Pith. sign in

REVIEW 4 major objections 5 minor 83 references

AS-FedBridge: Pseudo-Spike Bridge Distillation for Heterogeneous ANN-SNN Federated Learning

T0 review · 4 major / 5 minor · reviewed 2026-08-05 · deepseek-v4-flash

Pith's one-line read AS-FedBridge claims a shared pseudo-spike bridge lets mixed ANN-SNN federations train together without collapsing.

desk verdict A competent first benchmark for mixed ANN-SNN federated learning with a plausible bridge mechanism, but single-seed results and an inconsistent ablation leave the core mechanism's causal role unproven. read the letter →

arxiv 2608.03324 v1 pith:KKVREICM submitted 2026-08-04 cs.LG cs.NE

classification cs.LGcs.NE
keywords federatedlearningspikingneuralnetworksANN-SNNheterogeneityknowledgedistillationpseudo-spikeinterfacenon-IIDdataneuromorphiccomputingrepresentationalignment
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 targets a setting no prior method handles: a federated learning system in which some clients run ordinary artificial neural networks with continuous activations and others run spiking neural networks that communicate in discrete binary spikes over time. It claims that direct parameter sharing or existing logit/prototype distillation fails here because the two representations live in incompatible spaces, and it proposes AS-FedBridge, a lightweight global Bridge with a Pseudo-Spike Interface that converts continuous Bridge activations into a discretized firing-rate-like form and aligns them with SNN firing rates. The Bridge is aggregated on the server while all backbones stay private, and each client both extracts knowledge from the Bridge and injects its own knowledge back. Across CIFAR-10, CIFAR-100, Tiny-ImageNet, and CIFAR10-DVS under non-IID partitions, the paper reports consistently higher accuracy than six heterogeneous-FL baselines, with CKA/MMD/gradient analyses showing that better ANN-SNN alignment tracks better accuracy. If these results hold, mixed ANN-SNN federations become a practical way to combine high-accuracy and energy-efficient edge clients without converting one model type into the other.

What carries the argument

The load-bearing object is the Pseudo-Spike Interface inside the shared Bridge: a small set of Bridge activations are converted by $r = [\mathrm{ReLU}(a)/s]_0^1$, quantized to the finite set $\{0,1/T,\dots,1\}$ with a straight-through gradient, and aligned to SNN firing rates through a mean-squared-error loss. PSPR ($L_{\mathrm{PSPR}}$) pushes each channel's mean toward a threshold $V_{\mathrm{th}}$ and keeps its variance above a floor, so the quantizer receives well-populated inputs. This interface is what makes the continuous Bridge spike-compatible for SNN clients while remaining differentiable for ANN optimization.

What would settle it

On the same CIFAR-100 non-IID federation, run AS-FedBridge with the discrete quantizer $Q_T$ replaced by a purely continuous rate projection, or with the firing-rate alignment loss zeroed. If accuracy stays within noise or cross-client CKA does not degrade, then the discrete pseudo-spike mechanism is not what carries the gain. A second observation: if after training the Bridge's pseudo-spike activations concentrate on a single level of $R_T$ (e.g., all values at 0 or 1), the PSPR-plus-rate-alignment claim would be falsified.

Watch

Extended reading notes

Core claim

The central discovery is that ANN-SNN representational misalignment is not a barrier to collaborative learning; it can be mediated by a small shared model. AS-FedBridge's Bridge has a globally aggregated body and per-client heads; SNN clients route selected Bridge activations through a Pseudo-Spike Interface that computes bounded rate variables $r = [\mathrm{ReLU}(a)/s]_0^1$, quantizes them to the discrete firing-rate support $R_T = \{0,1/T,\dots,1\}$ with a straight-through estimator, and matches these rates to the temporally averaged firing rates of the SNN at a designated bottleneck. A new regularizer, PSPR, keeps the pre-quantization distribution from collapsing by pulling the mean of ea

Load-bearing premise

The whole improvement rests on the assumption that the straight-through gradient through the discrete pseudo-spike projection stays informative and that a time-averaged SNN firing rate is a rich enough target for cross-client alignment; the paper itself notes it offers no formal guarantee against complete saturation of the quantized interface.

