Pith. sign in

REVIEW 3 major objections 6 minor 27 references

Directed Structural Adaptation to Overcome Statistical Conflicts and Enable Continual Learning

T0 review · 3 major / 6 minor · reviewed 2026-08-11 · deepseek-v4-flash

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

Pith's one-line read Growing networks slip past statistical dead ends to learn new tasks

desk verdict Novel structural adaptation idea, but the guarantee is overclaimed and the experiments are too thin to back it. read the letter →

arxiv 2412.04190 v1 pith:D5WZSNHY submitted 2024-12-05 cs.LG cs.AI

Zeki Doruk Erden , Boi Faltings This is my paper · ORCID
classification cs.LGcs.AI
keywords directedstructuraladaptationcontinuallearningcatastrophicforgettingedge-nodeconversionmodulatorynodesnoveltydetectionstatisticalconflictsnetworkgrowth
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

The paper claims that networks need not be locked into fixed topologies. It proposes DIRAD, a growth procedure that adds edges and nodes only when existing parameters are exhausted, and in particular converts an edge whose per-sample gradients cancel out into a modulatory node whose deltas preserve each sample's direction, letting adaptation resume. On top of that it builds PREVAL, a continual-learning layer with no task labels: a second network predicts internal node states, mismatches flag new tasks, and new models are spawned and later selected by validation. If this works, continually learning systems could stop overwriting old knowledge and instead grow exactly the structure they need.

What carries the argument

The edge-node conversion (ENC): an edge whose immediate adaptive potential is exhausted (net gradient near zero) but whose total adaptive potential is nonzero is replaced by a new modulatory node with two multiplicative terms. The original source feeds term 0 with weight 1, the outgoing edge keeps the original weight, and term 1's bias and transfer function are set so the node is initially neutral; the term-1 deltas equal the original per-sample weight gradients. This transfers a scalar average that was stuck at zero into a vector of per-sample signals that can be aligned by additional edges, and repeated ENC chains recurse until some source correlates with the gradient. PREVAL's L1 network, a directed acyclic graph that predicts the states of L0's internal nodes from higher-level computation, is the machinery that converts the absence of task labels into novelty detection and model selection.

What would settle it

A streaming or interleaved learning setting where a new task arrives before L0/L1 stabilization: if the system cannot maintain performance or detect the new task, the block-wise assumption is shown to be load-bearing. Alternatively, a constructed batch where the per-sample gradients of an edge have zero covariance with every product of input activations, which would stall the ENC chain despite nonzero total adaptive potential.

Watch

Extended reading notes

Core claim

The central claim is that statistical conflicts within a batch, where the mean gradient is zero but individual sample gradients are nonzero, can be resolved by structural change rather than by more parameters or more data. The edge-node conversion takes an exhausted edge and replaces it with a two-term modulatory node; the per-sample weight gradients of the original edge become the deltas of the node's modulatory term, which can be aligned by finding a source that correlates with them. The paper shows that a chain of such conversions proceeds as long as any multiplicative combination of candidate sources correlates with the original gradient vector, a condition far weaker than zero mean gradient. This mechanism, DIRAD, yields solutions with tens of edges on a two-class MNIST task where a fixed fully connected network needs thousands. The PREVAL framework then stabilizes the task network, builds an L1 network that predicts the states of internal nodes, and uses prediction mismatches to detect new tasks and route samples to the right stored model without ever seeing task labels.

Load-bearing premise

The framework assumes that all data for one task arrives as a batch and that no new task appears until the current model has fully stabilized, so the setting is block-wise rather than a continuous stream.

Editorial extensions

If this is right

  • Systems using DIRAD could replace overparameterized fixed networks with much smaller grown ones on tasks where statistical conflicts dominate.
  • PREVAL's task detection and model routing provide a path to continual learning without task boundaries or labels.
  • Since L1 networks predict internal states, the approach offers a scalable way to detect novelty at the level of internal representations rather than just outputs.
  • The theoretical condition (nonzero covariance of some multiplicative input combination with the gradient) gives a target for proving or refuting optimality of the grown networks.

