pith. sign in

arxiv: 2606.00073 · v1 · pith:EKVANY6Tnew · submitted 2026-05-21 · 💻 cs.NE · cs.AI· cs.LG

Rare Events, Real Signals: Functional Ensembles as Units of Computation in Deep Spiking Networks

Aditi Aravind , Konstantinos Ladakis , Mario Alexios Savaglio , Stelios M. Smirnakis , Maria Papadopouli This is my paper

Pith reviewed 2026-06-30 15:58 UTC · model grok-4.3

classification 💻 cs.NE cs.AIcs.LG
keywords functional connectivityspiking neural networksfunctional ensemblescofiring eventsclass encodingdeep spiking networksresnet architecturesinformation transfer
0
0 comments X

The pith

Informative class encodings in spiking neural networks occur only during rare high-cofiring events of functional ensembles.

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

The paper introduces first-order functionally-connected ensembles in deep spiking networks, formed from significant correlations between adjacent layers. These ensembles' joint activity reliably forecasts later neuron responses in a size-dependent manner. Class-specific information appears consistently only when these ensembles cofire at high levels, which happens rarely. This points to computation being driven by infrequent but precise coordination rather than ongoing average firing. The structure depends on training, as it vanishes when weights are permuted, and breaks under input perturbations.

Core claim

First-order functionally-connected (1FC) groups are defined by a neuron's statistically significant pairwise correlations with neurons in the preceding layer. Their aggregate cofiring predicts downstream responses through a robust ReLU-like function whose gain increases with group size. Reliable encoding of the input class is observed exclusively during infrequent high 1FC cofiring events. These ensembles preserve functional connectivity principles from biological cortex, are shaped by learning, and their response profiles are disrupted by noise or adversarial inputs, particularly in early layers.

What carries the argument

The 1FC ensemble, formed via statistically significant pairwise correlations across layers, whose aggregate cofiring serves as the predictor for downstream activity and class encoding.

If this is right

  • Downstream responses follow a ReLU-like relationship from 1FC cofiring, with gain scaling by ensemble size.
  • High cofiring events are infrequent yet necessary for reliable class encoding.
  • Response profiles disrupt under random noise or adversarial perturbations, especially early and intermediate layers.
  • Functional connectivity structure emerges from learning and collapses under weight permutation.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • If the ensembles are the computational units, then network interventions could target cofiring patterns to alter behavior more precisely.
  • The concentration of signal in rare events implies that standard rate-based analyses may miss key computations in both artificial and biological networks.
  • This approach offers a way to perform high-resolution diagnostics on information pathways in trained models.
  • Similar ensemble-based encoding might explain efficient computation in biological sensory systems.

Load-bearing premise

Statistically significant pairwise correlations between neurons in adjacent layers define ensembles whose combined activity functions as an independent variable that predicts downstream neuronal responses.

What would settle it

Finding that class encoding remains reliable even during low 1FC cofiring periods, or that aggregate cofiring does not predict downstream activity in a consistent ReLU-like manner.

Figures

Figures reproduced from arXiv: 2606.00073 by Aditi Aravind, Konstantinos Ladakis, Maria Papadopouli, Mario Alexios Savaglio, Stelios M. Smirnakis.

Figure 1
Figure 1. Figure 1: Inter-Layer Functional Connectivity through Pairwise STTC. Pairwise STTC was computed between the spiking activity of neurons in consecutive layer pairs during inference over the test set of CIFAR100 dataset ("original" data). Columns correspond to layer pairs L2→L3 (green), L9→L10 (blue), and L16→L17 (orange); The final column shows the same analysis in mouse primary visual cortex (V1), computed across co… view at source ↗
Figure 2
Figure 2. Figure 2: Response of a Neuron as a Function of the Number of Cofiring of its Input 1FC group. Columns 1–3 show results from the SNN (for layer pairs L1→L2, L8→L9, L15→L16), and column 4 shows corresponding analyses from mouse primary visual cortex (V1). (A–D) Example response functions of single output neurons, showing firing probability as a function of the number of cofiring events in their respective input-layer… view at source ↗
Figure 3
Figure 3. Figure 3: Response characteristics as they change under different conditions. (A-C) Histograms of R2 values resulting from ReLU fits of neuronal responses as a function of their respective input￾layer 1FC group cofirings, for layer pairs L1→L2, L8→L9, L15→L16 for original input (black), random noise (maroon), and adversarial input (purple). (D-F) Histograms of the intersection points corresponding to the transition … view at source ↗
Figure 4
Figure 4. Figure 4: Slopes of response functions under different input conditions (A) Violin plots of neuronal response slopes for layer pairs L1→L2, L8→L9, and L15→L16. For each violin, the left half corresponds to neurons with MI below the 20th percentile, and the right half to neurons with MI above the 80th percentile. (B) Same as (A), but showing the difference in slopes between adversarial and original test, respectively… view at source ↗
Figure 5
Figure 5. Figure 5: Information encoding of neurons as a function of their 1FC cofiring. (A-Top) Heatmap of semantic rank similarity for L16 neurons in the top 20th percentile of class-wise mutual information (MI), shown as a function of L15-1FC group co-firing events under original input. Each bin indicates the probability that co-firing frames match the neuron’s preferred class. (A) The larger the L15-1FC cofiring, the high… view at source ↗
read the original abstract

