output neurons of a lower-layer motif are connected to the input…
Figure 5
Figure 5. Figure 5: Top-down connectivity and flow. The TD stream propagates feedback downward through two TD motifs associated with each BU motif. The computation consists of four operations. First, lateral BU–TD connections gate TD activity according to the active BU channel. Second, ea…
Figure 6
Figure 6. Figure 6: Learned complementary on-off activations. Activation maps from two representative on-off units sampled from the second convolutional layer of the Tiny ImageNet model. Activation maps are visualized using the activation-based method of Zeiler and Fergus [2014]. For each…

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

300 extracted references · 19 canonical work pages

  1. [1]

    Decoupling Gating from Linearity

    Decoupling gating from linearity , author=. arXiv preprint arXiv:1906.05032 , year=

    work page Pith review arXiv 1906

  2. [2]

    arXiv preprint arXiv:1711.05101 , year=

    Decoupled weight decay regularization , author=. arXiv preprint arXiv:1711.05101 , year=

    arXiv

  3. [3]

    Nature neuroscience , volume=

    Predictive coding in the visual cortex: a functional interpretation of some extra-classical receptive-field effects , author=. Nature neuroscience , volume=. 1999 , publisher=

    1999

  4. [4]

    Advances in Neural Information Processing Systems , volume=

    In defense of the unitary scalarization for deep multi-task learning , author=. Advances in Neural Information Processing Systems , volume=

  5. [5]

    Net2Brain: A Toolbox to compare artificial vision models with human brain responses

    Net2Brain: A Toolbox to compare artificial vision models with human brain responses , author=. arXiv preprint arXiv:2208.09677 , year=

    work page Pith review arXiv

  6. [6]

    PLoS computational biology , volume=

    Unveiling functions of the visual cortex using task-specific deep neural networks , author=. PLoS computational biology , volume=. 2021 , publisher=

    2021

  7. [7]

    nature , volume=

    Learning representations by back-propagating errors , author=. nature , volume=. 1986 , publisher=

    1986

  8. [8]

    CS 231N , volume=

    Tiny imagenet visual recognition challenge , author=. CS 231N , volume=

Show all 300 references
  1. [9]

    Disordered systems and biological organization , pages=

    Learning process in an asymmetric threshold network , author=. Disordered systems and biological organization , pages=. 1986 , publisher=

    1986

  2. [10]

    Joint european conference on machine learning and knowledge discovery in databases , pages=

    Difference target propagation , author=. Joint european conference on machine learning and knowledge discovery in databases , pages=. 2015 , organization=

    2015

  3. [11]

    arXiv preprint arXiv:2006.06438 , year=

    GAIT-prop: A biologically plausible learning rule derived from backpropagation of error , author=. arXiv preprint arXiv:2006.06438 , year=

    2006 arXiv

  4. [12]

    Proceedings of the national academy of sciences , volume=

    Neurons with graded response have collective computational properties like those of two-state neurons , author=. Proceedings of the national academy of sciences , volume=. 1984 , publisher=

    1984

  5. [13]

    arXiv preprint arXiv:1510.02777 , year=

    Early inference in energy-based models approximates back-propagation , author=. arXiv preprint arXiv:1510.02777 , year=

    arXiv

  6. [14]

    Frontiers in computational neuroscience , volume=

    Equilibrium propagation: Bridging the gap between energy-based models and backpropagation , author=. Frontiers in computational neuroscience , volume=. 2017 , publisher=

    2017

  7. [15]

    Frontiers in neuroscience , volume=

    Scaling equilibrium propagation to deep convnets by drastically reducing its gradient estimator bias , author=. Frontiers in neuroscience , volume=. 2021 , publisher=

    2021

  8. [16]

    Annals of the New York Academy of Sciences , volume=

    Beyond the feedforward sweep: feedback computations in the visual cortex , author=. Annals of the New York Academy of Sciences , volume=. 2020 , publisher=

    2020

  9. [17]

    arXiv preprint arXiv:2206.02629 , year=

    Backpropagation at the infinitesimal inference limit of energy-based models: Unifying predictive coding, equilibrium propagation, and contrastive hebbian learning , author=. arXiv preprint arXiv:2206.02629 , year=

    arXiv

  10. [18]

    Neural computation , volume=

    An approximation of the error backpropagation algorithm in a predictive coding network with local hebbian synaptic plasticity , author=. Neural computation , volume=. 2017 , publisher=

    2017

  11. [19]

    arXiv preprint arXiv:2009.05359 , year=

    Activation Relaxation: A Local Dynamical Approximation to Backpropagation in the Brain , author=. arXiv preprint arXiv:2009.05359 , year=

    2009 arXiv

  12. [20]

    Proceedings of the IEEE , volume=

    Gradient-based learning applied to document recognition , author=. Proceedings of the IEEE , volume=. 1998 , publisher=

    1998

  13. [21]

    arXiv preprint arXiv:1708.07747 , year=

    Fashion-mnist: a novel image dataset for benchmarking machine learning algorithms , author=. arXiv preprint arXiv:1708.07747 , year=

    arXiv

  14. [22]

    The journal of machine learning research , volume=

    Dropout: a simple way to prevent neural networks from overfitting , author=. The journal of machine learning research , volume=. 2014 , publisher=

    2014

  15. [23]

    arXiv preprint arXiv:2401.06209 , year=

    Eyes wide shut? exploring the visual shortcomings of multimodal llms , author=. arXiv preprint arXiv:2401.06209 , year=

    arXiv

  16. [24]

    The Eleventh International Conference on Learning Representations , year=

    When and why vision-language models behave like bags-of-words, and what to do about it? , author=. The Eleventh International Conference on Learning Representations , year=

  17. [25]

    arXiv preprint arXiv:1910.02720 , year=

    Meta-learning deep energy-based memory models , author=. arXiv preprint arXiv:1910.02720 , year=

    1910 arXiv

  18. [26]

    2009 , publisher=

    Learning multiple layers of features from tiny images , author=. 2009 , publisher=

    2009

  19. [27]

    PLoS computational biology , volume=

    Deep supervised, but not unsupervised, models may explain IT cortical representation , author=. PLoS computational biology , volume=. 2014 , publisher=

    2014

  20. [28]

    Journal of Neuroscience , volume=

    Deep neural networks reveal a gradient in the complexity of neural representations across the ventral stream , author=. Journal of Neuroscience , volume=. 2015 , publisher=

    2015

  21. [29]

    Annual review of vision science , volume=

    Deep neural networks: a new framework for modeling biological vision and brain information processing , author=. Annual review of vision science , volume=. 2015 , publisher=

    2015

  22. [30]

    2005 , publisher=

    The organization of behavior: A neuropsychological theory , author=. 2005 , publisher=

    2005

  23. [31]

    Science , volume=

    Regulation of synaptic efficacy by coincidence of postsynaptic APs and EPSPs , author=. Science , volume=. 1997 , publisher=

    1997

  24. [32]

    Nature , volume=

    A neuronal learning rule for sub-millisecond temporal coding , author=. Nature , volume=. 1996 , publisher=

    1996

  25. [33]

    arXiv preprint arXiv:2105.05592 , year=

    Image interpretation by iterative bottom-up top-down processing , author=. arXiv preprint arXiv:2105.05592 , year=

    arXiv

  26. [34]

    Proceedings of the National Academy of Sciences , volume=

    Human-like scene interpretation by a guided counterstream processing , author=. Proceedings of the National Academy of Sciences , volume=. 2023 , publisher=

    2023

  27. [35]

    Proceedings of the National Academy of Sciences , volume=

    Interactions between feedback and lateral connections in the primary visual cortex , author=. Proceedings of the National Academy of Sciences , volume=. 2017 , publisher=

