Pith. sign in

REVIEW 4 major objections 3 minor 46 references

Learning Visual Features Under Motion Invariance

T0 review · 4 major / 3 minor · reviewed 2026-08-14 · deepseek-v4-flash

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

Pith's one-line read This paper claims that enforcing motion invariance alone—via a variational 'cognitive action'—is enough to learn convolutional filters from unlabeled video.

desk verdict A serious variational framework for unsupervised video feature learning with an elegant motion-invariance core, but the central claim that quadratic surrogates preserve the mutual information stationary points is asserted without proof and needs to be fixed before the theory is trusted. read the letter →

arxiv 1909.00350 v3 pith:3IKPSBGD submitted 2019-09-01 cs.CV

Alessandro Betti , Marco Gori , Stefano Melacci This is my paper · ORCID
classification cs.CV MSC 68T45
keywords motioninvariancecognitiveactionunsupervisedlearningconvolutionalfiltersvariationalprinciplesvideoprocessingvisualfeaturesEuler-Lagrangeequations
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 argues that a single principle—visual features should stay constant as the pixels they describe move across the retina—is enough to derive an unsupervised learning rule for convolutional filters from raw video. The paper formulates learning as minimizing a 'cognitive action' functional that combines an information-theoretic term with a motion-invariance penalty and parsimony terms. Minimizing this action yields Euler-Lagrange equations that, on a discrete retina, become a fourth-order differential equation in time for the filter weights. Solving that equation gives filters that develop without labels, and the paper contends that this motion invariance subsumes translation, rotation, and scale invariance. The claim matters because it points to a principled, label-free alternative to supervised deep networks and offers an account of why temporal structure aids visual development.

What carries the argument

The central object is the cognitive action $A(\phi)$, a functional of the convolutional filters $\phi_{ij}(x-y,t)$ built from a maximum-entropy-style information index, a quadratic motion-invariance penalty $(\partial_t \Phi_i + v_j \partial_j \Phi_i)^2$ where $v$ is the optical flow, and spatial and temporal parsimony terms. The argument is carried by variational calculus: stationarity $\delta A(\phi)=0$ produces nonlocal integro-differential Euler-Lagrange equations; a causal retiming of the entropy term makes them time-local; and factoring the filters as a bell-shaped receptive field $G(x)\varphi_{ij}(x,t)$, with $G$ a Green's function of a self-adjoint operator, makes them space-local. On the discrete retina, the whole scheme collapses into a single fourth-order differential equation for the filter vector $q(t)$, and the reset-via-null-signal argument turns the accompanying boundary conditions into a causal learning rule.

What would settle it

A direct test is to compute, on a small video corpus, the true mutual information $I(Y;X,T,F)$ at the stationary filters found by solving Eq. (53) and compare it with the value of the quadratic surrogate at the same filters; if the surrogate's stationary points do not correspond to stationary points of the true mutual information, the theoretical justification collapses. Concretely, one could run two optimizations—one minimizing Eq. (21) and one minimizing the same action with the true information terms—and check whether the two filter trajectories converge to equivalent points.

Watch

Extended reading notes

Core claim

The central claim is that minimizing the cognitive action $A(\phi)$ in Eq. (21)—where the mutual information terms are replaced by the quadratic surrogates $(\int \Phi)^2$ and $\Phi^2$—leads to Euler-Lagrange equations for the filters, and that on a discrete retina these reduce to a local fourth-order time-variant differential equation, Eq. (53), for the vectorized filter weights $q(t)$. The equation is well-posed: under the coercivity conditions (48) the action admits a minimum, and the boundary conditions can be satisfied by injecting brief periods of null video signal, which act as a reset. With this scheme, convolutional filters emerge from natural video without labels, and the resulting motion-invariant features are claimed to provide the only invariance needed, with translation, rotation, and scale invariance following from it. Experiments on driving videos show the learned features, paired with a simple classifier, outperform sparse convolutional autoencoders and an RGB baseline on a five-class semantic labeling task.