We investigate how internal representations emerge across hierarchical processing systems by introducing a neuroscience-inspired framework for analyzing deep spiking neural networks (SNN) through the lens of functional connectivity. Drawing on concepts from systems neuroscience and information theory, we form the first-order functionally-connected (1FC) group of a neuron based on its statistically significant pairwise correlations with neurons from the previous layer of a trained SNN architecture. We then track its response properties during inference under various conditions. Our analysis shows that several principles of functional connectivity previously observed in biological cortex are preserved in spiking ResNet architectures. These 1FC ensembles display interesting properties: their aggregate cofiring reliably predicts downstream neuronal responses through a robust, ReLU-like input-output relationship, whose gain scales systematically with ensemble size. Reliable encoding of the presented class emerges only during high 1FC cofiring events, which themselves occur infrequently, indicating that informative representations are concentrated in rare but highly coordinated activity patterns. Under uniform random noise or adversarial perturbations, these response profiles are disrupted, particularly in early and intermediate layers. This enables a targeted high-resolution interrogation at specific nodes and pathways. We showed that the functional connectivity structure is shaped by learning and this structure breaks under weight permutation. These establish 1FC ensembles as a functionally meaningful substrate for input encoding and information transfer, with potential implications in designing targeted fine-grained diagnostics on the information flow.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

2 major / 1 minor

Summary. The paper introduces a neuroscience-inspired analysis framework for deep spiking neural networks that defines first-order functionally-connected (1FC) ensembles via statistically significant pairwise correlations between a neuron and the preceding layer. It reports that aggregate cofiring within these ensembles predicts downstream firing rates through a ReLU-like mapping whose gain increases with ensemble size, that class-selective encoding is reliable only during infrequent high-cofiring events, and that the connectivity structure is learning-dependent, disrupted by noise or adversarial inputs, and absent under weight permutation.

Significance. If the independence of 1FC ensembles from their defining correlations can be established with appropriate controls, the framework would supply a concrete, falsifiable method for identifying rare coordinated events as the primary carriers of class information in SNNs, offering a potential bridge between cortical functional-connectivity studies and mechanistic interpretability of artificial networks.

major comments (2)
  1. [Abstract] Abstract (paragraph on 1FC group formation): the definition of 1FC groups rests on the same pairwise-correlation threshold later used to identify cofiring events whose aggregate activity is claimed to act as an independent predictor of downstream responses; no surrogate-data tests, partial-correlation conditioning on earlier layers, or ablation against size-matched random groups are described, leaving open whether the reported ReLU-like mapping and predictive gain survive removal of transitive or shared-input effects.
  2. [Abstract] Abstract (claims on rarity and class encoding): the assertions that 'reliable encoding of the presented class emerges only during high 1FC cofiring events' and that 'informative representations are concentrated in rare but highly coordinated activity patterns' are presented without any reported quantitative metrics (R² values, p-values, sample sizes, or cross-validation procedures), which are required to assess whether the rarity claim is load-bearing or an artifact of post-hoc selection.
minor comments (1)
  1. The abstract refers to 'spiking ResNet architectures' without specifying depth, neuron model, or training dataset; these details are needed to evaluate reproducibility.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the detailed and constructive comments. We address each major point below and commit to revisions that strengthen the manuscript by adding the requested controls and metrics.

read point-by-point responses

  1. Referee: [Abstract] Abstract (paragraph on 1FC group formation): the definition of 1FC groups rests on the same pairwise-correlation threshold later used to identify cofiring events whose aggregate activity is claimed to act as an independent predictor of downstream responses; no surrogate-data tests, partial-correlation conditioning on earlier layers, or ablation against size-matched random groups are described, leaving open whether the reported ReLU-like mapping and predictive gain survive removal of transitive or shared-input effects.