    2017

  28. [36]

    Journal of Comparative Neurology , volume=

    Anatomy of hierarchy: feedforward and feedback pathways in macaque visual cortex , author=. Journal of Comparative Neurology , volume=. 2014 , publisher=

    2014

  29. [37]

    nature , volume=

    Deep learning , author=. nature , volume=. 2015 , publisher=

    2015

  30. [38]

    Neural networks , volume=

    Deep learning in neural networks: An overview , author=. Neural networks , volume=. 2015 , publisher=

    2015

  31. [39]

    , author=

    The recent excitement about neural networks. , author=. Nature , volume=

  32. [40]

    Cognitive science , volume=

    Competitive learning: From interactive activation to adaptive resonance , author=. Cognitive science , volume=. 1987 , publisher=

    1987

  33. [41]

    Proceedings of the IEEE conference on computer vision and pattern recognition , pages=

    Deep residual learning for image recognition , author=. Proceedings of the IEEE conference on computer vision and pattern recognition , pages=

  34. [42]

    arXiv preprint arXiv:2010.11929 , year=

    An image is worth 16x16 words: Transformers for image recognition at scale , author=. arXiv preprint arXiv:2010.11929 , year=

    2010 arXiv

  35. [43]

    International conference on machine learning , pages=

    Learning transferable visual models from natural language supervision , author=. International conference on machine learning , pages=. 2021 , organization=

    2021

  36. [44]

    arXiv preprint arXiv:2304.08485 , year=

    Visual instruction tuning , author=. arXiv preprint arXiv:2304.08485 , year=

    arXiv

  37. [45]

    arXiv preprint arXiv:2301.12597 , year=

    Blip-2: Bootstrapping language-image pre-training with frozen image encoders and large language models , author=. arXiv preprint arXiv:2301.12597 , year=

    arXiv

  38. [46]

    arXiv preprint arXiv:2308.12966 , year=

    Qwen-vl: A frontier large vision-language model with versatile abilities , author=. arXiv preprint arXiv:2308.12966 , year=

    arXiv

  39. [47]

    Proceedings of the thirteenth international conference on artificial intelligence and statistics , pages=

    Understanding the difficulty of training deep feedforward neural networks , author=. Proceedings of the thirteenth international conference on artificial intelligence and statistics , pages=. 2010 , organization=

    2010

  40. [48]

    International conference on machine learning , pages=

    Batch normalization: Accelerating deep network training by reducing internal covariate shift , author=. International conference on machine learning , pages=. 2015 , organization=

    2015

  41. [49]

    IEEE transactions on Information Theory , volume=

    The capacity of the Hopfield associative memory , author=. IEEE transactions on Information Theory , volume=. 1987 , publisher=

    1987

  42. [50]

    Neural computation , volume=

    A fast learning algorithm for deep belief nets , author=. Neural computation , volume=. 2006 , publisher=

    2006

  43. [51]

    Proceedings of the National Academy of Sciences , volume=

    Overparameterized neural networks implement associative memory , author=. Proceedings of the National Academy of Sciences , volume=. 2020 , publisher=

    2020

  44. [52]

    Proceedings of the national academy of sciences , volume=

    Performance-optimized hierarchical models predict neural responses in higher visual cortex , author=. Proceedings of the national academy of sciences , volume=. 2014 , publisher=

    2014

  45. [53]

    Science , volume=

    A synaptically controlled, associative signal for Hebbian plasticity in hippocampal neurons , author=. Science , volume=. 1997 , publisher=

    1997

  46. [54]

    Nature communications , volume=

    Random synaptic feedback weights support error backpropagation for deep learning , author=. Nature communications , volume=. 2016 , publisher=

    2016

  47. [55]

    arXiv preprint arXiv:1904.05391 , year=

    Deep learning without weight transport , author=. arXiv preprint arXiv:1904.05391 , year=

    1904 arXiv

  48. [56]

    Journal of mathematical biology , volume=

    Simplified neuron model as a principal component analyzer , author=. Journal of mathematical biology , volume=. 1982 , publisher=

    1982

  49. [57]

    arXiv preprint arXiv:1812.06488 , year=

    Feedback alignment in deep convolutional networks , author=. arXiv preprint arXiv:1812.06488 , year=

    arXiv

  50. [58]

    arXiv preprint arXiv:1811.03567 , year=

    Biologically-plausible learning algorithms can scale to large datasets , author=. arXiv preprint arXiv:1811.03567 , year=

    arXiv

  51. [59]

    arXiv preprint arXiv:2412.20018 , year=

    Self-assembly of a biologically plausible learning circuit , author=. arXiv preprint arXiv:2412.20018 , year=

    arXiv

  52. [60]

    Proceedings of 1994 IEEE International Conference on Neural Networks (ICNN'94) , volume=

    Backpropagation without weight transport , author=. Proceedings of 1994 IEEE International Conference on Neural Networks (ICNN'94) , volume=. 1994 , organization=

    1994

  53. [61]

    arXiv preprint arXiv:1706.03762 , year=

    Attention is all you need , author=. arXiv preprint arXiv:1706.03762 , year=

    arXiv

  54. [62]

    arXiv preprint arXiv:1312.6114 , year=

    Auto-encoding variational bayes , author=. arXiv preprint arXiv:1312.6114 , year=

    arXiv

  55. [63]

    Processes of change in brain and cognitive development: Attention and performance xxi , volume=

    How far can you go with Hebbian learning, and when does it lead you astray , author=. Processes of change in brain and cognitive development: Attention and performance xxi , volume=. 2006 , publisher=

    2006

  56. [64]

    Advances in neural information processing systems , volume=

    Neural discrete representation learning , author=. Advances in neural information processing systems , volume=

  57. [65]

    Advances in neural information processing systems , volume=

    Can the brain do backpropagation?—exact implementation of backpropagation in predictive coding networks , author=. Advances in neural information processing systems , volume=. 2020 , publisher=

    2020

  58. [66]

    International Conference on Medical image computing and computer-assisted intervention , pages=

    U-net: Convolutional networks for biomedical image segmentation , author=. International Conference on Medical image computing and computer-assisted intervention , pages=. 2015 , organization=

    2015

  59. [67]

    Deep learning in medical image analysis and multimodal learning for clinical decision support , pages=

    Unet++: A nested u-net architecture for medical image segmentation , author=. Deep learning in medical image analysis and multimodal learning for clinical decision support , pages=. 2018 , publisher=

    2018

  60. [68]

    ICASSP 2020-2020 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP) , pages=

    Unet 3+: A full-scale connected unet for medical image segmentation , author=. ICASSP 2020-2020 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP) , pages=. 2020 , organization=

    2020

  61. [69]

    arXiv preprint arXiv:2105.05537 , year=

    Swin-unet: Unet-like pure transformer for medical image segmentation , author=. arXiv preprint arXiv:2105.05537 , year=

    arXiv

  62. [70]

    Neural computation , volume=

    Contrastive similarity matching for supervised learning , author=. Neural computation , volume=. 2021 , publisher=

    2021

  63. [71]

    IEEE transactions on medical imaging , volume=

    H-DenseUNet: hybrid densely connected UNet for liver and tumor segmentation from CT volumes , author=. IEEE transactions on medical imaging , volume=. 2018 , publisher=

    2018

  64. [72]

    arXiv preprint arXiv:2007.03898 , year=

    Nvae: A deep hierarchical variational autoencoder , author=. arXiv preprint arXiv:2007.03898 , year=