Load-bearing premise

The result depends on the claim that replacing the mutual information terms with the quadratic surrogates $(\int \Phi)^2$ and $\Phi^2$ retains all the basic properties on the stationary points of the mutual information; no proof of that equivalence is given, and if it fails, the derived equations optimize a different objective than the information-theoretic one.

Editorial extensions

If this is right

  • Convolutional filters can be learned from raw video without supervision by numerically integrating Eq. (53), so label-hungry training on static image collections is not the only route to useful visual features.
  • Because translation, rotation, and scale invariance are claimed to follow from motion invariance, a single motion-coherence constraint should yield features stable under those transformations, reducing the need for explicit data augmentation.
  • The theory prescribes receptive-field structure and hierarchical layering rather than treating them as design choices: peaked bell-shaped filters are required for spatial locality, and deep stacks emerge naturally.
  • The reset argument implies that brief periods of null visual signal actively help learning by making the boundary conditions satisfiable, so temporal structure in the training stream is part of the learning mechanism.

Reading between the lines

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

  • If the quadratic-surrogate equivalence holds, the same variational derivation could be applied to other sensory modalities—audio or tactile streams—where a flow or motion field is available, yielding modality-specific unsupervised feature laws.
  • A concrete experiment beyond the paper is to train the same architecture on temporally shuffled frames and compare filter quality; the theory predicts a severe degradation, isolating motion coherence as the causal ingredient rather than mere video statistics.
  • One could test the developmental prediction directly by varying the blurring schedule and measuring whether an intermediate schedule, rather than the fastest or slowest, maximizes downstream task accuracy.
Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

4 major / 3 minor

Summary. The manuscript develops a variational theory for unsupervised learning of convolutional filters from video signals. The authors define a 'cognitive action' that combines information-theoretic terms, a motion-invariance penalty, and spatiotemporal parsimony terms, and they claim that its stationary points yield the learned filters via Euler-Lagrange equations. After a sequence of reductions intended to restore temporal and spatial locality, the theory leads to a fourth-order differential equation on the discrete retina. The paper further argues that motion invariance subsumes translation, rotation, and scale invariance, and it presents experiments on video feature extraction and on a BDD100K semantic-labeling transfer task. The central formal claims are the Euler-Lagrange equations (27) and (53), the existence of a minimum (Theorem 5), and the boundary-condition mechanism (Theorems 7 and 8).

Significance. If the theoretical core were fully established, the paper would offer a distinctive variational foundation for unsupervised video feature learning, connecting information-based feature extraction, motion coherence, receptive fields, and causal dynamics. The authors provide extensive symbolic derivations in the appendices, make code and data available, and report reproducible experimental comparisons on standard benchmarks. The conceptual contribution is real: it proposes a least-action principle for filter learning that is an alternative to gradient-based training of convolutional networks. However, the value of the contribution depends crucially on the unproved quadratic surrogate for mutual information and on imported well-posedness results, so the significance is conditional on closing those gaps.