Editorial extensions

If this is right

  • A federation can contain both ANN and SNN clients and still outperform either homogeneous group on its own; the paper reports AS-FedBridge beating ANN-only, SNN-only, and binary-ANN homogeneous federations on CIFAR-100.
  • Representation alignment is not just a diagnostic: the paper's CKA/MMD/gradient analyses tie improved alignment to improved accuracy, implying that alignment metrics can guide federated aggregation.
  • Varying the SNN time step $T$ yields a controllable accuracy/compute trade-off without changing model parameters, giving edge deployments a direct dial between energy use and performance.
  • The method transmits only a 3.93 MB Bridge body per round, so it fits communication budgets where full-model aggregation is impossible.

Reading between the lines

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

  • Beyond the paper, the pseudo-spike interface is a general pattern: any federated system whose clients speak in different discrete codes could use a similar quantized bottleneck with straight-through gradients, not just ANN-SNN pairs.
  • A testable extension is to replace the time-averaged firing-rate target with richer SNN summaries such as spike timing or first-spike latency; if the paper's alignment story is right, such targets should further raise cross-client transfer.
  • The paper's own ablation suggests the interface mainly helps ANN clients receive SNN knowledge (removing it costs 4.20 ANN points but slightly helps SNN); a follow-up could deliberately strengthen the reverse path and look for further gains.
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

4 major / 5 minor

Summary. The paper proposes AS-FedBridge, a federated learning framework for mixed ANN-SNN clients. The central idea is to interpose a lightweight, server-aggregated Bridge between private ANN and SNN backbones, with a Pseudo-Spike Interface that projects continuous Bridge activations into spike-rate space, a Pseudo-Spike Polarization Regularizer (PSPR) that shapes the pre-quantization distribution, and bidirectional logit/rate distillation. The paper reports experiments on CIFAR-10, CIFAR-100, Tiny-ImageNet, and CIFAR10-DVS under Dirichlet non-IID partitions, comparing against several heterogeneous FL baselines, and claims consistent accuracy gains while mitigating scale, architecture, and client-composition heterogeneity. It also provides communication/computation cost analysis, ablations, and representation-alignment diagnostics (CKA, MMD, gradient cosine similarity).

Significance. If the claims hold, AS-FedBridge would be a meaningful contribution: it is, to my knowledge, the first systematic framework for mixed ANN-SNN federated learning, and the benchmark it establishes is valuable. The paper is thorough in matching baseline communication budgets, in costing computation via FLOPs/SOPs with explicit hardware-agnostic separation, and in providing multiple heterogeneity scenarios and post-training alignment diagnostics. The method itself is reasonably novel, combining distillation, a discrete pseudo-spike bottleneck, and a short-list of regularizers. However, the central empirical claim rests on single-seed comparisons and on a component-ablation table whose internal inconsistencies weaken causal attribution. The positive CKA/MMD/gradient correlations are suggestive but, as presented, descriptive rather than causal. The paper is promising but needs stronger statistical grounding and more careful causal language before the main claims are fully supported.