    2007 arXiv

  65. [73]

    arXiv preprint arXiv:1905.10985 , year=

    AI-GAs: AI-generating algorithms, an alternate paradigm for producing general artificial intelligence , author=. arXiv preprint arXiv:1905.10985 , year=

    1905 arXiv

  66. [74]

    Humanities and Social Sciences Communications , volume=

    Why general artificial intelligence will not be realized , author=. Humanities and Social Sciences Communications , volume=. 2020 , publisher=

    2020

  67. [75]

    Weak AI

    " Weak AI" is Likely to Never Become" Strong AI", So What is its Greatest Value for us? , author=. arXiv preprint arXiv:2103.15294 , year=

    arXiv

  68. [76]

    Neuron , volume=

    Preference distributions of primary motor cortex neurons reflect control solutions optimized for limb biomechanics , author=. Neuron , volume=. 2013 , publisher=

    2013

  69. [77]

    arXiv preprint arXiv:1407.7906 , year=

    How auto-encoders could provide credit assignment in deep networks via target propagation , author=. arXiv preprint arXiv:1407.7906 , year=

    arXiv

  70. [78]

    Advances in Neural Information Processing Systems , volume=

    A theoretical framework for target propagation , author=. Advances in Neural Information Processing Systems , volume=

  71. [79]

    Nature Reviews Neuroscience , volume=

    Backpropagation and the brain , author=. Nature Reviews Neuroscience , volume=. 2020 , publisher=

    2020

  72. [80]

    International Conference on Machine Learning , pages=

    Towards scaling difference target propagation by learning backprop targets , author=. International Conference on Machine Learning , pages=. 2022 , organization=

    2022

  73. [81]

    , author=

    Target Propagation in Recurrent Neural Networks. , author=. J. Mach. Learn. Res. , volume=

  74. [82]

    arXiv preprint arXiv:2112.01453 , year=

    Target Propagation via Regularized Inversion , author=. arXiv preprint arXiv:2112.01453 , year=

    arXiv

  75. [83]

    The Neuroscientist , volume=

    Bottom-up and top-down attention: different processes and overlapping neural systems , author=. The Neuroscientist , volume=. 2014 , publisher=

    2014

  76. [84]

    Advances in Neural Information Processing Systems , volume=

    Gradient surgery for multi-task learning , author=. Advances in Neural Information Processing Systems , volume=

  77. [85]

    2021 IEEE International Conference on Image Processing (ICIP) , pages=

    Multi-task learning by a top-down control network , author=. 2021 IEEE International Conference on Image Processing (ICIP) , pages=. 2021 , organization=

    2021

  78. [86]

    arXiv preprint arXiv:2009.09796 , year=

    Multi-task learning with deep neural networks: A survey , author=. arXiv preprint arXiv:2009.09796 , year=

    2009 arXiv

  79. [87]

    Advances in Neural Information Processing Systems , volume=

    Multi-task reinforcement learning with soft modularization , author=. Advances in Neural Information Processing Systems , volume=

  80. [88]

    arXiv preprint arXiv:2201.11665 , year=

    Error-driven Input Modulation: Solving the Credit Assignment Problem without a Backward Pass , author=. arXiv preprint arXiv:2201.11665 , year=

    arXiv

  81. [89]

    International Conference on Machine Learning , pages=

    Burst-dependent plasticity and dendritic amplification support target-based learning and hierarchical imitation learning , author=. International Conference on Machine Learning , pages=. 2022 , organization=

    2022

  82. [90]

    Trends in cognitive sciences , volume=

    Theories of error back-propagation in the brain , author=. Trends in cognitive sciences , volume=. 2019 , publisher=

    2019

  83. [91]

    International conference on machine learning , pages=

    Decoupled neural interfaces using synthetic gradients , author=. International conference on machine learning , pages=. 2017 , organization=

    2017

  84. [92]

    arXiv preprint arXiv:2012.13255 , year=

    Intrinsic dimensionality explains the effectiveness of language model fine-tuning , author=. arXiv preprint arXiv:2012.13255 , year=

    2012 arXiv

  85. [93]

    Neuron , volume=

    A top-down cortical circuit for accurate sensory perception , author=. Neuron , volume=. 2015 , publisher=

    2015

  86. [94]

    arXiv preprint arXiv:2110.08744 , year=

    A model for full local image interpretation , author=. arXiv preprint arXiv:2110.08744 , year=

    arXiv

  87. [95]

    Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages=

    An iterative and cooperative top-down and bottom-up inference network for salient object detection , author=. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages=

  88. [96]

    Proceedings of the IEEE/CVF Winter Conference on Applications of Computer Vision , pages=

    Unsupervised Sounding Object Localization with Bottom-Up and Top-Down Attention , author=. Proceedings of the IEEE/CVF Winter Conference on Applications of Computer Vision , pages=

  89. [97]

    Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages=

    Monocular 3D multi-person pose estimation by integrating top-down and bottom-up networks , author=. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages=

  90. [98]

    Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages=

    Sketch, ground, and refine: Top-down dense video captioning , author=. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages=

  91. [99]

    Proceedings of the IEEE/CVF conference on computer vision and pattern recognition , pages=

    Blendmask: Top-down meets bottom-up for instance segmentation , author=. Proceedings of the IEEE/CVF conference on computer vision and pattern recognition , pages=

  92. [100]

    Proceedings of the IEEE/CVF International Conference on Computer Vision , pages=

    Multi-instance pose networks: Rethinking top-down pose estimation , author=. Proceedings of the IEEE/CVF International Conference on Computer Vision , pages=

  93. [101]

    Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages=

    Exploring bottom-up and top-down cues with attentive learning for webly supervised object detection , author=. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages=

  94. [102]

    Nature , volume=

    Vector-based navigation using grid-like representations in artificial agents , author=. Nature , volume=. 2018 , publisher=

    2018

  95. [103]

    Advances in neural information processing systems , volume=

    Generalisation of structural knowledge in the hippocampal-entorhinal system , author=. Advances in neural information processing systems , volume=

  96. [104]

    Nature neuroscience , volume=

    Using goal-driven deep learning models to understand sensory cortex , author=. Nature neuroscience , volume=. 2016 , publisher=

    2016

  97. [105]

    Trends in cognitive sciences , volume=

    Parallel distributed processing theory in the age of deep networks , author=. Trends in cognitive sciences , volume=. 2017 , publisher=

    2017

  98. [106]

    International Conference on Machine Learning , pages=

    Proving the lottery ticket hypothesis: Pruning is all you need , author=. International Conference on Machine Learning , pages=. 2020 , organization=

    2020

  99. [107]

    arXiv preprint arXiv:1803.03635 , year=

    The lottery ticket hypothesis: Finding sparse, trainable neural networks , author=. arXiv preprint arXiv:1803.03635 , year=

    arXiv

  100. [108]

    Advances in Neural Information Processing Systems , volume=

    The elastic lottery ticket hypothesis , author=. Advances in Neural Information Processing Systems , volume=

  101. [109]

    Advances in Neural Information Processing Systems , volume=

    You are caught stealing my winning lottery ticket! Making a lottery ticket claim its ownership , author=. Advances in Neural Information Processing Systems , volume=

  102. [110]

    Neural computation , volume=

    Biologically plausible error-driven learning using local activation differences: The generalized recirculation algorithm , author=. Neural computation , volume=. 1996 , publisher=

    1996

  103. [111]

    arXiv preprint arXiv:2107.12979 , year=

    Predictive coding: a theoretical and experimental review , author=. arXiv preprint arXiv:2107.12979 , year=

    arXiv

  104. [112]

    Neural Computation , volume=

    Predictive coding approximates backprop along arbitrary computation graphs , author=. Neural Computation , volume=. 2022 , publisher=