Reading between the lines

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

  • If the covariance condition can be shown to be equivalent to a global optimum condition, DIRAD would connect structural adaptation to statistical independence or optimal transport criteria; the authors leave this open.
  • The ENC mechanism suggests a general principle: when a parameter's average update is zero but per-sample updates are not, the right move is to promote that parameter to a state variable, a trick that could be applied beyond this architecture, for example in feature selection or mixture models.
  • PREVAL's L1 prediction errors could in principle be used as a confidence signal to abstain or request labels in human-in-the-loop settings, which the paper does not explore.
  • A testable extension: use PREVAL with different base learners, such as random forests, to see if novelty detection via internal-state prediction transfers beyond gradient-trained networks.
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 / 6 minor

Summary. The paper proposes DIRAD, a structural-adaptation method in which a network grows edges and modulatory nodes in a greedy, gradient-directed manner, and PREVAL, a task-label-free continual-learning wrapper that uses per-node prediction validation (L1 networks) to detect new tasks and select among per-task models. The central mechanism is edge-node conversion (ENC): when an edge's net batch gradient vanishes despite nonzero per-sample gradients (a "statistical conflict"), the edge is replaced by a modulatory two-term node that inherits the per-sample gradients as deltas, allowing aligned modulation to resume adaptation. The paper provides an illustrative signed-XOR run, reports single-task and three-task MNIST experiments (downscaled to 14x14, 8 runs) showing roughly 90% single-task accuracy and 67-71% accuracy after three tasks, with network sizes reportedly two orders of magnitude smaller than a small fully-connected baseline, and it argues that Eq. (21) is the condition under which ENC chains cannot escape exhaustion.

Significance. The core idea is genuinely interesting: transferring edge-weight gradients into the deltas of a modulatory node converts an unusable zero-mean gradient vector into a usable per-sample vector, and the signed-XOR demonstration in Figure 1 illustrates a concrete escape from a statistical conflict that a static network cannot resolve. The PREVAL decomposition into new-task detection and sample-to-model assignment is a clean and useful framing of task-free continual learning, and the paper is admirably transparent: it explicitly flags the block-wise batch assumption (footnote 4), the semi-formal character of the theoretical argument, the unverified connection of Eq. (21) to global optimality, and the intrinsic limits of task discernability. The appendix gives a nearly complete algorithmic and parametric specification (Algorithm 1; delta_min = 0.01, R1 = 5, R2 = 0.1, gamma = 2, Tconf = 1.5, TSV = 0.01, epsilon_IS = 0.2), which makes the experiments reproducible in principle, although no code or seed policy is supplied.