major comments (4)
  1. [Section 5, Eq. (21)] The paper asserts, immediately before Eq. (21), that replacing the mutual-information terms in Eq. (20) by the quadratic surrogates (∫Φ)² and Φ² 'retains all the basic properties on the stationary points of the mutual information,' but no proof or reference is supplied. This equivalence is load-bearing: Theorems 1–3 and the discrete equation (53) are derived from the surrogate action (21), not from the mutual-information action (20). If the stationary points are not preserved, the central claim that the scheme learns features by maximizing mutual information is unsupported. Please either prove the equivalence under explicit conditions on f, Φ, and the admissible filter class, or restate the theory as being about the quadratic surrogate and treat the mutual-information connection only as motivation.
  2. [Section 5, Theorem 3 and Eq. (40)] The spatial-localization step is presented as an equivalence, but it relies on approximations that are not controlled. Theorem 4 only proves that L_σ^m G_σ converges to the delta distribution as σ→0; for finite σ, which is the operating regime of the experiments with finite-width Gaussian receptive fields, Eq. (42) holds only approximately. Moreover, the proof requires L*G = δ on a bounded retina X with G(∂X)=0, whereas a Gaussian does not vanish exactly on a finite boundary, and the existence and boundary behavior of the adjoint field Λ solving LΛ = Δ... are assumed rather than established. Please state the precise functional setting, provide error bounds as a function of σ, or explicitly label Eq. (40) as an approximate localization.
  3. [Section 6, Eq. (52) and Theorems 7–8] The boundary-condition mechanism is not self-contained and rests on an unjustified manipulation of the video signal. The proofs of Theorems 7 and 8 are deferred to reference [32], and the assertion that inserting null-signal intervals 'does not change the information structure' of the video is stated without proof. This is not a minor point: the reset mechanism is used both to satisfy the boundary conditions (52) and in the experiments (Section 8.1, reset thresholds ε_j). Without an argument that the stationary points of the learning objective are preserved under such resets, the well-posedness and causal interpretation of the learning dynamics remain conditional. Please either prove invariance of the relevant stationary points under resets, or state the reset operation as an additional modeling assumption and analyze its effect on the objective.
  4. [Appendix C, definition of M_αβ] In Appendix C, the matrix M_αβ is defined as ˙χ^i_α (g_x γ^x_α γ^x_β) δ_ij ˙χ^j_β, which includes the derivative ˙χ and the index contraction. Under this definition, the third term in the expansion is not a quadratic form in ˙χ, and Proposition 3's expression M(q) = (1/2)∫ ˙q M♮ ˙q is inconsistent. Since M♮ appears in the Euler-Lagrange equation (53) through Z2 and λM, this error affects the central discrete derivation. Please correct the definition to M_αβ = g_x γ^x_α γ^x_β and verify the subsequent vectorization identities.
minor comments (3)
  1. [Throughout] The manuscript contains numerous typos and infelicities ('Mathermatics' in the affiliation, 'porpuse', 'ill-position', 'assolve'); a careful copyedit is needed.
  2. [Section 4, Eq. (34)] In the temporal-locality approximation (34), the integrand appears to contain both h(t) outside and f(x,t) = h(t)g(x−a(t)) inside the frame integral, which would give an extra power of h(t). Please check whether the factor should be h(t)(∫ g Φ)² rather than h(t)(∫ g Φ f)².
  3. [Section 1 and Section 8.2, Table 4] The claim that motion invariance is 'the only invariance that we need' is stronger than what is demonstrated; translation, rotation, and scale invariance are discussed informally only. The experimental conclusion that cognitive-action models outperform autoencoders is also only true when RGB information is appended; without RGB, the comparison is mixed, e.g., cal-7L has mean IoU 26.22 versus autoenc-7L 27.25.

Circularity Check

1 steps flagged · score 4.0 of 10

Well-posedness and reset guarantees are imported from the authors' own [32]; the core variational derivation is not itself circular.

  1. self citation load bearing [Section 6, Theorems 5, 7 and 8 (after Eq. 47 and Eq. 57); also Proposition 4]
    "The following theorem, that is a straightforward extension of a results appeared in [32], offers an important result on the well-posedness of learning. Theorem 5. If the following coercivity conditions ... hold true then functional Γ, defined by Eq. 47, admits a minimum ... Proof. The proof is the same as the one in [32] once one observes that M(q) ≥ 0 ... Theorem 7. We can always choose the system parameters of Eq. (57) ... Proof. See [32] for the proof. Theorem 8. ... Proof. See [32] for the proof."

    The paper's guarantees that the cognitive action admits a minimum and that the null-video reset mechanism can satisfy the boundary conditions (52) are load-bearing: they are what make Eq. (53) a causal, well-posed learning law. Yet Theorems 5, 7 and 8 are not proved in this manuscript; they are all deferred to [32], a prior paper by the same authors. The abstract's promise of a 'well-posed computational scheme' therefore rests on a self-citation chain rather than on a proof contained here. This is not a minor citation: Section 6 explicitly uses Theorems 7–8 to glue the Cauchy initial-value problem to the boundary conditions through inserted null-video segments, and without [32] Eq. (53) lacks a supported existence-and-boundary argument in this text.