    2022

  105. [113]

    Neural networks , volume=

    Optimal unsupervised learning in a single-layer linear feedforward neural network , author=. Neural networks , volume=. 1989 , publisher=

    1989

  106. [114]

    Advances in Neural Information Processing Systems , volume=

    Convergence and alignment of gradient descent with random backpropagation weights , author=. Advances in Neural Information Processing Systems , volume=

  107. [115]

    Advances in neural information processing systems , volume=

    Direct feedback alignment provides learning in deep neural networks , author=. Advances in neural information processing systems , volume=

  108. [116]

    The Journal of Physiology , volume=

    Mapping receptive fields in primary visual cortex , author=. The Journal of Physiology , volume=. 2004 , publisher=

    2004

  109. [117]

    Nature reviews neuroscience , volume=

    Interneurons of the neocortical inhibitory system , author=. Nature reviews neuroscience , volume=. 2004 , publisher=

    2004

  110. [118]

    Neuroimage , volume=

    Cortical hierarchy, dual counterstream architecture and the importance of top-down generative networks , author=. Neuroimage , volume=. 2021 , publisher=

    2021

  111. [119]

    Advances in neural information processing systems , volume=

    Assessing the scalability of biologically-motivated deep learning algorithms and architectures , author=. Advances in neural information processing systems , volume=

  112. [120]

    Frontiers in neuroscience , volume=

    Direct feedback alignment with sparse connections for local learning , author=. Frontiers in neuroscience , volume=. 2019 , publisher=

    2019

  113. [121]

    arXiv preprint arXiv:1906.04554 , year=

    Principled training of neural networks with direct feedback alignment , author=. arXiv preprint arXiv:1906.04554 , year=

    1906 arXiv

  114. [122]

    Advances in neural information processing systems , volume=

    Direct feedback alignment scales to modern deep learning tasks and architectures , author=. Advances in neural information processing systems , volume=

  115. [123]

    Frontiers in neuroscience , volume=

    Learning without feedback: Fixed random learning signals allow for feedforward training of deep neural networks , author=. Frontiers in neuroscience , volume=. 2021 , publisher=

    2021

  116. [124]

    arXiv preprint arXiv:2108.13446 , year=

    Benchmarking the Accuracy and Robustness of Feedback Alignment Algorithms , author=. arXiv preprint arXiv:2108.13446 , year=

    arXiv

  117. [125]

    1996 , publisher=

    Rethinking innateness: A connectionist perspective on development , author=. 1996 , publisher=

    1996

  118. [126]

    Neural computing surveys , volume=

    A brief history of connectionism , author=. Neural computing surveys , volume=. 1998 , publisher=

    1998

  119. [127]

    1987 , publisher=

    Parallel Distributed Processing, Volume 2: Explorations in the Microstructure of Cognition: Psychological and Biological Models , author=. 1987 , publisher=

    1987

  120. [128]

    Current opinion in neurobiology , volume=

    An integrative computational architecture for object-driven cortex , author=. Current opinion in neurobiology , volume=. 2019 , publisher=

    2019

  121. [129]

    Trends in cognitive sciences , volume=

    Catastrophic forgetting in connectionist networks , author=. Trends in cognitive sciences , volume=. 1999 , publisher=

    1999

  122. [130]

    Psychology of learning and motivation , volume=

    Catastrophic interference in connectionist networks: The sequential learning problem , author=. Psychology of learning and motivation , volume=. 1989 , publisher=

    1989

  123. [131]

    European conference on computer vision , pages=

    Gdumb: A simple approach that questions our progress in continual learning , author=. European conference on computer vision , pages=. 2020 , organization=

    2020

  124. [132]

    Advances in neural information processing systems , volume=

    Gradient episodic memory for continual learning , author=. Advances in neural information processing systems , volume=

  125. [133]

    Proceedings of the AAAI Conference on Artificial Intelligence , volume=

    Measuring catastrophic forgetting in neural networks , author=. Proceedings of the AAAI Conference on Artificial Intelligence , volume=

  126. [134]

    arXiv preprint arXiv:1606.04671 , year=

    Progressive neural networks , author=. arXiv preprint arXiv:1606.04671 , year=

    arXiv

  127. [135]

    Proceedings of the national academy of sciences , volume=

    Overcoming catastrophic forgetting in neural networks , author=. Proceedings of the national academy of sciences , volume=. 2017 , publisher=

    2017

  128. [136]

    arXiv preprint arXiv:1607.00122 , year=

    Less-forgetting learning in deep neural networks , author=. arXiv preprint arXiv:1607.00122 , year=

    arXiv

  129. [137]

    IEEE transactions on pattern analysis and machine intelligence , volume=

    Learning without forgetting , author=. IEEE transactions on pattern analysis and machine intelligence , volume=. 2017 , publisher=

    2017

  130. [138]

    Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages=

    Representation Compensation Networks for Continual Semantic Segmentation , author=. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages=

  131. [139]

    arXiv preprint arXiv:2106.09685 , year=

    Lora: Low-rank adaptation of large language models , author=. arXiv preprint arXiv:2106.09685 , year=

    arXiv

  132. [140]

    China national conference on Chinese computational linguistics , pages=

    How to fine-tune bert for text classification? , author=. China national conference on Chinese computational linguistics , pages=. 2019 , organization=

    2019

  133. [141]

    arXiv preprint arXiv:2202.09061 , year=

    Vlp: A survey on vision-language pre-training , author=. arXiv preprint arXiv:2202.09061 , year=

    arXiv

  134. [142]

    Expert Systems with Applications , volume=

    A new image classification method using CNN transfer learning and web data augmentation , author=. Expert Systems with Applications , volume=. 2018 , publisher=

    2018

  135. [143]

    International Conference on Machine Learning , pages=

    Parameter-efficient transfer learning for NLP , author=. International Conference on Machine Learning , pages=. 2019 , organization=

    2019

  136. [144]

    European conference on computer vision , pages=

    Facial landmark detection by deep multi-task learning , author=. European conference on computer vision , pages=. 2014 , organization=

    2014

  137. [145]

    Proceedings of the IEEE conference on computer vision and pattern recognition , pages=

    Instance-aware semantic segmentation via multi-task network cascades , author=. Proceedings of the IEEE conference on computer vision and pattern recognition , pages=

  138. [146]

    Proceedings of the IEEE/CVF conference on computer vision and pattern recognition , pages=

    End-to-end multi-task learning with attention , author=. Proceedings of the IEEE/CVF conference on computer vision and pattern recognition , pages=

  139. [147]

    Proceedings of the 24th ACM SIGKDD international conference on knowledge discovery & data mining , pages=

    Modeling task relationships in multi-task learning with multi-gate mixture-of-experts , author=. Proceedings of the 24th ACM SIGKDD international conference on knowledge discovery & data mining , pages=

  140. [148]

    Proceedings of the IEEE conference on computer vision and pattern recognition , pages=

    Cross-stitch networks for multi-task learning , author=. Proceedings of the IEEE conference on computer vision and pattern recognition , pages=

  141. [149]

    Proceedings of the AAAI Conference on Artificial Intelligence , volume=

    Latent multi-task architecture learning , author=. Proceedings of the AAAI Conference on Artificial Intelligence , volume=

  142. [150]

    Proceedings of the IEEE/CVF conference on computer vision and pattern recognition , pages=

    Nddr-cnn: Layerwise feature fusing in multi-task cnns by neural discriminative dimensionality reduction , author=. Proceedings of the IEEE/CVF conference on computer vision and pattern recognition , pages=

  143. [151]