major comments (4)
  1. [B.1 (Common Experimental Settings); Tables 1, 2, 6] All experiments use one fixed seed (Section B.1: 'the random seed is fixed to 42'), and the main tables report single-run accuracy. Given the small margins over the strongest baselines (0.46–2.89 points in Table 1), the headline claim that AS-FedBridge 'consistently demonstrates' superior accuracy is not statistically pinned down. The internal inconsistency in Table 6 (see next comment) further suggests run-to-run variance is nontrivial. Please report mean and standard deviation over at least three independent seeds, and perform significance testing against the strongest baseline in each setting.
  2. [Table 6 (Additional Component Ablations)] The component ablation is internally inconsistent. Removing the entire Pseudo-spike Interface (PSPR + Q_T + L_rate) yields 69.83% avg, but removing only PSPR yields 69.25% avg — a strict superset removal hurts less than a subset removal. Similarly, removing Q_T alone gives 69.71% and removing L_rate alone gives 70.37%, while the full model gives 71.64%. Without error bars, these non-monotonic results are best explained by noise, and the claimed 1.81-point contribution of the Pseudo-spike Interface is not reliable. This is load-bearing because the method's novelty and the paper's main narrative both center on that interface. Please repeat the ablation with multiple seeds and report variance; if the non-monotonicity persists, discuss why removing a superset can improve over removing a component.
  3. [C.1 (Theoretical Role and Gradient of PSPR)] Section C.1 explicitly concedes: 'this provides a practical recovery path, although it does not constitute a formal guarantee against complete saturation.' The straight-through gradient in Eq. (46) vanishes when the ReLU output is non-positive or the normalized value is clipped at a boundary. The continuous logit branch can keep the Bridge trainable, but that means the pseudo-spike alignment component can silently stop contributing while the overall method still learns via CE/KD. Given that the ablation-based evidence for the component's causal role is weakened by the inconsistency in Table 6, the paper needs either (a) an empirical saturation analysis (e.g., fraction of pseudo-spike units with zero gradient during training, and how it evolves), or (b) a mechanism that avoids the saturation failure. As written, the central claim that the Pseudo-Spike Interface is what resolves ANN-SNN mi
  4. [Mechanism Analysis (Figures 4, 5, 8 and accompanying text)] The CKA, MMD, and gradient-cosine analyses are post-training diagnostics and are not used as training objectives. The text overstates their causal role: e.g., 'This analysis establishes a positive correlation where higher ANN-SNN similarity more-likely yields higher global accuracy, proving the absolute necessity of our shared Bridge' (Section Mechanism Analysis) and 'resolving severe optimization conflicts.' Correlation between alignment and accuracy does not prove that the Bridge is necessary or that alignment causes the gain. Please temper the causal language, or provide an intervention (e.g., an objective that explicitly optimizes one of these alignment measures) to support the causal claim.
minor comments (5)
  1. [Figure 5 caption / Section Mechanism Analysis] Typo in Figure 5 caption: 's, resolving severe optimization conflicts.'; delete the stray 's'. Also, 'Accruacy' appears in Figure 7 axis label.
  2. [Introduction (second paragraph)] The sentence 'This assumption completely breaks down in mixed ANN-SNN federations' has an orphaned 'andThis' in the text; fix the formatting and the missing space.
  3. [Abstract] The abstract claims 'robust performance gains' and 'consistent' superiority, but the single-seed evidence does not support robustness in the statistical sense. Adjust the wording to match the reported evidence level.
  4. [B.4 (Motivation and Mechanism Analysis)] The motivation experiment in Figure 1 uses 'sampled non-IID subsets of CIFAR-10/100' but the exact partition parameters (e.g., α, number of clients) are not given in the main text; please provide them or reference Section B.4 earlier.
  5. [Table 3 / Table 10] The two tables overlap heavily; consider merging or clearly cross-referencing them so the reader does not have to compare two large computation tables. Also, the 'near' column headers and the double asterisk footnote could be clarified.