major comments (3)
  1. [Appendix Full Theoretical Description of DIRAD, Eq. (21)] The argument around Eq. (21) establishes only a necessary condition for failure, while the text claims sufficiency. The derivation shows that if Cov(prod_{x in A} a_x^m, dC^m/dw_ij) = 0 for every subset A of N, then every edge created in any recursive ENC chain has zero net gradient. That yields only the contrapositive: if some chain reaches a nonzero gradient, Eq. (21) must fail. It does not show that when Eq. (21) fails, the particular chain traversed by Algorithm 1 finds a nonzero gradient. Source selection in Algorithm 1 (and in the edge-generation rule it implements) maximizes |sum_m a_i^m delta_j^m| over single sources; when all single-source correlations vanish, ties are broken arbitrarily, and nothing prevents choosing a source i that belongs to no subset A with nonzero covariance. Since every product formed in every subsequent ENC chain then contains i, the chain can stall forever even though Eq. (21) fails. A concrete instance: three binary inputs x0, x1, x2 with x2 independent of (x0, x1) and mean zero, and per-sample gradients g^m = x0^m x1^m (a signed-XOR pattern); all singleton covariances Cov(xi, g) vanish while Cov(x0 x1, g) is nonzero, so Eq. (21) fails, yet any chain that starts with x2 only ever forms products containing x2, all of which have zero covariance with g, and adaptation never proceeds. The main-text claim (paragraph following Eq. (2)) that "the chain of ENC operations will continue, resulting eventually in nonzero net gradients" as long as Eq. (21) does not hold is therefore unsupported, and this bears directly on the abstract's "strong guarantees" claim. The appended hedge that the argument is semi-formal and limited by finite step sizes does not cover this gap, because the issue is logical rather than numerical. The fix is in scope: either state the result as a necessary condition, or modify the generative process to search over sequences of sources and prove a matching sufficiency result.
  2. [Experiments and Results, Tables 1-6] The reported experiments are too thin to support the breadth of the empirical claims. Tables 1-2 average 8 runs without any measure of dispersion; per-run Tables 3-6 show large variance, including complete failures on individual classes (e.g., TCP = 0.10 run 3 has 0.0 accuracy on both Task-2 classes; TCP = 0.20 run 4 ends with 0.0 on class 0 in T3). The parenthesized values in Table 1 that exclude runs with non-detected tasks remove 1-3 of 8 runs (the ND column), so the headline retention figures (e.g., ALL+3 = 1.01 for TCP = 0.10) are computed on a self-selected subset. No error bars or confidence intervals are given, and there is no comparison against any fixed-topology or continual-learning baseline, so the claims that DIRAD networks are "orders-of-magnitude simpler" and that PREVAL retention exceeds 85% per new task cannot be evaluated quantitatively. The complexity comparison in the text (6 nodes and 15 edges versus 3296 edges for a 16-hidden-unit fully connected net) is made against a hypothetical architecture rather than an actually trained fixed-topology network, and the acknowledged lower single-task accuracy (about 90% on downscaled MNIST relative to standard MNIST classifiers) should be reported side by side with the complexity numbers. Additionally, the number of confidently-predicted (CP) nodes per configuration is never reported, so the ratios N_conf/N_CP that drive the validation decisions have unknown scale.
  3. [Footnote 4 and Detailed Experimental Settings] The continual-learning claim is validated only under a block-wise assumption that is stated clearly in footnote 4 ("a batch for one task is available to the system until the end of its adaptation, and no new task is provided until system is stabilized for current one") and restated in the appendix. Under this assumption, the branch "if all models are stabilized" is always the one taken when a task changes, so the new-task detection mechanism is never exercised while a model is still adapting; the streaming case is untested. Since the abstract promises a workflow "without needing task labels anywhere," the main text should state that the experiments cover only the block-wise regime and should discuss what changes (e.g., how a partially stabilized L0/L1 pair would be re-opened) if a new task arrives mid-adaptation.
minor comments (6)
  1. [Appendix, Eq. (8) and Eq. (13)] Equation (8) contains a garbled index: "y in srci(y)" should read "y in src_i(x)", and the derivative in Eq. (13) is written "dC^m / z^m_{k,1}" but should be "dC^m / dz^m_{k,1}".
  2. [Figure 1 caption] The caption's notation "Ge" and "dC/dwe" is not defined in the caption and should be typeset as G_e = dC/dw_e; the four parenthesized signs per variable should be mapped explicitly to the four samples of the signed-XOR task, since the caption is otherwise hard to parse.
  3. [Eq. (2) and Eq. (21)] The covariance notation Cov(...) is used over an implicit empirical batch distribution even though the underlying quantities carry sample superscripts; the paper should state that the covariance is taken over the uniform batch distribution, in which the gradient has zero mean by the exhaustion condition.
  4. [References and Related Work] The related-work section does not cite any recent task-free or class-incremental continual-learning methods, which is surprising given that PREVAL is explicitly task-free; engaging that line of work would strengthen the positioning, and the reference list also contains formatting inconsistencies (e.g., "Castro, Liskov et al. 1999" in the text).
  5. [Table 2] Table 2 reports ratios that can exceed 1 (e.g., ALL+3 = 1.01 for TCP = 0.10), but the text never explains that values above 1 are possible or that the parenthesized and unparenthesized columns average over different run subsets; a footnote should clarify both points.
  6. [Detailed Experimental Settings] No random seeds or code are provided, so the 8-run averages in Tables 1-2 are not reproducible as reported; please specify a seed policy or release the implementation.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: DIRAD's derivation is self-contained; the Eq. 21 concern is a proof gap, not a circular reduction.

full rationale

I found no circular dependency in this paper. The theoretical condition in Eq. 21 is derived from the algorithm's own definitions, but it is presented as a design analysis of when the recursive ENC chain can proceed, not as a fitted prediction or as a result imported from the authors' prior work. The experimental accuracies are obtained by gradient descent and generative processes on held-out test data, with the PREVAL thresholds (notably TCP) varied over a range and reported, so no fitted parameter is renamed as a prediction. There are no load-bearing self-citations: the only supporting citation for the neutrality principle is an external book (Wagner 2011), and the predictive-coding references are explicitly offered only as a possible interpretation. The skeptical concern that Eq. 21 is necessary but not sufficient for the greedy ENC chain to find a nonzero gradient is a soundness/completeness objection, not a circularity: the derivation would be invalid if the sufficiency claim fails, but the claimed result is not equivalent to its inputs by construction. Accordingly, the paper receives a circularity score of 0.