    European Conference on Computer Vision , pages=

    Mti-net: Multi-scale task interaction networks for multi-task learning , author=. European Conference on Computer Vision , pages=. 2020 , organization=

    2020

  144. [152]

    Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition , pages=

    Pad-net: Multi-tasks guided prediction-and-distillation network for simultaneous depth estimation and scene parsing , author=. Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition , pages=

  145. [153]

    Proceedings of the IEEE/CVF conference on computer vision and pattern recognition , pages=

    Pattern-affinitive propagation across depth, surface normal and semantic segmentation , author=. Proceedings of the IEEE/CVF conference on computer vision and pattern recognition , pages=

  146. [154]

    Proceedings of the IEEE/CVF International Conference on Computer Vision , pages=

    Many task learning with task routing , author=. Proceedings of the IEEE/CVF International Conference on Computer Vision , pages=

  147. [155]

    Representation learning using multi-task deep neural networks for semantic classification and information retrieval , author=

  148. [156]

    arXiv preprint arXiv:1511.06114 , year=

    Multi-task sequence to sequence learning , author=. arXiv preprint arXiv:1511.06114 , year=

    arXiv

  149. [157]

    Proceedings of the 54th Annual Meeting of the Association for Computational Linguistics (Volume 2: Short Papers) , pages=

    Deep multi-task learning with low level tasks supervised at lower layers , author=. Proceedings of the 54th Annual Meeting of the Association for Computational Linguistics (Volume 2: Short Papers) , pages=

  150. [158]

    arXiv preprint arXiv:1611.01587 , year=

    A joint many-task model: Growing a neural network for multiple nlp tasks , author=. arXiv preprint arXiv:1611.01587 , year=

    arXiv

  151. [159]

    Proceedings of the AAAI Conference on Artificial Intelligence , volume=

    A hierarchical multi-task approach for learning embeddings from semantic tasks , author=. Proceedings of the AAAI Conference on Artificial Intelligence , volume=

  152. [160]

    Proceedings of the IEEE conference on computer vision and pattern recognition , pages=

    Fully-adaptive feature sharing in multi-task networks with applications in person attribute classification , author=. Proceedings of the IEEE conference on computer vision and pattern recognition , pages=

  153. [161]

    arXiv preprint arXiv:1904.02920 , year=

    Branched multi-task networks: deciding what layers to share , author=. arXiv preprint arXiv:1904.02920 , year=

    1904 arXiv

  154. [162]

    arXiv preprint arXiv:1701.08734 , year=

    Pathnet: Evolution channels gradient descent in super neural networks , author=. arXiv preprint arXiv:1701.08734 , year=

    arXiv

  155. [163]

    arXiv preprint arXiv:1711.00108 , year=

    Beyond shared hierarchies: Deep multitask learning through soft layer ordering , author=. arXiv preprint arXiv:1711.00108 , year=

    arXiv

  156. [164]

    arXiv preprint arXiv:1808.07658 , year=

    Exploring shared structures and hierarchies for multiple nlp tasks , author=. arXiv preprint arXiv:1808.07658 , year=

    arXiv

  157. [165]

    Advances in Neural Information Processing Systems , volume=

    Adashare: Learning what to share for efficient deep multi-task learning , author=. Advances in Neural Information Processing Systems , volume=

  158. [166]

    Proceedings of the European Conference on Computer Vision (ECCV) , pages=

    Piggyback: Adapting a single network to multiple tasks by learning to mask weights , author=. Proceedings of the European Conference on Computer Vision (ECCV) , pages=

  159. [167]

    arXiv preprint arXiv:1908.04339 , year=

    Feature partitioning for efficient multi-task architectures , author=. arXiv preprint arXiv:1908.04339 , year=

    1908 arXiv

  160. [168]

    Proceedings of the IEEE/CVF International Conference on Computer Vision , pages=

    Stochastic filter groups for multi-task cnns: Learning specialist and generalist convolution kernels , author=. Proceedings of the IEEE/CVF International Conference on Computer Vision , pages=

  161. [169]

    Proceedings of the AAAI Conference on Artificial Intelligence , volume=

    Learning sparse sharing architectures for multiple tasks , author=. Proceedings of the AAAI Conference on Artificial Intelligence , volume=

  162. [170]

    arXiv preprint arXiv:1711.01239 , year=

    Routing networks: Adaptive selection of non-linear functions for multi-task learning , author=. arXiv preprint arXiv:1711.01239 , year=

    arXiv

  163. [171]

    arXiv preprint arXiv:1807.04640 , year=

    Automatically composing representation transformations as a means for generalization , author=. arXiv preprint arXiv:1807.04640 , year=

    arXiv

  164. [172]

    Advances in neural information processing systems , volume=

    Modular networks: Learning to decompose neural computation , author=. Advances in neural information processing systems , volume=

  165. [173]

    Proceedings of the IEEE/CVF international conference on computer vision , pages=

    Deep elastic networks with model selection for multi-task learning , author=. Proceedings of the IEEE/CVF international conference on computer vision , pages=

  166. [174]

    Proceedings of the IEEE conference on computer vision and pattern recognition , pages=

    Multi-task learning using uncertainty to weigh losses for scene geometry and semantics , author=. Proceedings of the IEEE conference on computer vision and pattern recognition , pages=

  167. [175]

    International conference on machine learning , pages=

    Gradnorm: Gradient normalization for adaptive loss balancing in deep multitask networks , author=. International conference on machine learning , pages=. 2018 , organization=

    2018

  168. [176]

    Proceedings of the IEEE/CVF conference on computer vision and pattern recognition , pages=

    Pyramidal person re-identification via multi-loss dynamic training , author=. Proceedings of the IEEE/CVF conference on computer vision and pattern recognition , pages=

  169. [177]

    Proceedings of the European conference on computer vision (ECCV) , pages=

    Dynamic task prioritization for multitask learning , author=. Proceedings of the European conference on computer vision (ECCV) , pages=

  170. [178]

    Automatic differentiation in PyTorch , author=

  171. [179]

    arXiv preprint arXiv:1909.06434 , year=

    Adaptive scheduling for multi-task learning , author=. arXiv preprint arXiv:1909.06434 , year=

    1909 arXiv

  172. [180]

    Low resource dependency parsing: Cross-lingual parameter sharing in a neural network parser , author=. Proceedings of the 53rd annual meeting of the Association for Computational Linguistics and the 7th international joint conference on natural language processing (volume 2: s...

  173. [181]

    arXiv preprint arXiv:1606.04038 , year=

    Trace norm regularised deep multi-task learning , author=. arXiv preprint arXiv:1606.04038 , year=

    arXiv

  174. [182]

    International Conference on Machine Learning , pages=

    Deep asymmetric multi-task feature learning , author=. International Conference on Machine Learning , pages=. 2018 , organization=

    2018

  175. [183]

    Recursive routing networks: Learning to compose modules for language understanding , author=. Proceedings of the 2019 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, Volume 1 (Long and Short Papers) , pages=

    2019

  176. [184]

    Multi-task learning for multiple language translation , author=. Proceedings of the 53rd Annual Meeting of the Association for Computational Linguistics and the 7th International Joint Conference on Natural Language Processing (Volume 1: Long Papers) , pages=

  177. [185]

    arXiv preprint arXiv:1702.06053 , year=

    Learning to multi-task by active sampling , author=. arXiv preprint arXiv:1702.06053 , year=

    arXiv

  178. [186]

    Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages=

    Attentive single-tasking of multiple tasks , author=. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages=