Circularity Check

0 steps flagged · score 2.0 of 10

No significant circularity: the central claim is an external benchmark comparison; alignment diagnostics are post-training and not training objectives; self-citations are background/baselines only.

full rationale

AS-FedBridge's central claim is an empirical accuracy comparison against six heterogeneous-FL baselines plus standalone/centralized references on four datasets. Nothing in the method defines accuracy in terms of the alignment losses: Lrate (Eq. 5) and LPSPR (Eq. 6) are training objectives, and the CKA/MMD/gradient-cosine analyses (Figs. 4-5, Sec. A.3) are explicitly post-training diagnostics ("MMD2 is used only as a post-training diagnostic and is not included in the training objective"), so the observed alignment-accuracy correlation is not a fitted quantity renamed as a prediction. The pseudo-spike interface is not imported from a citation; it is defined by the paper's own equations, and the cited TET/NLD components are external. The paper cites prior work by overlapping authors (MH-pFLID, ANN-SNN conversion works), but only as a reproduced baseline and background, not as a uniqueness theorem or as justification for the bridge design. The admitted lack of a saturation guarantee (Sec. C.1) and the single-seed non-monotonicity of Table 6 are statistical/robustness concerns, not evidence that an output reduces to an input by construction. On the available text, the derivation is self-contained against external benchmarks, so circularity is negligible.

Assumptions & free parameters 8 free parameters · 4 assumptions · 2 invented entities

The central claim depends on a set of hand-chosen hyperparameters and on assumptions that the pseudo-spike quantization and firing-rate alignment transfer enough information across ANN-SNN clients. The paper's own ablations show the pseudo-spike interface is worth 1.81 points, so the performance hinges on that unproven mechanism.

free parameters (8)
  • Distillation temperature tau = 2.0
    Set manually in Table 11; standard in KD, no sensitivity analysis provided.
  • KD weight alpha_kd schedule = 0.10 to 0.025 (ANN), 0.07 to 0.025 (SNN)
    Chosen by hand with a schedule; no theoretical justification.
  • Bridge injection weights alpha_teach and alpha_ce = 0.16 to 0.05, 1.10 to 0.55
    Manually tuned; central to the bidirectional distillation objective (Eqs. 11-12).
  • Rate alignment weight alpha_rate = 0.005
    Chosen by hand; controls Lrate contribution to SNN injection (Eq. 12).
  • PSPR parameters alpha_pspr, lambda_var, V_th, k_d, sigma_min = 0.10, 1.0, 0.5, 1.0, 0.05
    Hand-set hyperparameters for the pseudo-spike regularizer (Eq. 6); no sensitivity analysis except Bridge width.
  • Proximal weight alpha_prox = 1e-4
    Set manually for Bridge injection regularization (Eq. 11-12).
  • SNN distillation weights lambda_nld and lambda_tet = 0.08, 1e-4
    Hand-chosen for the SNN extraction objective (Eq. 10).
  • Bridge width = 1.0x
    Default capacity; Table 9 shows sensitivity, so performance depends on this choice.
assumptions (4)
  • domain assumption The straight-through estimator gives usable gradients through the pseudo-spike quantization (Eq. 3, C.1).
    The paper states quantization does not block surrogate gradients but provides no formal guarantee against saturation.
  • domain assumption The temporally averaged SNN firing rate r_i is a sufficient target for aligning Bridge representations with spike-based clients (Eqs. 4-5).
    The method matches only the bottleneck rate statistics, assuming this captures cross-client transferable knowledge.
  • domain assumption The four benchmark datasets and the Dirichlet partition with alpha=0.1 represent realistic mixed ANN-SNN federated settings.
    The paper's empirical claims generalize from these datasets and partitions.
  • standard math Standard measures (CKA, MMD, gradient cosine) faithfully quantify representation alignment relevant to FL collaboration.
    These are accepted diagnostics, though their causal link to accuracy is asserted rather than derived.
invented entities (2)
  • Pseudo-Spike Interface
    purpose: Projects continuous Bridge activations into a discrete spike-rate-compatible space for alignment with SNN firing rates (Eqs. 1-3).
    A new architectural component; its utility is supported only by the paper's own ablations and experiments, with no external validation.
  • Shared Pseudo-Spike Bridge
    purpose: A lightweight mediator that aggregates global knowledge on the server and enables bidirectional distillation between ANN and SNN clients.
    The Bridge body is a new mediator concept in FL; no external evidence yet exists.