    Authors: We agree that demonstrating independence from transitive and shared-input effects is essential for the framework's validity. In the revised manuscript we will add surrogate-data controls that shuffle spike times while preserving per-neuron rates, partial-correlation analyses conditioned on preceding-layer activity, and direct comparisons against size-matched random ensembles. These tests and their outcomes will be reported in the Methods and Results sections to confirm that the ReLU-like mapping and ensemble-size-dependent gain remain after removal of the identified confounds. revision: yes

  2. Referee: [Abstract] Abstract (claims on rarity and class encoding): the assertions that 'reliable encoding of the presented class emerges only during high 1FC cofiring events' and that 'informative representations are concentrated in rare but highly coordinated activity patterns' are presented without any reported quantitative metrics (R² values, p-values, sample sizes, or cross-validation procedures), which are required to assess whether the rarity claim is load-bearing or an artifact of post-hoc selection.

    Authors: We accept that explicit quantitative metrics are required to substantiate the rarity and selectivity claims. The revised manuscript will include R² values for the cofiring-to-response fits, p-values for statistical comparisons of class-encoding reliability across cofiring regimes, explicit sample sizes (number of events, trials, and networks), and any cross-validation details. These metrics will be added to the abstract and the corresponding results paragraphs. revision: yes

Circularity Check

0 steps flagged

No significant circularity in derivation chain

full rationale

The paper presents an empirical analysis framework for SNNs: 1FC groups are formed via statistically significant pairwise correlations with the prior layer, after which observational properties (aggregate cofiring, ReLU-like input-output mapping, gain scaling with size, rarity of informative events) are measured on held-out inference data. No equation or step equates the claimed functional relationships to the group-definition criterion by construction; the ReLU-like profile and downstream prediction are data-driven findings, not definitional identities or refitted parameters. The structure is shaped by learning (verified via permutation controls) and the method is self-contained against external benchmarks without load-bearing self-citation or ansatz smuggling.

Axiom & Free-Parameter Ledger

1 free parameters · 1 axioms · 1 invented entities

The analysis rests on the assumption that pairwise correlation significance defines meaningful functional groups and that the observed input-output mapping is not an artifact of the correlation definition itself.

free parameters (1)
  • statistical significance threshold for pairwise correlations
    Used to decide membership in each 1FC group; value not stated in abstract.
axioms (1)
  • domain assumption Statistically significant pairwise correlations between neurons in adjacent layers capture functionally relevant connectivity
    Invoked when forming the 1FC group from correlation structure.
invented entities (1)
  • 1FC ensemble no independent evidence
    purpose: Proposed unit of computation and information transfer
    Newly defined grouping whose properties are then measured; no independent evidence outside the correlation definition is supplied.

pith-pipeline@v0.9.1-grok · 5802 in / 1461 out tokens · 39320 ms · 2026-06-30T15:58:28.758231+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

63 extracted references · 28 canonical work pages · 7 internal anchors

  1. [1]

    There Will Be a Scientific Theory of Deep Learning

    Jamie Simon, Daniel Kunin, Alexander Atanasov, Enric Boix-Adserà, Blake Bordelon, Jeremy Cohen, Nikhil Ghosh, Florentin Guth, Arthur Jacot, Mason Kamb, et al. There will be a scientific theory of deep learning.arXiv preprint arXiv:2604.21691, 2026

    work page internal anchor Pith review Pith/arXiv arXiv 2026

  2. [2]

    The platonic representation hypothesis

    Minhyuk Huh, Brian Cheung, Tongzhou Wang, and Phillip Isola. The platonic representation hypothesis. InInternational Conference on Machine Learning (ICML), 2024

    2024

  3. [3]

    Network dissec- tion: Quantifying interpretability of deep visual representations

    David Bau, Bolei Zhou, Aditya Khosla, Aude Oliva, and Antonio Torralba. Network dissec- tion: Quantifying interpretability of deep visual representations. InProceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR), 2017

    2017

  4. [4]

    Similarity of neural network representations revisited

    Simon Kornblith, Mohammad Norouzi, Honglak Lee, and Geoffrey Hinton. Similarity of neural network representations revisited. InInternational Conference on Machine Learning (ICML), 2019

    2019

  5. [5]

    Representation learning: A review and new perspectives.IEEE Transactions on Pattern Analysis and Machine Intelligence, 35(8): 1798–1828, 2013

    Yoshua Bengio, Aaron Courville, and Pascal Vincent. Representation learning: A review and new perspectives.IEEE Transactions on Pattern Analysis and Machine Intelligence, 35(8): 1798–1828, 2013

    2013

  6. [6]