  179. [187]

    arXiv preprint arXiv:1806.08028 , year=

    Gradient adversarial training of neural networks , author=. arXiv preprint arXiv:1806.08028 , year=

    arXiv

  180. [188]

    arXiv preprint arXiv:1511.06295 , year=

    Policy distillation , author=. arXiv preprint arXiv:1511.06295 , year=

    arXiv

  181. [189]

    arXiv preprint arXiv:1511.06342 , year=

    Actor-mimic: Deep multitask and transfer reinforcement learning , author=. arXiv preprint arXiv:1511.06342 , year=

    arXiv

  182. [190]

    Advances in neural information processing systems , volume=

    Distral: Robust multitask reinforcement learning , author=. Advances in neural information processing systems , volume=

  183. [191]

    arXiv preprint arXiv:1904.09482 , year=

    Improving multi-task deep neural networks via knowledge distillation for natural language understanding , author=. arXiv preprint arXiv:1904.09482 , year=

    1904 arXiv

  184. [192]

    arXiv preprint arXiv:1907.04829 , year=

    Bam! born-again multi-task networks for natural language understanding , author=. arXiv preprint arXiv:1907.04829 , year=

    1907 arXiv

  185. [193]

    Advances in neural information processing systems , volume=

    Multi-task learning as multi-objective optimization , author=. Advances in neural information processing systems , volume=

  186. [194]

    Advances in Neural Information Processing Systems , volume=

    Just pick a sign: Optimizing deep multitask models with gradient sign dropout , author=. Advances in Neural Information Processing Systems , volume=

  187. [195]

    Advances in neural information processing systems , volume=

    Pareto multi-task learning , author=. Advances in neural information processing systems , volume=

  188. [196]

    Frontiers in Systems Neuroscience , volume=

    Shaping the cortical landscape: Functions and mechanisms of top-down cortical feedback pathways , author=. Frontiers in Systems Neuroscience , volume=. 2020 , publisher=

    2020

  189. [197]

    arXiv preprint arXiv:2110.04544 , year=

    Clip-adapter: Better vision-language models with feature adapters , author=. arXiv preprint arXiv:2110.04544 , year=

    arXiv

  190. [198]

    Proceedings of the IEEE/CVF conference on computer vision and pattern recognition , pages=

    Improved baselines with visual instruction tuning , author=. Proceedings of the IEEE/CVF conference on computer vision and pattern recognition , pages=

  191. [199]

    International Conference on Machine Learning , pages=

    Zero-shot text-to-image generation , author=. International Conference on Machine Learning , pages=. 2021 , organization=

    2021

  192. [200]

    Science , volume=

    A general reinforcement learning algorithm that masters chess, shogi, and Go through self-play , author=. Science , volume=. 2018 , publisher=

    2018

  193. [201]

    Advances in Neural Information Processing Systems , volume=

    Only train once: A one-shot neural network training and pruning framework , author=. Advances in Neural Information Processing Systems , volume=

  194. [202]

    Advances in Neural Information Processing Systems , volume=

    Pruning randomly initialized neural networks with iterative randomization , author=. Advances in Neural Information Processing Systems , volume=

  195. [203]

    Advances in Neural Information Processing Systems , volume=

    Sparse training via boosting pruning plasticity with neuroregeneration , author=. Advances in Neural Information Processing Systems , volume=

  196. [204]

    Advances in Neural Information Processing Systems , volume=

    Why lottery ticket wins? a theoretical perspective of sample complexity on sparse neural networks , author=. Advances in Neural Information Processing Systems , volume=

  197. [205]

    arXiv preprint arXiv:1608.07639 , year=

    Learning to generalize to new compositions in image understanding , author=. arXiv preprint arXiv:1608.07639 , year=

    arXiv

  198. [206]

    Advances in Neural Information Processing Systems , volume=

    Global convergence of gradient descent for asymmetric low-rank matrix factorization , author=. Advances in Neural Information Processing Systems , volume=

  199. [207]

    arXiv preprint arXiv:1810.02054 , year=

    Gradient descent provably optimizes over-parameterized neural networks , author=. arXiv preprint arXiv:1810.02054 , year=

    arXiv

  200. [208]

    arXiv preprint arXiv:1609.04747 , year=

    An overview of gradient descent optimization algorithms , author=. arXiv preprint arXiv:1609.04747 , year=

    arXiv

  201. [209]

    Machine learning , volume=

    Multitask learning , author=. Machine learning , volume=. 1997 , publisher=

    1997

  202. [210]

    Advances in Neural Information Processing Systems , volume=

    Conflict-averse gradient descent for multi-task learning , author=. Advances in Neural Information Processing Systems , volume=

  203. [211]

    Proceedings of the IEEE/CVF International Conference on Computer Vision , pages=

    Task Switching Network for Multi-task Learning , author=. Proceedings of the IEEE/CVF International Conference on Computer Vision , pages=

  204. [212]

    Proceedings of the IEEE conference on Computer Vision and Pattern Recognition , pages=

    Packnet: Adding multiple tasks to a single network by iterative pruning , author=. Proceedings of the IEEE conference on Computer Vision and Pattern Recognition , pages=

  205. [213]

    International Conference on Machine Learning , pages=

    Forget-free continual learning with winning subnetworks , author=. International Conference on Machine Learning , pages=. 2022 , organization=

    2022

  206. [214]

    Advances in neural information processing systems , volume=

    Deconstructing lottery tickets: Zeros, signs, and the supermask , author=. Advances in neural information processing systems , volume=

  207. [215]

    Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages=

    What's hidden in a randomly weighted neural network? , author=. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages=

  208. [216]

    International Conference on Machine Learning , pages=

    The combinatorial brain surgeon: Pruning weights that cancel one another in neural networks , author=. International Conference on Machine Learning , pages=. 2022 , organization=

    2022

  209. [217]

    Advances in neural information processing systems , volume=

    Second order derivatives for network pruning: Optimal brain surgeon , author=. Advances in neural information processing systems , volume=

  210. [218]

    Advances in Neural Information Processing Systems , volume=

    Drawing robust scratch tickets: Subnetworks with inborn robustness are found within randomly initialized networks , author=. Advances in Neural Information Processing Systems , volume=

  211. [219]

    Advances in neural information processing systems , volume=

    One ticket to win them all: generalizing lottery ticket initializations across datasets and optimizers , author=. Advances in neural information processing systems , volume=

  212. [220]

    arXiv preprint arXiv:1903.04476 , year=

    Continual learning via neural pruning , author=. arXiv preprint arXiv:1903.04476 , year=

    1903 arXiv

  213. [221]

    International Conference on Machine Learning , pages=

    Overcoming catastrophic forgetting with hard attention to the task , author=. International Conference on Machine Learning , pages=. 2018 , organization=

    2018

  214. [222]

    International Conference on Learning Representations , year=

    Long live the lottery: The existence of winning tickets in lifelong learning , author=. International Conference on Learning Representations , year=

  215. [223]

    Advances in Neural Information Processing Systems , volume=

    Pruning neural networks without any data by iteratively conserving synaptic flow , author=. Advances in Neural Information Processing Systems , volume=

  216. [224]

    Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages=

    Continual Object Detection via Prototypical Task Correlation Guided Gating Mechanism , author=. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages=

  217. [225]

    Elife , volume=

    Dynamic representation of partially occluded objects in primate prefrontal and visual cortex , author=. Elife , volume=. 2017 , publisher=

    2017

  218. [226]

    Nature Reviews Neuroscience , volume=

    Top-down influences on visual processing , author=. Nature Reviews Neuroscience , volume=. 2013 , publisher=

    2013

  219. [227]

    Advances in neural information processing systems , volume=

    Visual instruction tuning , author=. Advances in neural information processing systems , volume=