how reviews work

0 comments
Cite this review

Pith. "Pith review of AS-FedBridge: Pseudo-Spike Bridge Distillation for Heterogeneous ANN-SNN Federated Learning." pith.science (2026). https://pith.science/paper/KKVREICM

@misc{pith2026260803324,
  author       = {Pith},
  title        = {Pith review of: AS-FedBridge: Pseudo-Spike Bridge Distillation for Heterogeneous ANN-SNN Federated Learning},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/KKVREICM}},
  note         = {Machine review of arXiv:2608.03324}
}
read the original abstract

Federated learning enables collaborative model training across distributed edge devices while strictly preserving data privacy. To facilitate practical deployment on resource-constrained edge devices, Spiking Neural Networks (SNNs) have emerged as a promising alternative to traditional Artificial Neural Networks (ANNs) due to their sparse computing mechanisms and high energy efficiency. However, jointly training ANNs and SNNs exposes a challenge of representational misalignment, which is intrinsically caused by differences in information representation, specifically the semantic gap between continuous real-valued activations in ANNs and discrete spatio-temporal spikes in SNNs. To overcome this barrier, we propose AS-FedBridge, a novel federated learning framework tailored for mixed ANN-SNN clients. AS-FedBridge features a lightweight Bridge equipped with a Pseudo-Spike Interface, which effectively projects continuous signals into a spike-compatible space to facilitate ANN-SNN alignment. Given the absence of existing mixed ANN-SNN federated frameworks, we establish a comprehensive benchmark to evaluate against multiple advanced heterogeneous FL methods. Our empirical analysis demonstrates a positive correlation between the degree of ANN-SNN alignment and the collaborative FL performance. Across four datasets, AS-FedBridge consistently demonstrates advanced accuracy while mitigating extreme scale, architecture, and client heterogeneity challenge. Furthermore, our framework enables a highly controllable trade-off between model performance and resource efficiency. AS-FedBridge accomplishes these robust performance gains while introducing only marginal computational overhead, establishing a robust and practical foundation for mixed ANN-SNN federated learning systems.

Figures

Figures reproduced from arXiv: 2608.03324 by the authors.

Figure 1
Figure 1. Representation analysis of a mixed ANN-SNN fed [PITH_FULL_IMAGE:figures/full_fig_p001_1.png] view at source ↗
Figure 2
Figure 2. Overall pipeline of AS-FedBridge. The server aggregates only the shared Bridge body, while heterogeneous ANN [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗
Figure 3
Figure 3. SNN clients side bidirectional distillation in AS [PITH_FULL_IMAGE:figures/full_fig_p003_3.png] view at source ↗
Figures from the paper (5 more)
Figure 4
Figure 4. Figure 4: Representation alignment evaluated via cross-client CKA alongside model accuracy on CIFAR-10 and CIFAR-100. [PITH_FULL_IMAGE:figures/full_fig_p005_4.png]
Figure 5
Figure 5. Figure 5: Gradient cosine similarity alongside model accuracy on CIFAR-10 and CIFAR-100. AS-FedBridge ensures broadly [PITH_FULL_IMAGE:figures/full_fig_p005_5.png]
Figure 6
Figure 6. Figure 6: Layer-wise distributions of Bridge pseudo-spike [PITH_FULL_IMAGE:figures/full_fig_p005_6.png]
Figure 7
Figure 7. Figure 7: Impact of varying SNN client numbers on overall accuracy within a ten-client CIFAR-100 federation. [PITH_FULL_IMAGE:figures/full_fig_p007_7.png]
Figure 8
Figure 8. Figure 8: Feature-distribution alignment measured by Maximum Mean Discrepancy (MMD) and collaborative accuracy on IID [PITH_FULL_IMAGE:figures/full_fig_p012_8.png]

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

83 extracted references · 25 canonical work pages

  1. [1]