Assumptions & free parameters 8 free parameters · 5 assumptions · 3 invented entities

No code or data is released, so none of the introduced components has independent falsifiable evidence outside the paper's own small-scale experiments. The hyperparameters are hand-chosen and affect the reported performance; the theoretical condition in Eq. (21) is a domain assumption whose validity is not checked on the data.

free parameters (8)
  • delta threshold (delta_min) = 0.01
    Deltas below this value are treated as zero when checking exhaustion and computing adaptive potentials (Appendix 'Supplementary Mechanisms'). It directly controls when edge generation and ENC are triggered.
  • ENC ratio parameter R1 = 5
    Used in Eq. (23) to decide when an edge's immediate adaptive potential is considered exhausted for ENC.
  • exhaustion ratio R2 = 0.1
    An edge is regarded as having exhausted immediate AP if |dC/dw| < R2 * |w| (Appendix). It speeds up adaptation by triggering generative processes earlier.
  • learning rate gamma = 2
    Parameter adaptation rate used in gradient updates (Appendix 'Detailed Experimental Settings').
  • acceptable output mismatch = 0.01 (99%)
    Mismatches below this magnitude at the output are treated as zero, effectively setting the stopping tolerance for L0 adaptation and capping achievable accuracy.
  • refraction period = 5 steps
    During this period after a generative process, no other GP can take place for a node/edge, limiting the speed of complexity increase.
  • stabilization window = 50 steps
    L0 and L1 networks are stabilized when no total prediction error decrease is observed for 50 steps, which determines how thoroughly each task is learned.
  • PREVAL thresholds Tconf, TSV, eps_IS = Tconf=1.5, TSV=0.01, eps_IS=0.2
    Control CP-node conflict detection, sample validation ratio, and batch validation margin. These are hand-chosen and directly affect task detection and model assignment.
assumptions (5)
  • domain assumption Network is a directed acyclic graph (DAG) throughout generative processes.
    Assumed in the description and enforced in implementation (Appendix 'Full Theoretical Description of DIRAD'). This restricts the class of representable functions and avoids recurrence.
  • domain assumption Neutrality of generative processes: no existing node's response or adaptive signal changes when a new edge or node is added.
    Core design principle used in the derivations of edge generation and ENC (Appendix sections). It ensures structural changes do not immediately harm performance.
  • domain assumption A non-zero covariance between the target edge gradient and at least one multiplicative combination of input states is required for adaptation to proceed (Eq. 21).
    The theoretical guarantee of ENC chains rests on this condition; the paper does not verify it on the datasets and notes that it is a theoretical condition limited by practical considerations.
  • ad hoc to paper Task batches are block-wise: a batch for one task is fully available until its model stabilizes, and no new task arrives during that period.
    Stated in footnote 4 and in the Appendix 'Detailed Experimental Settings'. This simplifies the continual learning scenario relative to online or streaming settings and is not standard for all continual learning benchmarks.
  • standard math Backpropagation and the multivariable chain rule for arbitrary DAGs are valid.
    Used throughout the gradient derivations in the appendix.
invented entities (3)
  • Modulatory node (two-term node with sigma0(z)=z and sigma1(z)=4/(1+e^{-Kz})-1)
    purpose: Realizes edge-node conversion to transfer per-sample edge gradients to node deltas and escape statistical conflicts.
    New architectural component introduced in this paper; its utility is only demonstrated on the paper's own MNIST toy experiments, with no external predictive handle.
  • L1 prediction network
    purpose: Predicts L0 node states for novelty detection and model validation.
    A secondary network that predicts internal states of the task network; no independent evidence is provided beyond the reported proof-of-concept.
  • Model (distinct network per task in PREVAL)
    purpose: Encapsulates a single task's response so new tasks can be handled without overwriting old ones.
    The notion of a 'model' as a separate network is defined by the authors; the paper shows it in experiments without external validation.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Directed Structural Adaptation to Overcome Statistical Conflicts and Enable Continual Learning." pith.science (2026). https://pith.science/paper/D5WZSNHY

@misc{pith2026241204190,
  author       = {Pith},
  title        = {Pith review of: Directed Structural Adaptation to Overcome Statistical Conflicts and Enable Continual Learning},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/D5WZSNHY}},
  note         = {Machine review of arXiv:2412.04190}
}
read the original abstract