    Cerebral cortex: architecture, intracortical connections, motor projections

    Nó R de Lorente. Cerebral cortex: architecture, intracortical connections, motor projections. Physiology of the nervous system, pages 288–330, 1949

    1949

  7. [7]

    Markus Meister, Rachel O. L. Wong, Denis A. Baylor, and Carla J. Shatz. Synchronous bursts of action potentials in ganglion cells of the developing mammalian retina.Science, 252(5008):939–943, May 1991. ISSN 1095-9203. doi: 10.1126/science.2035024. URL http://dx.doi.org/10.1126/science.2035024

    work page doi:10.1126/science.2035024 1991

  8. [8]

    Abeles, H

    M. Abeles, H. Bergman, E. Margalit, and E. Vaadia. Spatiotemporal firing patterns in the frontal cortex of behaving monkeys.Journal of Neurophysiology, 70(4):1629–1638, October 1993. doi: 10.1152/jn.1993.70.4.1629. URLhttps://doi.org/10.1152/jn.1993.70.4.1629

    work page doi:10.1152/jn.1993.70.4.1629 1993

  9. [9]

    Neuronal assemblies: Single cortical neurons are obedient members of a huge orchestra.Biopolymers, 68(3):422–436, February 2003

    Amiram Grinvald, Amos Arieli, Misha Tsodyks, and Tal Kenet. Neuronal assemblies: Single cortical neurons are obedient members of a huge orchestra.Biopolymers, 68(3):422–436, February 2003. doi: 10.1002/bip.10273. URLhttps://doi.org/10.1002/bip.10273

    work page doi:10.1002/bip.10273 2003

  10. [10]

    The first stage of perception: growth of the assembly.The Organization of Behavior, 4(60):78–60, 1949

    Donald O Hebb. The first stage of perception: growth of the assembly.The Organization of Behavior, 4(60):78–60, 1949

    1949

  11. [11]

    Spontaneous and driven cortical activity: implications for computation.Curr Opin Neurobiol, 19(4):439–444, August 2009

    Dario L Ringach. Spontaneous and driven cortical activity: implications for computation.Curr Opin Neurobiol, 19(4):439–444, August 2009

    2009

  12. [12]

    Peterka, and Rafael Yuste

    Luis Carrillo-Reid, Weijian Yang, Yuki Bando, Darcy S. Peterka, and Rafael Yuste. Imprinting and recalling cortical ensembles.Science, 353(6300):691–694, August 2016. doi: 10.1126/ science.aaf7560. URLhttps://doi.org/10.1126/science.aaf7560

    work page doi:10.1126/science.aaf7560 2016

  13. [13]

    The structures and functions of correlations in neural population codes.Nature Reviews Neuroscience, 23(9), 2022

    Stefano Panzeri, Monica Moroni, Houman Safaai, and Christopher D Harvey. The structures and functions of correlations in neural population codes.Nature Reviews Neuroscience, 23(9), 2022. 11

    2022

  14. [14]

    Neural correlations, population coding and computation.Nature reviews neuroscience, 7(5), 2006

    Bruno B Averbeck, Peter E Latham, and Alexandre Pouget. Neural correlations, population coding and computation.Nature reviews neuroscience, 7(5), 2006

    2006

  15. [15]

    R. Yuste. From the neuron doctrine to neural networks.Nature Reviews Neuroscience, 2015. URLhttps://doi.org/10.1038/nrn3962

    work page doi:10.1038/nrn3962 2015

  16. [16]

    Attractor dynamics of network up states in the neocortex.Nature, 423(6937):283–288, May 2003

    Rosa Cossart, Dmitriy Aronov, and Rafael Yuste. Attractor dynamics of network up states in the neocortex.Nature, 423(6937):283–288, May 2003. ISSN 1476-4687. doi: 10.1038/nature01614. URLhttp://dx.doi.org/10.1038/nature01614

    work page doi:10.1038/nature01614 2003

  17. [17]

    Jae-eun Kang Miller, Inbal Ayzenshtat, Luis Carrillo-Reid, and Rafael Yuste. Visual stimuli recruit intrinsically generated cortical ensembles.Proceedings of the National Academy of Sci- ences of the United States of America, 111(38):E4053—61, September 2014. ISSN 0027-8424. doi: 10.1073/pnas.1406077111. URLhttps://europepmc.org/articles/PMC4183303

    work page doi:10.1073/pnas.1406077111 2014

  18. [18]

    Cell assemblies at multiple time scales with arbitrary lag constellations.Elife, 6:e19428, 2017

    Eleonora Russo and Daniel Durstewitz. Cell assemblies at multiple time scales with arbitrary lag constellations.Elife, 6:e19428, 2017