full rationale

The variational part of the paper is self-contained: once the cognitive action (21) is accepted as the objective, the Euler-Lagrange equations (27), the spatial-localization reduction (40), and the discrete fourth-order equation (53) follow by standard calculus of variations. Embedding the motion-invariance penalty in the action is a postulate, not a circular prediction; the paper does not claim to derive that principle from a deeper premise. The more serious issue is correctness rather than circularity: Section 5 asserts that replacing the mutual-information entropies by the quadratic surrogates (∫Φ)² and Φ² 'retains all the basic properties on the stationary points of the mutual information,' but supplies no proof. If that asserted equivalence fails, the equations derived in this paper optimize the surrogate objective, not the claimed information-theoretic one. I do not count that as circular because the variational computation is a consequence of the substituted objective, not an equation that reduces to its own input. The concrete circularity concern is the self-citation load-bearing chain: the existence of the minimum and the reset/boundary mechanism are imported verbatim from [32], an earlier paper by the same authors, so the well-posedness claim is not independently established in this manuscript. The experimental section, by contrast, is self-contained against external benchmarks (HOHA2, BDD100K) and does not involve fitting-then-predicting the same quantity. Overall, the central derivation retains independent mathematical content, but the well-posedness/reset machinery is carried by the authors' own prior work, which warrants a score of 4 rather than a clean 0–2.

Assumptions & free parameters 6 free parameters · 7 assumptions · 2 invented entities

The framework rests on several hand-chosen multipliers, an unproved quadratic surrogate for mutual information, the assumption of known optical flow, and an approximation of Green's functions by Gaussians. The reset mechanism is a practical hack to satisfy boundary conditions. The theory's contribution is therefore a principled objective, but the specific form of that objective is largely postulated.

free parameters (6)
  • lambda_M = varied over {0, 1e-8, 1e-6, 1e-4, 1e-2, 1, 1e2}; set to 1e-6 for ca-1L
    Weights the motion-invariance penalty in the cognitive action; selected based on validation MI, not the final task labels.
  • lambda_P, lambda_K = positive multipliers, values not fully specified
    Weights for spatial and temporal parsimony terms in Eq. (21); chosen by hand.
  • mu, nu, gamma, k = k in [1e-19, 1e-3], theta=1e-4
    Coefficients in the discrete regularization R(q) (Eq. 46-47); chosen to satisfy stability conditions of Proposition 4.
  • eta (blurring schedule) = 0.0005
    Controls the gradual introduction of video detail in Eq. (59); the blurring plan is a hand-chosen annealing schedule.
  • reset thresholds = epsilon_j = 300*n
    Derivative magnitudes that trigger a reset of the video signal to null; tuned to keep the dynamics well-behaved.
  • lambda_C, lambda_1, lambda_0 = lambda_C=1; lambda_1, lambda_0 positive
    Multipliers for entropy, normalization and positivity constraints in Eq. (21); soft constraints with hand-set weights.
assumptions (7)
  • domain assumption Ergodic-like factorization of the visual measure: f(x,t)=g(x-a(t))h(t)
    Introduced in Section 4 (Eq. 8) to replace the joint probability measure over space, time, and frames with a product that separates the focus-of-attention distribution from the temporal weighting. This factorization is essential for the variational formulation but is not derived from data.
  • ad hoc to paper The quadratic surrogate for mutual information preserves the stationary points of the original objective
    Section 5, after Eq. (21): the paper replaces Phi log Phi with Phi^2 and (int Phi)^2, asserting this 'retains all the basic properties on the stationary points of the mutual information' without proof. The entire Euler-Lagrange derivation depends on this replacement.
  • standard math Convolution is commutative on the bounded retina under the stated support assumptions
    Section 5, Eq. (23): commutativity requires the color field to have support in X_R and the filters in X_r with r<=R, with zero extension. This is a technical assumption needed for the variational simplifications.
  • ad hoc to paper G is a Green's function of a self-adjoint operator L with G(deltaX)=0, and a Gaussian approximates this
    Theorem 3 assumes the existence of such G and L; Theorem 4 shows a Gaussian and truncated differential operator satisfy the hypothesis only in the limit sigma->0. The localization of Eq. (37) into Eq. (40) relies on this approximation.
  • ad hoc to paper Injecting null-signal 'reset' segments does not change the information structure of the video and restores boundary conditions
    Section 6 introduces segments of null video to make the derivatives of q vanish at interval endpoints, asserting this is equivalent to the original visual environment. This manipulation is central to the causal solving of Eq. (53) but is not empirically validated.
  • domain assumption The optical flow field v is given and accurate
    The motion-invariance term (Eq. 16) and all subsequent equations assume a known velocity field. In experiments v is estimated with OpenCV, so its errors propagate into the learning dynamics.
  • domain assumption The temporal weighting h(t) is monotone increasing with specific exponential form
    Eq. (49) sets h(t)=theta exp(theta t)/(exp(theta T)-1) to create dissipation and a time direction; the well-posedness and boundary-condition arguments rely on this form.
invented entities (2)
  • Adjoint field Lambda_ij
    purpose: Introduced in Eq. (40) to convert the nonlocal integral equations into a local system of differential equations.
    A mathematical auxiliary with no independent physical or empirical handle; it is a computational device within the theory.
  • Motion invariance principle
    purpose: Postulates that visual features should be constant along optical-flow trajectories, serving as the core penalty of the cognitive action.
    The paper asserts this is the fundamental visual invariance, but it is not derived from more basic principles and no independent test is provided.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Learning Visual Features Under Motion Invariance." pith.science (2026). https://pith.science/paper/3IKPSBGD