Adaptive networks today rely on overparameterized fixed topologies that cannot break through the statistical conflicts they encounter in the data they are exposed to, and are prone to "catastrophic forgetting" as the network attempts to reuse the existing structures to learn new task. We propose a structural adaptation method, DIRAD, that can complexify as needed and in a directed manner without being limited by statistical conflicts within a dataset. We then extend this method and present the PREVAL framework, designed to prevent "catastrophic forgetting" in continual learning by detection of new data and assigning encountered data to suitable models adapted to process them, without needing task labels anywhere in the workflow. We show the reliability of the DIRAD in growing a network with high performance and orders-of-magnitude simpler than fixed topology networks; and demonstrate the proof-of-concept operation of PREVAL, in which continual adaptation to new tasks is observed while being able to detect and discern previously-encountered tasks.

Figures

Figures reproduced from arXiv: 2412.04190 by the authors.

Figure 1
Figure 1. A simplified illustrative case of the path of adap [PITH_FULL_IMAGE:figures/full_fig_p003_1.png] view at source ↗
Figure 2
Figure 2. Simplified representation of PREVAL flow. [PITH_FULL_IMAGE:figures/full_fig_p006_2.png] view at source ↗
Figure 3
Figure 3. Sample progress of adaptation and complexity on a [PITH_FULL_IMAGE:figures/full_fig_p006_3.png] view at source ↗

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

27 extracted references · 14 canonical work pages

  1. [1]

    , " * write output.state after.block = add.period write newline

    ENTRY address archivePrefix author booktitle chapter edition editor eid eprint howpublished institution isbn journal key month note number organization pages publisher school series title type volume year label extra.label sort.label short.list INTEGERS output.state before.all mid.sentence after.sentence after.block FUNCTION init.state.consts #0 'before.a...

  2. [2]

    write newline

    " write newline "" before.all 'output.state := FUNCTION n.dashify 't := "" t empty not t #1 #1 substring "-" = t #1 #2 substring "--" = not "--" * t #2 global.max substring 't := t #1 #1 substring "-" = "-" * t #2 global.max substring 't := while if t #1 #1 substring * t #2 global.max substring 't := if while FUNCTION word.in bbl.in capitalize " " * FUNCT...

  3. [3]

    Baker, B.; Gupta, O.; Naik, N.; and Raskar, R. 2016. Designing neural network architectures using reinforcement learning. arXiv preprint arXiv:1611.02167

    arXiv 2016

  4. [4]

    Castro, M.; Liskov, B.; et al. 1999. Practical byzantine fault tolerance. In OsDI, volume 99, 173--186

    work page 1999

  5. [5]

    Dai, X.; Yin, H.; and Jha, N. K. 2019. NeST: A neural network synthesis tool based on a grow-and-prune paradigm. IEEE Transactions on Computers, 68(10): 1487--1497

    work page 2019

  6. [6]

    Evci, U.; van Merrienboer, B.; Unterthiner, T.; Vladymyrov, M.; and Pedregosa, F. 2022. Gradmax: Growing neural networks using gradient information. arXiv preprint arXiv:2201.05125

    arXiv 2022

  7. [7]

    J.; and Barucca, P

    Fadda, E.; He, J.; Tessone, C. J.; and Barucca, P. 2022. Consensus formation on heterogeneous networks. EPJ Data Science, 11(1): 34

    work page 2022

  8. [8]

    A.; and Pascanu, R

    Hadsell, R.; Rao, D.; Rusu, A. A.; and Pascanu, R. 2020. Embracing change: Continual learning in deep neural networks. Trends in cognitive sciences, 24(12): 1028--1040

    work page 2020

Show all 27 references
  1. [9]

    J.; Wright, C

    Jacobson, M. J.; Wright, C. Q.; Jiang, N.; Rodriguez-Rivera, G.; and Xue, Y. 2022. Task Detection in Continual Learning via Familiarity Autoencoders. In 2022 IEEE International Conference on Systems, Man, and Cybernetics (SMC), 1--8. IEEE

    2022

  2. [10]

    A.; Milan, K.; Quan, J.; Ramalho, T.; Grabska-Barwinska, A.; et al

    Kirkpatrick, J.; Pascanu, R.; Rabinowitz, N.; Veness, J.; Desjardins, G.; Rusu, A. A.; Milan, K.; Quan, J.; Ramalho, T.; Grabska-Barwinska, A.; et al. 2017. Overcoming catastrophic forgetting in neural networks. Proceedings of the national academy of sciences, 114(13): 3521--3526