    2017

  19. [19]

    Neuronal ensembles: Building blocks of neural circuits.Neuron, 2024

    Rafael Yuste, Rosa Cossart, and Emre Yaksi. Neuronal ensembles: Building blocks of neural circuits.Neuron, 2024. ISSN 0896-6273. doi: https://doi.org/10.1016/j.neuron.2023.12.008. URLhttps://www.sciencedirect.com/science/article/pii/S0896627323009674

    work page doi:10.1016/j.neuron.2023.12.008 2024

  20. [20]

    Sponta- neously emerging cortical representations of visual attributes.Nature, 425(6961):954–956, Oct 2003

    Tal Kenet, Dmitri Bibitchkov, Misha Tsodyks, Amiram Grinvald, and Amos Arieli. Sponta- neously emerging cortical representations of visual attributes.Nature, 425(6961):954–956, Oct 2003. ISSN 1476-4687. doi: 10.1038/nature02078. URL https://doi.org/10.1038/ nature02078

    work page doi:10.1038/nature02078 2003

  21. [21]

    Artur Luczak, Peter Barthó, and Kenneth D. Harris. Spontaneous events outline the realm of possible sensory responses in neocortical populations.Neuron, 62(3):413–425, May 2009. doi: 10.1016/j.neuron.2009.03.014. URL https://doi.org/10.1016/j.neuron.2009.03. 014

    work page doi:10.1016/j.neuron.2009.03.014 2009

  22. [22]

    Meyer, and Stelios M

    Ganna Palagina, Jochen F. Meyer, and Stelios M. Smirnakis. Inhibitory units: An organizing nidus for feature-selective sub-networks in area v1.bioRxiv, 2018. doi: 10.1101/282988. URL https://www.biorxiv.org/content/early/2018/03/15/282988

    work page doi:10.1101/282988 2018

  23. [23]

    Brain orchestra under spontaneous conditions: Identifying communication modules from the func- tional architecture of area v1.bioRxiv, 2024

    Maria Papadopouli, Ioannis Smyrnakis, Emmanouil Koniotakis, Mario-Alexios Savaglio, Christina Brozi, Eleftheria Psilou, Ganna Palagina, and Stelios Manolis Smirnakis. Brain orchestra under spontaneous conditions: Identifying communication modules from the func- tional architecture of area v1.bioRxiv, 2024. doi: 10.1101/2024.02.29.582364. URL https://www.b...

    work page doi:10.1101/2024.02.29.582364 2024

  24. [24]

    Keliris, Ganna Palagina, Stelios M

    Mario Alexios Savaglio, Christina Brozi, Eleftheria Psilou, Chryssalenia Koumpouzi, Marianna Papadostefanaki, Christos Vasilakos, Spyros Nessis, Emmanouel Smyrnakis, Vasileios Akoutas, Georgios A. Keliris, Ganna Palagina, Stelios M. Smirnakis, and Maria Papadopouli. Direction of motion decoding in mouse v1: Neuron predictive power relates to functional co...

    2025

  25. [25]

    Smirnakis, and Maria Papadopouli

    Mario Alexios Savaglio, Christina Brozi, Stelios M. Smirnakis, and Maria Papadopouli. On tem- poral robustness & brain-state stability of functional connectivity in mouse primary visual area v1 compared to higher visual area al. InIEEE 25th International Conference on Bioinformatics and Bioengineering (BIBE) Best Paper Award, 2025

    2025

  26. [26]

    Decoding the brain: From neural representations to mechanistic models

    Mackenzie Weygandt Mathis, Adriana Perez Rotondo, Edward F Chang, Andreas S Tolias, and Alexander Mathis. Decoding the brain: From neural representations to mechanistic models. Cell, 187(21):5814–5832, 2024

    2024

  27. [27]

    Deep learning in spiking neural networks.Neural networks, 111:47–63, 2019

    Amirhossein Tavanaei, Masoud Ghodrati, Saeed Reza Kheradpisheh, Timothée Masquelier, and Anthony Maida. Deep learning in spiking neural networks.Neural networks, 111:47–63, 2019

    2019

  28. [28]

    From attribution maps to human-understandable explanations through concept relevance propagation.Nature machine intelligence, 5(9):1006– 1019, 2023

    Reduan Achtibat, Maximilian Dreyer, Ilona Eisenbraun, Sebastian Bosse, Thomas Wiegand, Wojciech Samek, and Sebastian Lapuschkin. From attribution maps to human-understandable explanations through concept relevance propagation.Nature machine intelligence, 5(9):1006– 1019, 2023. 12

    2023

  29. [29]