  220. [228]

    arXiv preprint arXiv:2312.16602 , year=

    Visual Instruction Tuning towards General-Purpose Multimodal Model: A Survey , author=. arXiv preprint arXiv:2312.16602 , year=

    arXiv

  221. [229]

    Advances in Neural Information Processing Systems , volume=

    Instructblip: Towards general-purpose vision-language models with instruction tuning , author=. Advances in Neural Information Processing Systems , volume=

  222. [230]

    arXiv preprint arXiv:2310.09478 , year=

    Minigpt-v2: large language model as a unified interface for vision-language multi-task learning , author=. arXiv preprint arXiv:2310.09478 , year=

    arXiv

  223. [231]

    arXiv preprint arXiv:2304.10592 , year=

    Minigpt-4: Enhancing vision-language understanding with advanced large language models , author=. arXiv preprint arXiv:2304.10592 , year=

    arXiv

  224. [232]

    arXiv preprint arXiv:2306.14824 , year=

    Kosmos-2: Grounding multimodal large language models to the world , author=. arXiv preprint arXiv:2306.14824 , year=

    arXiv

  225. [233]

    Proceedings of the National Academy of Sciences , volume=

    Task context impacts visual object processing differentially across the cortex , author=. Proceedings of the National Academy of Sciences , volume=. 2014 , publisher=

    2014

  226. [234]

    Journal of cognitive neuroscience , volume=

    Spatial and feature-selective attention have distinct, interacting effects on population-level tuning , author=. Journal of cognitive neuroscience , volume=. 2022 , publisher=

    2022

  227. [235]

    Frontiers in neuroscience , volume=

    Neural responses to naturalistic clips of behaving animals in two different task contexts , author=. Frontiers in neuroscience , volume=. 2018 , publisher=

    2018

  228. [236]

    Journal of Neuroscience , volume=

    Task-dependent warping of semantic representations during search for visual action categories , author=. Journal of Neuroscience , volume=. 2022 , publisher=

    2022

  229. [237]

    Frontiers in neuroscience , volume=

    Goal-directed processing of naturalistic stimuli modulates large-scale functional connectivity , author=. Frontiers in neuroscience , volume=. 2019 , publisher=

    2019

  230. [238]

    Advances in Neural Information Processing Systems , volume=

    Correlative Information Maximization: A Biologically Plausible Approach to Supervised Deep Neural Networks without Weight Symmetry , author=. Advances in Neural Information Processing Systems , volume=

  231. [239]

    IEEE transactions on pattern analysis and machine intelligence , volume=

    Multi-task learning for dense prediction tasks: A survey , author=. IEEE transactions on pattern analysis and machine intelligence , volume=. 2021 , publisher=

    2021

  232. [240]

    arXiv preprint arXiv:2111.10603 , year=

    A closer look at loss weighting in multi-task learning , author=. arXiv preprint arXiv:2111.10603 , year=

    arXiv

  233. [241]

    Advances in Neural Information Processing Systems , volume=

    Do Current Multi-Task Optimization Methods in Deep Learning Even Help? , author=. Advances in Neural Information Processing Systems , volume=

  234. [242]

    arXiv preprint arXiv:2212.13345 , year=

    The forward-forward algorithm: Some preliminary investigations , author=. arXiv preprint arXiv:2212.13345 , year=

    arXiv

  235. [243]

    2021 , publisher=

    A computational perspective on visual attention , author=. 2021 , publisher=

    2021

  236. [244]

    Advances in neural information processing systems , volume=

    Dynamic routing between capsules , author=. Advances in neural information processing systems , volume=

  237. [245]

    2021 , organization=

    Towards impartial multi-task learning , author=. 2021 , organization=

    2021

  238. [246]

    Elife , volume=

    Dendritic NMDA receptors in parvalbumin neurons enable strong and stable neuronal assemblies , author=. Elife , volume=. 2019 , publisher=

    2019

  239. [247]

    Proceedings of the AAAI Conference on Artificial Intelligence , volume=

    Tdaf: Top-down attention framework for vision tasks , author=. Proceedings of the AAAI Conference on Artificial Intelligence , volume=

  240. [248]

    Proceedings of the AAAI Conference on Artificial Intelligence , volume=

    Close the loop: a unified bottom-up and top-down paradigm for joint image deraining and segmentation , author=. Proceedings of the AAAI Conference on Artificial Intelligence , volume=

  241. [249]

    Advances in Neural Information Processing Systems , volume=

    Local plasticity rules can learn deep representations using self-supervised contrastive predictions , author=. Advances in Neural Information Processing Systems , volume=

  242. [250]

    International Conference on Machine Learning , pages=

    Minimizing control for credit assignment with strong feedback , author=. International Conference on Machine Learning , pages=. 2022 , organization=

    2022

  243. [251]

    Proceedings of the 40th International Conference on Machine Learning , pages =

    Flexible Phase Dynamics for Bio-Plausible Contrastive Learning , author =. Proceedings of the 40th International Conference on Machine Learning , pages =. 2023 , editor =

    2023

  244. [252]

    Proceedings of the 8th International Conference on Sciences of Electronics, Technologies of Information and Telecommunications (SETIT’18), Vol

    Strided convolution instead of max pooling for memory efficiency of convolutional neural networks , author=. Proceedings of the 8th International Conference on Sciences of Electronics, Technologies of Information and Telecommunications (SETIT’18), Vol. 1 , pages=. 2020 , organization=

    2020

  245. [253]

    Proceedings of the 40th International Conference on Machine Learning , pages =

    Dual Propagation: Accelerating Contrastive Hebbian Learning with Dyadic Neurons , author =. Proceedings of the 40th International Conference on Machine Learning , pages =. 2023 , editor =

    2023

  246. [254]

    Proceedings of the AAAI conference on artificial intelligence , volume=

    Rumor detection on social media with bi-directional graph convolutional networks , author=. Proceedings of the AAAI conference on artificial intelligence , volume=

  247. [255]

    Proceedings of the AAAI Conference on Artificial Intelligence , volume=

    SLOTH: Structured Learning and Task-Based Optimization for Time Series Forecasting on Hierarchies , author=. Proceedings of the AAAI Conference on Artificial Intelligence , volume=

  248. [256]

    Proceedings of the AAAI Conference on Artificial Intelligence , author=

    Efficient End-to-End Video Question Answering with Pyramidal Multimodal Transformer , volume=. Proceedings of the AAAI Conference on Artificial Intelligence , author=. 2023 , month=

    2023

  249. [257]

    Proceedings of the AAAI Conference on Artificial Intelligence , volume=

    Classification by attention: Scene graph classification with prior knowledge , author=. Proceedings of the AAAI Conference on Artificial Intelligence , volume=

  250. [258]

    Proceedings of the AAAI Conference on Artificial Intelligence , volume=

    Reverse differentiation via predictive coding , author=. Proceedings of the AAAI Conference on Artificial Intelligence , volume=

  251. [259]

    Proceedings of the AAAI Conference on Artificial Intelligence , volume=

    Learning by competition of self-interested reinforcement learning agents , author=. Proceedings of the AAAI Conference on Artificial Intelligence , volume=

  252. [260]

    Proceedings of the AAAI Conference on Artificial Intelligence , volume=

    Fixed-Weight Difference Target Propagation , author=. Proceedings of the AAAI Conference on Artificial Intelligence , volume=

  253. [261]

    Proceedings of the AAAI Conference on Artificial Intelligence , volume=

    Backpropagation-Free Deep Learning with Recursive Local Representation Alignment , author=. Proceedings of the AAAI Conference on Artificial Intelligence , volume=

  254. [262]