@misc{pith2026190900350,
  author       = {Pith},
  title        = {Pith review of: Learning Visual Features Under Motion Invariance},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/3IKPSBGD}},
  note         = {Machine review of arXiv:1909.00350}
}
read the original abstract

Humans are continuously exposed to a stream of visual data with a natural temporal structure. However, most successful computer vision algorithms work at image level, completely discarding the precious information carried by motion. In this paper, we claim that processing visual streams naturally leads to formulate the motion invariance principle, which enables the construction of a new theory of learning that originates from variational principles, just like in physics. Such principled approach is well suited for a discussion on a number of interesting questions that arise in vision, and it offers a well-posed computational scheme for the discovery of convolutional filters over the retina. Differently from traditional convolutional networks, which need massive supervision, the proposed theory offers a truly new scenario for the unsupervised processing of video signals, where features are extracted in a multi-layer architecture with motion invariance. While the theory enables the implementation of novel computer vision systems, it also sheds light on the role of information-based principles to drive possible biological solutions.

Figures

Figures reproduced from arXiv: 1909.00350 by the authors.

Figure 1
Figure 1. Convolutional computation in a deep network. The input is processed by convolutional filters [PITH_FULL_IMAGE:figures/full_fig_p012_1.png] view at source ↗
Figure 2
Figure 2. Motion invariance in the feature extraction process. The symbol y1, that defines a fea￾tures at the beginning of motion (t = 0 s), must be coherently extracted during the movement — see the enforcement of the coherence requirement at t = 1, 2 s. computational structure with a homogeneous convolutional-based computation, which yields the features at the p convolutional layer. The theory proposed in this paper focuses… view at source ↗
Figure 3
Figure 3. Conceptual map of Sections 4–5. Each arrow indicate the main dependencies of each section. [PITH_FULL_IMAGE:figures/full_fig_p019_3.png] view at source ↗
Figures from the paper (7 more)
Figure 4
Figure 4. Figure 4: Without explicit constraints in the action that enforces probabilistic normalization the outcomes of [PITH_FULL_IMAGE:figures/full_fig_p037_4.png]
Figure 6
Figure 6. Figure 6: We selected the “stability, reality” configuration of Fig. 5, that fulfils the condi [PITH_FULL_IMAGE:figures/full_fig_p053_6.png]
Figure 5
Figure 5. Figure 5: Comparing 4 configurations of the parameters, characterized by different properties in terms of [PITH_FULL_IMAGE:figures/full_fig_p055_5.png]
Figure 6
Figure 6. Figure 6: Different number of features and filter sizes (1st column: [PITH_FULL_IMAGE:figures/full_fig_p056_6.png]
Figure 7
Figure 7. Figure 7: Three different blurring plans: slow, fast, faster (i.e., no bluring). We consider [PITH_FULL_IMAGE:figures/full_fig_p057_7.png]
Figure 8
Figure 8. Figure 8: Sample frames from the BDD100K dataset. Conclusion This paper proposes a learning theory that formalizes the problem of learning vi￾sual features from streams of visual data. The theory, which is based on the ergodic assumption of the video signal, provides a reformula…
Figure 9
Figure 9. Figure 9: Plots of ρσ when m = 1 for various values of σ: σ = 1 in (a), σ = 0.1 in (b), σ = 0.01 in (c) and σ = 0.001 in (d). 65 [PITH_FULL_IMAGE:figures/full_fig_p065_9.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