    How important is a neuron

    Kedar Dhamdhere, Mukund Sundararajan, and Qiqi Yan. How important is a neuron. In International Conference on Learning Representations, 2019. URL https://openreview. net/forum?id=SylKoo0cKm

    2019

  30. [30]

    Understanding the role of individual units in a deep neural network.Proceedings of the National Academy of Sciences, 117(48):30071–30078, 2020

    David Bau, Jun-Yan Zhu, Hendrik Strobelt, Agata Lapedriza, Bolei Zhou, and Antonio Torralba. Understanding the role of individual units in a deep neural network.Proceedings of the National Academy of Sciences, 117(48):30071–30078, 2020. doi: 10.1073/pnas.1907375117. URL https://www.pnas.org/doi/abs/10.1073/pnas.1907375117

    work page doi:10.1073/pnas.1907375117 2020

  31. [31]

    The building blocks of interpretability.Distill, 2018

    Chris Olah, Arvind Satyanarayan, Ian Johnson, Shan Carter, Ludwig Schubert, Katherine Ye, and Alexander Mordvintsev. The building blocks of interpretability.Distill, 2018. doi: 10.23915/distill.00010. https://distill.pub/2018/building-blocks

    work page doi:10.23915/distill.00010 2018

  32. [32]

    https://distill.pub/2020/circuits/zoom-in

    Chris Olah, Nick Cammarata, Ludwig Schubert, Gabriel Goh, Michael Petrov, and Shan Carter. Zoom in: An introduction to circuits.Distill, 2020. doi: 10.23915/distill.00024.001. https://distill.pub/2020/circuits/zoom-in

    work page doi:10.23915/distill.00024.001 2020

  33. [33]

    Locating and Editing Factual Associations in GPT

    Kevin Meng, David Bau, Alex Andonian, and Yonatan Belinkov. Locating and editing fac- tual associations in GPT.Advances in Neural Information Processing Systems, 36, 2022. arXiv:2202.05262

    work page internal anchor Pith review Pith/arXiv arXiv 2022

  34. [34]

    Learning Deep Features for Discriminative Localization

    Bolei Zhou, Aditya Khosla, Àgata Lapedriza, Aude Oliva, and Antonio Torralba. Learning deep features for discriminative localization.CoRR, abs/1512.04150, 2015. URL http: //arxiv.org/abs/1512.04150

    work page internal anchor Pith review Pith/arXiv arXiv 2015

  35. [35]

    Grad-CAM: Visual explanations from deep networks via gradient- based localization.Int

    Ramprasaath R Selvaraju, Michael Cogswell, Abhishek Das, Ramakrishna Vedantam, Devi Parikh, and Dhruv Batra. Grad-CAM: Visual explanations from deep networks via gradient- based localization.Int. J. Comput. Vis., 128(2):336–359, February 2020

    2020

  36. [36]

    Ablation-CAM: Visual explanations for deep convolutional network via gradient-free localization

    Saurabh Desai and Harish G Ramaswamy. Ablation-CAM: Visual explanations for deep convolutional network via gradient-free localization. In2020 IEEE Winter Conference on Applications of Computer Vision (WACV). IEEE, March 2020

    2020

  37. [37]

    Bowers, Gaurav Malhotra, Federico Adolfi, Marin Dujmovi´c, Milton L

    Jeffrey S. Bowers, Gaurav Malhotra, Federico Adolfi, Marin Dujmovi´c, Milton L. Montero, Valerio Biscione, Guillermo Puebla, John H. Hummel, and Rachel F. Heaton. On the importance of severely testing deep learning models of cognition.Cognitive Systems Research, 82:101158,

  38. [38]

    doi: 10.1016/j.cogsys.2023.101158

    work page doi:10.1016/j.cogsys.2023.101158 2023

  39. [39]

    Deep residual learning for image recognition

    Kaiming He, Xiangyu Zhang, Shaoqing Ren, and Jian Sun. Deep residual learning for image recognition. InProceedings of the IEEE conference on computer vision and pattern recognition, pages 770–778, 2016

    2016

  40. [40]

    Learning multiple layers of features from tiny images

    Alex Krizhevsky. Learning multiple layers of features from tiny images. Technical re- port, University of Toronto, 2009. URL https://api.semanticscholar.org/CorpusID: 18268744

    2009

  41. [41]

    Explaining and Harnessing Adversarial Examples

    Ian J. Goodfellow, Jonathon Shlens, and Christian Szegedy. Explaining and harnessing adver- sarial examples.CoRR, abs/1412.6572, 2014. URLhttps://api.semanticscholar.org/ CorpusID:6706414

    work page internal anchor Pith review Pith/arXiv arXiv 2014

  42. [42]