    2017

  3. [11]

    Kowaliw, T.; Bredeche, N.; Chevallier, S.; and Doursat, R. 2014. Artificial neurogenesis: An introduction and selective review. Growing Adaptive Machines: Combining Development and Learning in Artificial Neural Networks, 1--60

    2014

  4. [12]

    Li, K.; Li, H.; Wang, H.; An, H.; Lu, P.; Yi, P.; and Zhu, F. 2020. PoV: an efficient voting-based consensus algorithm for consortium blockchains. Frontiers in Blockchain, 3: 11

    2020

  5. [13]

    Liu, H.; Simonyan, K.; Vinyals, O.; Fernando, C.; and Kavukcuoglu, K. 2017. Hierarchical representations for efficient architecture search. arXiv preprint arXiv:1711.00436

    2017 arXiv

  6. [14]

    Liu, H.; Simonyan, K.; and Yang, Y. 2018. Darts: Differentiable architecture search. arXiv preprint arXiv:1806.09055

    2018 arXiv

  7. [15]

    Miikkulainen, R.; Liang, J.; Meyerson, E.; Rawal, A.; Fink, D.; Francon, O.; Raju, B.; Shahrzad, H.; Navruzyan, A.; Duffy, N.; et al. 2019. Evolving deep neural networks. In Artificial intelligence in the age of neural networks and brain computing, 293--312. Elsevier

    2019

  8. [16]

    Millidge, B.; Seth, A.; and Buckley, C. L. 2021. Predictive coding: a theoretical and experimental review. arXiv preprint arXiv:2107.12979

    2021 arXiv

  9. [17]

    Mitchell, R.; Mundt, M.; and Kersting, K. 2023. Self Expanding Neural Networks. arXiv preprint arXiv:2307.04526

    2023 arXiv

  10. [18]

    I.; Kemker, R.; Part, J

    Parisi, G. I.; Kemker, R.; Part, J. L.; Kanan, C.; and Wermter, S. 2019. Continual lifelong learning with neural networks: A review. Neural networks, 113: 54--71

    2019

  11. [19]

    A.; Clifton, D

    Pimentel, M. A.; Clifton, D. A.; Clifton, L.; and Tarassenko, L. 2014. A review of novelty detection. Signal processing, 99: 215--249

    2014

  12. [20]

    A.; Rabinowitz, N

    Rusu, A. A.; Rabinowitz, N. C.; Desjardins, G.; Soyer, H.; Kirkpatrick, J.; Kavukcuoglu, K.; Pascanu, R.; and Hadsell, R. 2016. Progressive neural networks. arXiv preprint arXiv:1606.04671

    2016 arXiv

  13. [21]

    Shin, R.; Packer, C.; and Song, D. 2018. Differentiable neural network architecture search

    2018

  14. [22]

    Spratling, M. W. 2017. A review of predictive coding algorithms. Brain and cognition, 112: 92--97

    2017

  15. [23]

    O.; Clune, J.; Lehman, J.; and Miikkulainen, R

    Stanley, K. O.; Clune, J.; Lehman, J.; and Miikkulainen, R. 2019. Designing neural networks through neuroevolution. Nature Machine Intelligence, 1(1): 24--35

    2019

  16. [24]

    V.; Montone, G.; and O’Regan, J

    Terekhov, A. V.; Montone, G.; and O’Regan, J. K. 2015. Knowledge transfer in deep block-modular neural networks. In Biomimetic and Biohybrid Systems: 4th International Conference, Living Machines 2015, Barcelona, Spain, July 28-31, 2015, Proceedings 4, 268--279. Springer

    2015

  17. [25]

    Wagner, A. 2011. The origins of evolutionary innovations: a theory of transformative change in living systems. OUP Oxford

    2011

  18. [26]

    Yoon, J.; Yang, E.; Lee, J.; and Hwang, S. J. 2017. Lifelong learning with dynamically expandable networks. arXiv preprint arXiv:1708.01547

    2017 arXiv

  19. [27]

    Zhuang, F.; Qi, Z.; Duan, K.; Xi, D.; Zhu, Y.; Zhu, H.; Xiong, H.; and He, Q. 2020. A comprehensive survey on transfer learning. Proceedings of the IEEE, 109(1): 43--76

    2020

Pith tools

Reviewed August 11, 2026 · model on record in the stance chip above.