    Proceedings of the IEEE conference on computer vision and pattern recognition , pages=

    Neural module networks , author=. Proceedings of the IEEE conference on computer vision and pattern recognition , pages=

  255. [263]

    arXiv preprint arXiv:1910.04915 , year=

    Flexible multi-task networks by learning parameter allocation , author=. arXiv preprint arXiv:1910.04915 , year=

    1910 arXiv

  256. [264]

    Advances in Neural Information Processing Systems , volume=

    Dynamic inference with neural interpreters , author=. Advances in Neural Information Processing Systems , volume=

  257. [265]

    Advances in Neural Information Processing Systems , volume=

    Is a modular architecture enough? , author=. Advances in Neural Information Processing Systems , volume=

  258. [266]

    The Journal of Machine Learning Research , volume=

    Switch transformers: Scaling to trillion parameter models with simple and efficient sparsity , author=. The Journal of Machine Learning Research , volume=. 2022 , publisher=

    2022

  259. [267]

    International Conference on Machine Learning , pages=

    Glam: Efficient scaling of language models with mixture-of-experts , author=. International Conference on Machine Learning , pages=. 2022 , organization=

    2022

  260. [268]

    Global workspace theory, a rigorous scientific theory of consciousness , author=

    In the theatre of consciousness. Global workspace theory, a rigorous scientific theory of consciousness , author=. Journal of consciousness Studies , volume=. 1997 , publisher=

    1997

  261. [269]

    Robotics, AI, and Humanity: Science, Ethics, and Policy , pages=

    What is consciousness, and could machines have it? , author=. Robotics, AI, and Humanity: Science, Ethics, and Policy , pages=. 2021 , publisher=

    2021

  262. [270]

    Proceedings of the IEEE international conference on computer vision , pages=

    Deep learning face attributes in the wild , author=. Proceedings of the IEEE international conference on computer vision , pages=

  263. [271]

    arXiv preprint arXiv:2308.07870 , year=

    Brain-inspired computational intelligence via predictive coding , author=. arXiv preprint arXiv:2308.07870 , year=

  264. [272]

    Elife , volume=

    Local online learning in recurrent networks with random feedback , author=. Elife , volume=. 2019 , publisher=

    2019

  265. [273]

    Nature communications , volume=

    A solution to the learning dilemma for recurrent networks of spiking neurons , author=. Nature communications , volume=. 2020 , publisher=

    2020

  266. [274]

    Advances in Neural Information Processing Systems , volume=

    Beyond accuracy: generalization properties of bio-plausible temporal credit assignment rules , author=. Advances in Neural Information Processing Systems , volume=

  267. [275]

    Advances in neural information processing systems , volume=

    Updates of equilibrium prop match gradients of backprop through time in an RNN with static input , author=. Advances in neural information processing systems , volume=

  268. [276]

    Plos one , volume=

    On the relationship between predictive coding and backpropagation , author=. Plos one , volume=. 2022 , publisher=

    2022

  269. [277]

    Advances in Neural Information Processing Systems , volume=

    Constrained predictive coding as a biologically plausible model of the cortical hierarchy , author=. Advances in Neural Information Processing Systems , volume=

  270. [278]

    Vision research , volume=

    Visual attention: The past 25 years , author=. Vision research , volume=. 2011 , publisher=

    2011

  271. [279]

    Current biology , volume=

    Visual attention: bottom-up versus top-down , author=. Current biology , volume=. 2004 , publisher=

    2004

  272. [280]

    Nature reviews neuroscience , volume=

    Computational modelling of visual attention , author=. Nature reviews neuroscience , volume=. 2001 , publisher=

    2001

  273. [281]

    arXiv preprint arXiv:2307.16125 , year=

    Seed-bench: Benchmarking multimodal llms with generative comprehension , author=. arXiv preprint arXiv:2307.16125 , year=

    arXiv

  274. [282]

    arXiv preprint arXiv:2305.10355 , year=

    Evaluating object hallucination in large vision-language models , author=. arXiv preprint arXiv:2305.10355 , year=

    arXiv

  275. [283]

    Cerebral cortex , volume=

    Sequence seeking and counter streams: a computational model for bidirectional information flow in the visual cortex , author=. Cerebral cortex , volume=. 1995 , publisher=

    1995

  276. [284]

    Cognitive psychology , volume=

    A feature-integration theory of attention , author=. Cognitive psychology , volume=. 1980 , publisher=

    1980

  277. [285]

    International conference on machine learning , pages=

    Scaling up visual and vision-language representation learning with noisy text supervision , author=. International conference on machine learning , pages=. 2021 , organization=

    2021

  278. [286]

    Proceedings of the IEEE/CVF international conference on computer vision , pages=

    Emerging properties in self-supervised vision transformers , author=. Proceedings of the IEEE/CVF international conference on computer vision , pages=

  279. [287]

    Advances in Neural Information Processing Systems , volume=

    Cambrian-1: A fully open, vision-centric exploration of multimodal llms , author=. Advances in Neural Information Processing Systems , volume=

  280. [288]

    Advances in neural information processing systems , volume=

    Instructblip: Towards general-purpose vision-language models with instruction tuning , author=. Advances in neural information processing systems , volume=

  281. [289]

    Advances in neural information processing systems , volume=

    Flamingo: a visual language model for few-shot learning , author=. Advances in neural information processing systems , volume=

  282. [290]

    Advances in neural information processing systems , volume=

    Instruction-guided visual masking , author=. Advances in neural information processing systems , volume=

  283. [291]

    arXiv preprint arXiv:2303.04671 , year=

    Visual chatgpt: Talking, drawing and editing with visual foundation models , author=. arXiv preprint arXiv:2303.04671 , year=

    arXiv

  284. [292]

    Proceedings of the IEEE/CVF international conference on computer vision , pages=

    Vipergpt: Visual inference via python execution for reasoning , author=. Proceedings of the IEEE/CVF international conference on computer vision , pages=

  285. [293]

    Advances in neural information processing systems , volume=

    Imagenet classification with deep convolutional neural networks , author=. Advances in neural information processing systems , volume=

  286. [294]

    arXiv preprint arXiv:1409.1556 , year=

    Very deep convolutional networks for large-scale image recognition , author=. arXiv preprint arXiv:1409.1556 , year=

    arXiv

  287. [295]

    Advances in Neural Information Processing Systems , volume=

    Pretraining with random noise for fast and robust learning without weight transport , author=. Advances in Neural Information Processing Systems , volume=

  288. [296]

    International conference on machine learning , pages=

    Efficientnet: Rethinking model scaling for convolutional neural networks , author=. International conference on machine learning , pages=. 2019 , organization=

    2019

  289. [297]

    Proceedings of the IEEE conference on computer vision and pattern recognition , pages=

    Rich feature hierarchies for accurate object detection and semantic segmentation , author=. Proceedings of the IEEE conference on computer vision and pattern recognition , pages=

  290. [298]

    Advances in neural information processing systems , volume=

    Faster r-cnn: Towards real-time object detection with region proposal networks , author=. Advances in neural information processing systems , volume=

  291. [299]

    Proceedings of the Computer Vision and Pattern Recognition Conference , pages=

    FRAMES-VQA: Benchmarking Fine-Tuning Robustness across Multi-Modal Shifts in Visual Question Answering , author=. Proceedings of the Computer Vision and Pattern Recognition Conference , pages=

  292. [300]

    arXiv preprint arXiv:2306.14565 , year=

    Mitigating hallucination in large multi-modal models via robust instruction tuning , author=. arXiv preprint arXiv:2306.14565 , year=

    arXiv

Pith tools

Reviewed August 10, 2026 · model on record in the stance chip above.