    Similarity of neural network models: A survey of functional and representational measures.ACM Comput

    Max Klabunde, Tobias Schumacher, Markus Strohmaier, and Florian Lemmerich. Similarity of neural network models: A survey of functional and representational measures.ACM Comput. Surv., 57(9):1–52, September 2025

    2025

  43. [43]

    Localizing Model Behavior with Path Patching

    Nicholas Goldowsky-Dill, Christopher Macleod, Lucas Jun Koba Sato, and Aryaman Arora. Localizing model behavior with path patching.ArXiv, abs/2304.05969, 2023. URL https: //api.semanticscholar.org/CorpusID:258079237

    work page internal anchor Pith review Pith/arXiv arXiv 2023

  44. [44]

    Bridging the black box: A survey on mechanistic interpretability in ai.ACM Comput

    Shriyank Somvanshi, Md Monzurul Islam, Amir Rafe, Anannya Ghosh Tusti, Arka Chakraborty, Anika Baitullah, Tausif Islam Chowdhury, Nawaf Alnawmasi, Anandi Dutta, and Subasish Das. Bridging the black box: A survey on mechanistic interpretability in ai.ACM Comput. Surv., 58 (8), February 2026. ISSN 0360-0300. doi: 10.1145/3787104. URL https://doi.org/10. 114...

    work page doi:10.1145/3787104 2026

  45. [45]

    Understanding intermediate layers using linear classifier probes

    Guillaume Alain and Yoshua Bengio. Understanding intermediate layers using linear clas- sifier probes.ArXiv, abs/1610.01644, 2016. URL https://api.semanticscholar.org/ CorpusID:9794990

    work page internal anchor Pith review Pith/arXiv arXiv 2016

  46. [46]

    Svcca: Singular vector canonical correlation analysis for deep learning dynamics and interpretability

    Maithra Raghu, Justin Gilmer, Jason Yosinski, and Jascha Sohl-Dickstein. Svcca: Singular vector canonical correlation analysis for deep learning dynamics and interpretability. In I. Guyon, U. V . Luxburg, S. Bengio, H. Wallach, R. Fergus, S. Vishwanathan, and R. Garnett, editors,Advances in Neural Information Processing Systems 30, pages 6076–6085. Curran...

    2017

  47. [47]

    Interpret neural networks by identifying critical data routing paths

    Yulong Wang, Hang Su, Bo Zhang, and Xiaolin Hu. Interpret neural networks by identifying critical data routing paths. In2018 IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 8906–8914. IEEE, June 2018

    2018

  48. [48]

    Visual explanations from spiking neural networks using inter-spike intervals.Sci

    Kim Youngeun and Panda Priyadarshini. Visual explanations from spiking neural networks using inter-spike intervals.Sci. Rep., 11(1):19037, September 2021

    2021

  49. [49]

    Correlations and neuronal population information.Annual review of neuroscience, 39, 2016

    Adam Kohn, Ruben Coen-Cagli, Ingmar Kanitscheider, and Alexandre Pouget. Correlations and neuronal population information.Annual review of neuroscience, 39, 2016

    2016

  50. [50]

    Carsen Stringer, Marius Pachitariu, Nicholas Steinmetz, Charu Bai Reddy, Matteo Carandini, and Kenneth D. Harris. Spontaneous behaviors drive multidimensional, brainwide activity. Science, 364(6437), April 2019. doi: 10.1126/science.aav7893. URLhttps://doi.org/10. 1126/science.aav7893

    work page doi:10.1126/science.aav7893 2019

  51. [51]

    Cortical areas interact through a communication subspace.Neuron, 102(1):249–259.e4, February 2019

    João D Semedo, Amin Zandvakili, Christian K Machens, Byron M Yu, and Adam Kohn. Cortical areas interact through a communication subspace.Neuron, 102(1):249–259.e4, February 2019

    2019

  52. [52]

    Coordinated neuronal ensembles in primary auditory cortical columns.eLife, 7:e35587, jun 2018

    Jermyn Z See, Craig A Atencio, Vikaas S Sohal, and Christoph E Schreiner. Coordinated neuronal ensembles in primary auditory cortical columns.eLife, 7:e35587, jun 2018. ISSN 2050-084X. doi: 10.7554/eLife.35587. URLhttps://doi.org/10.7554/eLife.35587

    work page doi:10.7554/elife.35587 2018

  53. [53]