    Proceedings of the AAAI Conference on Artificial Intelligence , volume=

    SFedHIFI: Fire Rate-Based Heterogeneous Information Fusion for Spiking Federated Learning , author=. Proceedings of the AAAI Conference on Artificial Intelligence , volume=

  2. [2]

    IEEE Transactions on Signal Processing , volume=

    Federated learning with spiking neural networks , author=. IEEE Transactions on Signal Processing , volume=. 2021 , publisher=

  3. [3]

    Proceedings of the Int'l ACM Symposium on Mobility Management and Wireless Access , pages=

    HFedSNN: Efficient hierarchical federated learning using spiking neural networks , author=. Proceedings of the Int'l ACM Symposium on Mobility Management and Wireless Access , pages=

  4. [4]

    IEEE Transactions on Vehicular Technology , volume=

    Efficient federated learning with spike neural networks for traffic sign recognition , author=. IEEE Transactions on Vehicular Technology , volume=. 2022 , publisher=

  5. [5]

    Advances in Neural Information Processing Systems , volume=

    Temporal effective batch normalization in spiking neural networks , author=. Advances in Neural Information Processing Systems , volume=

  6. [6]

    Proceedings of the AAAI conference on artificial intelligence , volume=

    Going deeper with directly-trained larger spiking neural networks , author=. Proceedings of the AAAI conference on artificial intelligence , volume=

  7. [7]

    Exploiting Label Skewness for Spiking Neural Networks in Federated Learning

    Exploiting label skewness for spiking neural networks in federated learning , author=. arXiv preprint arXiv:2412.17305 , year=

  8. [8]

    Proceedings of the AAAI Conference on Artificial Intelligence , volume=

    A closer look at knowledge distillation in spiking neural network training , author=. Proceedings of the AAAI Conference on Artificial Intelligence , volume=

Show all 83 references
  1. [9]

    MH-pFLID: Model Heterogeneous personalized Federated Learning via Injection and Distillation for Medical Data Analysis , booktitle =

    Luyuan Xie and Manqing Lin and Tianyu Luan and Cong Li and Yuejian Fang and Qingni Shen and Zhonghai Wu , editor =. MH-pFLID: Model Heterogeneous personalized Federated Learning via Injection and Distillation for Medical Data Analysis , booktitle =. 2024 , url =

  2. [10]

    International conference on machine learning , pages=

    Exploiting shared representations for personalized federated learning , author=. International conference on machine learning , pages=. 2021 , organization=

  3. [11]

    Artificial intelligence and statistics , pages=

    Communication-efficient learning of deep networks from decentralized data , author=. Artificial intelligence and statistics , pages=. 2017 , organization=

  4. [12]

    Proceedings of Machine learning and systems , volume=

    Federated optimization in heterogeneous networks , author=. Proceedings of Machine learning and systems , volume=

  5. [13]

    International conference on machine learning , pages=

    Scaffold: Stochastic controlled averaging for federated learning , author=. International conference on machine learning , pages=. 2020 , organization=

  6. [14]

    Advances in neural information processing systems , volume=

    Tackling the objective inconsistency problem in heterogeneous federated optimization , author=. Advances in neural information processing systems , volume=

  7. [15]

    arXiv preprint arXiv:2003.00295 , year=

    Adaptive federated optimization , author=. arXiv preprint arXiv:2003.00295 , year=

  8. [16]

    arXiv preprint arXiv:2111.04263 , year=

    Federated learning based on dynamic regularization , author=. arXiv preprint arXiv:2111.04263 , year=

  9. [17]

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

    Model-contrastive federated learning , author=. Proceedings of the IEEE/CVF conference on computer vision and pattern recognition , pages=

  10. [18]

    arXiv preprint arXiv:2102.07623 , year=

    Fedbn: Federated learning on non-iid features via local batch normalization , author=. arXiv preprint arXiv:2102.07623 , year=