46 extracted references · 41 canonical work pages

  1. [32]

    Betti, M

    A. Betti, M. Gori, S. Melacci, Cognitive Action Laws: The Case of Visual Fea- tures, IEEE transactions on neural networks and learning systems online first

    work page

  2. [1]

    Marr, Vision, Freeman, San Francisco, 1982, partially reprinted in [46]

    D. Marr, Vision, Freeman, San Francisco, 1982, partially reprinted in [46]

    work page 1982

  3. [2]

    J. Deng, W. Dong, R. Socher, L.-J. Li, K. Li, L. Fei-Fei, ImageNet: A Large-Scale Hierarchical Image Database, in: CVPR09, 2009

    work page 2009

  4. [3]

    T. A. Poggio, F. Anselmi, Visual Cortex and Deep Networks: Learning Invariant Representations, 1st Edition, The MIT Press, 2016

    work page 2016

  5. [4]

    D. G. Lowe, Distinctive image features from scale-invariant keypoints, Int. J. Comput. Vision 60 (2) (2004) 91–110. doi:10.1023/B:VISI. 0000029664.99615.94. URL http://dx.doi.org/10.1023/B:VISI.0000029664.99615. 94

    arXiv 2004

  6. [5]

    B. K. Horn, B. Schunck, Determining optical flow, Artificial Intelligence 17 (1-3) (1981) 185–203. 68

    work page 1981

  7. [6]

    Baker, D

    S. Baker, D. Scharstein, J. P. Lewis, S. Roth, M. J. Black, R. Szeliski, A database and evaluation methodology for optical flow, Int. J. Comput. Vision 92 (1) (2011) 1–31. doi:10.1007/s11263-010-0390-2 . URL http://dx.doi.org/10.1007/s11263-010-0390-2

    work page doi:10.1007/s11263-010-0390-2 2011

  8. [7]

    M. Gori, M. Lippi, M. Maggini, S. Melacci, Semantic video labeling by devel- opmental visual agents, Computer Vision and Image Understanding 146 (2016) 9–26. doi:10.1016/j.cviu.2016.02.011. URL https://doi.org/10.1016/j.cviu.2016.02.011

    work page doi:10.1016/j.cviu.2016.02.011 2016

Show all 46 references
  1. [8]

    Betti, M

    A. Betti, M. Gori, The principle of least cognitive action, Theor. Comput. Sci. 633 (2016) 83–99. doi:10.1016/j.tcs.2015.06.042. URL https://doi.org/10.1016/j.tcs.2015.06.042

    2016 doi

  2. [9]

    Wiskott, T

    L. Wiskott, T. J. Sejnowski, Slow feature analysis: Unsupervised learning of in- variances, Neural computation 14 (4) (2002) 715–770

    2002

  3. [10]

    Wiskott, Slow feature analysis: A theoretical analysis of optimal free re- sponses, Neural Computation 15 (9) (2003) 2147–2177

    L. Wiskott, Slow feature analysis: A theoretical analysis of optimal free re- sponses, Neural Computation 15 (9) (2003) 2147–2177

    2003

  4. [11]

    L. Sun, K. Jia, T.-H. Chan, Y . Fang, G. Wang, S. Yan, Dl-sfa: deeply-learned slow feature analysis for action recognition, in: Proceedings of the IEEE conference on computer vision and pattern recognition, 2014, pp. 2625–2632

    2014

  5. [12]

    F. J. Huang, Y .-L. Boureau, Y . LeCun, et al., Unsupervised learning of invariant feature hierarchies with applications to object recognition, in: Computer Vision and Pattern Recognition, 2007. CVPR’07. IEEE Conference on, IEEE, 2007, pp. 1–8

    2007

  6. [13]

    Kavukcuoglu, P

    K. Kavukcuoglu, P. Sermanet, Y . lan Boureau, K. Gregor, M. Mathieu, Y . L. Cun, Learning convolutional feature hierarchies for visual recognition, in: J. D. Lafferty, C. K. I. Williams, J. Shawe-Taylor, R. S. Zemel, A. Culotta (Eds.), Ad- vances in Neural Information Processi...