    Alexander Bae, Mahaly Baptiste, Maya R

    J. Alexander Bae, Mahaly Baptiste, Maya R. Baptiste, Caitlyn A. Bishop, Agnes L. Bodor, Derrick Brittain, Victoria Brooks, JoAnn Buchanan, Daniel J. Bumbarger, Manuel A. Castro, Brendan Celii, Erick Cobos, Forrest Collman, Nuno Maçarico da Costa, Bethanny Danskin, Sven Dorkenwald, Leila Elabbady, Paul G. Fahey, Tim Fliss, Emmanouil Froudarakis, Jay Gager,...

    work page doi:10.1038/s41586-025-08790-w 2025

  54. [54]

    Using goal-driven deep learning models to understand sensory cortex.Nat

    Daniel L K Yamins and James J DiCarlo. Using goal-driven deep learning models to understand sensory cortex.Nat. Neurosci., 19(3):356–365, March 2016

    2016

  55. [55]

    Deep residual network predicts cortical representation and organization of visual features for rapid categorization.Sci

    Haiguang Wen, Junxing Shi, Wei Chen, and Zhongming Liu. Deep residual network predicts cortical representation and organization of visual features for rapid categorization.Sci. Rep., 8 (1):3752, February 2018. 14

    2018

  56. [56]

    Deep residual learning in spiking neural networks

    Wei Fang, Zhaofei Yu, Yanqi Chen, Tiejun Huang, Timothée Masquelier, and Yonghong Tian. Deep residual learning in spiking neural networks. In M. Ranzato, A. Beygelz- imer, Y . Dauphin, P.S. Liang, and J. Wortman Vaughan, editors,Advances in Neu- ral Information Processing Systems, volume 34, pages 21056–21069. Curran Associates, Inc., 2021. URL https://pr...

    2021

  57. [57]

    Incorporating learnable membrane time constant to enhance learning of spiking neural networks

    Wei Fang, Zhaofei Yu, Yanqi Chen, Timothee Masquelier, Tiejun Huang, and Yonghong Tian. Incorporating learnable membrane time constant to enhance learning of spiking neural networks. InProceedings of the IEEE/CVF International Conference on Computer Vision (ICCV), pages 2661–2671, October 2021

    2021

  58. [58]

    Cutts and Stephen J

    Catherine S. Cutts and Stephen J. Eglen. Detecting pairwise correlations in spike trains: An objective comparison of methods and application to the study of retinal waves.Journal of Neuroscience, 34(43):14288–14303, 2014. ISSN 0270-6474. doi: 10.1523/JNEUROSCI. 2767-14.2014. URLhttps://www.jneurosci.org/content/34/43/14288

    work page doi:10.1523/jneurosci 2014

  59. [59]

    Deep learning.nature, 521(7553):436–444, 2015

    Yann LeCun, Yoshua Bengio, and Geoffrey Hinton. Deep learning.nature, 521(7553):436–444, 2015

    2015

  60. [60]

    Deep sparse rectifier neural networks

    Xavier Glorot, Antoine Bordes, and Yoshua Bengio. Deep sparse rectifier neural networks. In Geoffrey Gordon, David Dunson, and Miroslav Dudík, editors,Proceedings of the Fourteenth International Conference on Artificial Intelligence and Statistics, volume 15 ofProceedings of Machine Learning Research, pages 315–323, Fort Lauderdale, FL, USA, 11–13 Apr 201...

    2011

  61. [61]

    Zeiler and Rob Fergus

    Matthew D. Zeiler and Rob Fergus. Visualizing and understanding convolutional networks. In David Fleet, Tomas Pajdla, Bernt Schiele, and Tinne Tuytelaars, editors,Computer Vision – ECCV 2014, pages 818–833, Cham, 2014. Springer International Publishing

    2014

  62. [62]

    Learning Transferable Visual Models From Natural Language Supervision

    Alec Radford, Jong Wook Kim, Chris Hallacy, Aditya Ramesh, Gabriel Goh, Sandhini Agar- wal, Girish Sastry, Amanda Askell, Pamela Mishkin, Jack Clark, Gretchen Krueger, and Ilya Sutskever. Learning transferable visual models from natural language supervision.CoRR, abs/2103.00020, 2021. URLhttps://arxiv.org/abs/2103.00020

    work page internal anchor Pith review Pith/arXiv arXiv 2021

  63. [63]

    ray fish

    Aditi Aravind. Sttcpy: Efficient spike time tiling coefficient calculations for large-scale spiking datasets, 2024. URLhttps://github.com/aditiaravind/STTCPy. 15 A Technical appendices and supplementary material A.1 SNN Architecture and Training Details From a practical standpoint, hierarchical representation learning necessitates sufficient network depth...

    2024