    2010

  7. [14]

    X. Wang, A. Gupta, Unsupervised learning of visual representations using videos, in: The IEEE International Conference on Computer Vision (ICCV), 2015

    2015

  8. [15]

    Goroshin, J

    R. Goroshin, J. Bruna, J. Tompson, D. Eigen, Y . LeCun, Unsupervised learning of spatiotemporally coherent metrics, in: Proceedings of the IEEE international conference on computer vision, 2015, pp. 4086–4093

    2015

  9. [16]

    Y . Li, M. Paluri, J. M. Rehg, P. Doll´ar, Unsupervised learning of edges, in: Pro- ceedings of the IEEE Conference on Computer Vision and Pattern Recognition, 2016, pp. 1619–1627

    2016

  10. [17]

    Pathak, R

    D. Pathak, R. B. Girshick, P. Doll ´ar, T. Darrell, B. Hariharan, Learning features by watching objects move., in: CVPR, V ol. 1, 2017, p. 7

    2017

  11. [18]

    Tavanaei, T

    A. Tavanaei, T. Masquelier, A. S. Maida, Acquisition of visual features through probabilistic spike-timing-dependent plasticity, CoRR abs/1606.01102. arXiv: 1606.01102. URL http://arxiv.org/abs/1606.01102

    arXiv

  12. [19]

    H. Lee, R. Grosse, R. Ranganath, A. Y . Ng, Convolutional deep belief net- works for scalable unsupervised learning of hierarchical representations, in: Pro- ceedings of the 26th Annual International Conference on Machine Learning, ICML ’09, ACM, New York, NY , USA, 2009, pp. ...

    2009

  13. [20]

    Ranzato, F

    M. Ranzato, F. J. Huang, Y . Boureau, Y . LeCun, Unsupervised learning of invariant feature hierarchies with applications to object recognition, in: 2007 IEEE Computer Society Conference on Computer Vision and Pattern Recog- nition (CVPR 2007), 18-23 June 2007, Minneapolis, Mi...

    2007

  14. [21]

    Goroshin, J

    R. Goroshin, J. Bruna, J. Tompson, D. Eigen, Y . LeCun, Unsupervised learning of spatiotemporally coherent metrics, in: 2015 IEEE International Conference on 70 Computer Vision, ICCV 2015, Santiago, Chile, December 7-13, 2015, 2015, pp. 4086–4093. doi:10.1109/ICCV.2015.465. UR...

    2015 doi

  15. [22]

    M. A. Goodale, A. D. Milner, Separate visual pathways for perception and action, Trends in Neurosciences 15 (1) (1992) 20–25

    1992

  16. [23]

    Hubel, T

    D. Hubel, T. Wiesel, Receptive fields, binocular interaction, and functional ar- chitecture in the cat’s visual cortex, Journal of Physiology (London) 160 (1962) 106–154

    1962

  17. [24]

    D. V ., D. Y . Teller, Visual acuity in human infants: A review and comparison of behavioral and electrophysiological studies, Vision Research 18

  18. [25]

    B. O., J. Atkinson, Development of human visual function, Vision Research 51 (2011) 1588–1609

    2011

  19. [26]

    A. T., N. Y . N., C. Cirelli, G. Tononi, F. Itzhak, Single-neuron activity and eye movements during human rem sleep and awake vision, Nature October 2014

    2014

  20. [27]

    E. T. Jaynes, Information theory and statistical mechanics, The Physical Review 106 (1957) 620–630

    1957

  21. [28]

    Zanca, M

    D. Zanca, M. Gori, Variational laws of visual attention for dynamic scenes, in: NIPS, 2017, pp. 3826–3835

    2017

  22. [29]

    Zanca, S

    D. Zanca, S. Melacci, M. Gori, Gravitational laws of focus of attention, IEEE transactions on pattern analysis and machine intelligence

  23. [30]

    Gori, Machine Learning: A Constrained-Based Approach, Morgan Kauffman, 2018

    M. Gori, Machine Learning: A Constrained-Based Approach, Morgan Kauffman, 2018

    2018

  24. [31]

    Melacci, M

    S. Melacci, M. Gori, Unsupervised learning by minimal entropy encoding, IEEE Trans. Neural Netw. Learning Syst. 23 (12) (2012) 1849–1861.doi:10.1109/ TNNLS.2012.2216899. URL https://doi.org/10.1109/TNNLS.2012.2216899 71

    2012

  25. [33]

    K. A. Krueger, P. Dayan, Flexible shaping: How learning in small steps helps, Cognition 110 (3) (2009) 380–394

    2009

  26. [34]

    B. F. Skinner, Reinforcement today., American Psychologist 13 (3) (1958) 94

    1958

  27. [35]

    S. A. Huettel, P. B. Mack, G. McCarthy, Perceiving patterns in random series: dynamic processing of sequence in prefrontal cortex, Nature neuroscience 5 (5) (2002) 485–490

    2002

  28. [36]

    Moody, C

    J. Moody, C. Darken, Learning with localized receptive fields, in: D. Touretzky, G. Hinton, T. Sejnowski (Eds.), Proceedings of the 1988 Connectionist Models Summer School, Morgan Kaufmann, San Mateo, Pittsburg 1988, 1988, pp. 133– 143

    1988

  29. [37]

    Stokbro, D

    K. Stokbro, D. Umberger, J. Hertz, Exploiting neurons with localized receptive fields to learn chaos, Preprint 90/28 S, Nordita, Copenhagen, Denmark (1990)

    1990

  30. [38]

    D. H. Hubel, T. N. Wiesel, Receptive fields and functional architecture of monkey striate cortex, Journal of Physiology (London) 195 (1968) 215–243

    1968

  31. [39]

    W. Luo, Y . Li, R. Urtasun, R. S. Zemel, Understanding the effective receptive field in deep convolutional neural networks, CoRR abs/1701.04128

    arXiv

  32. [40]

    Marszałek, I

    M. Marszałek, I. Laptev, C. Schmid, Actions in context, in: IEEE Conference on Computer Vision & Pattern Recognition, 2009

    2009

  33. [41]

    F. Yu, W. Xian, Y . Chen, F. Liu, M. Liao, V . Madhavan, T. Darrell, Bdd100k: A diverse driving video database with scalable annotation tooling, arXiv preprint arXiv:1805.04687

    arXiv

  34. [42]

    F. Yu, V . Koltun, T. Funkhouser, Dilated residual networks, in: Proceedings of the IEEE conference on computer vision and pattern recognition, 2017, pp. 472–480. 72

    2017

  35. [43]

    Ronneberger, P

    O. Ronneberger, P. Fischer, T. Brox, U-net: Convolutional networks for biomedi- cal image segmentation, in: International Conference on Medical image comput- ing and computer-assisted intervention, Springer, 2015, pp. 234–241

    2015

  36. [44]

    W. Luo, J. Li, J. Yang, W. Xu, J. Zhang, Convolutional sparse autoencoders for image classification, IEEE transactions on neural networks and learning systems 29 (7) (2017) 3289–3294

    2017

  37. [45]

    Masci, U

    J. Masci, U. Meier, D. Cires ¸an, J. Schmidhuber, Stacked convolutional auto- encoders for hierarchical feature extraction, in: International Conference on Arti- ficial Neural Networks, Springer, 2011, pp. 52–59

    2011

  38. [46]

    Anderson, E

    J. Anderson, E. Rosenfeld (Eds.), Neurocomputing: Foundations of Research, MIT Press, Cambridge, 1988. 73

    1988

Pith tools

Reviewed August 14, 2026 · model on record in the stance chip above.