  11. [19]

    International conference on machine learning , pages=

    Ditto: Fair and robust federated learning through personalization , author=. International conference on machine learning , pages=. 2021 , organization=

  12. [20]

    arXiv preprint arXiv:1912.00818 , year=

    Federated learning with personalization layers , author=. arXiv preprint arXiv:1912.00818 , year=

  13. [21]

    Foundations and trends in machine learning , volume=

    Advances and open problems in federated learning , author=. Foundations and trends in machine learning , volume=. 2021 , publisher=

  14. [22]

    arXiv preprint arXiv:1910.03581 , year=

    Fedmd: Heterogenous federated learning via model distillation , author=. arXiv preprint arXiv:1910.03581 , year=

  15. [23]

    arXiv preprint arXiv:2006.16765 , year=

    Federated mutual learning , author=. arXiv preprint arXiv:2006.16765 , year=

  16. [24]

    Proceedings of the AAAI conference on artificial intelligence , volume=

    Fedproto: Federated prototype learning across heterogeneous clients , author=. Proceedings of the AAAI conference on artificial intelligence , volume=

  17. [25]

    Proceedings of the AAAI conference on artificial intelligence , volume=

    Fedtgp: Trainable global prototypes with adaptive-margin-enhanced contrastive learning for data and model heterogeneity in federated learning , author=. Proceedings of the AAAI conference on artificial intelligence , volume=

  18. [26]

    9th International Conference on Learning Representations,

    Enmao Diao and Jie Ding and Vahid Tarokh , title =. 9th International Conference on Learning Representations,. 2021 , url =

  19. [27]

    Stich and Martin Jaggi , editor =

    Tao Lin and Lingjing Kong and Sebastian U. Stich and Martin Jaggi , editor =. Ensemble Distillation for Robust Model Fusion in Federated Learning , booktitle =. 2020 , url =

  20. [28]

    FedBE: Making Bayesian Model Ensemble Applicable to Federated Learning , booktitle =

    Hong. FedBE: Making Bayesian Model Ensemble Applicable to Federated Learning , booktitle =. 2021 , url =

  21. [29]

    Data-Free Knowledge Distillation for Heterogeneous Federated Learning , booktitle =

    Zhuangdi Zhu and Junyuan Hong and Jiayu Zhou , editor =. Data-Free Knowledge Distillation for Heterogeneous Federated Learning , booktitle =. 2021 , url =

  22. [30]

    Nature communications , volume=

    Communication-efficient federated learning via knowledge distillation , author=. Nature communications , volume=. 2022 , publisher=

  23. [31]

    Advances in neural information processing systems , volume=

    Fedrolex: Model-heterogeneous federated learning with rolling sub-model extraction , author=. Advances in neural information processing systems , volume=

  24. [32]

    Advances in Neural Information Processing Systems , volume=

    Fjord: Fair and accurate federated learning under heterogeneous targets with ordered dropout , author=. Advances in Neural Information Processing Systems , volume=

  25. [33]

    The Eleventh International Conference on Learning Representations , year=

    Depthfl: Depthwise federated learning for heterogeneous clients , author=. The Eleventh International Conference on Learning Representations , year=

  26. [34]

    Advances in Neural Information Processing Systems , volume=

    FedFree: Breaking knowledge-sharing barriers through layer-wise alignment in heterogeneous federated learning , author=. Advances in Neural Information Processing Systems , volume=

  27. [35]

    Advances in Neural Information Processing Systems , volume=

    Feature distillation is the better choice for model-heterogeneous federated learning , author=. Advances in Neural Information Processing Systems , volume=

  28. [36]

    Nature , volume=

    Towards spike-based machine intelligence with neuromorphic computing , author=. Nature , volume=. 2019 , publisher=

  29. [37]

    IEEE Signal Processing Magazine , volume=

    Surrogate gradient learning in spiking neural networks , author=. IEEE Signal Processing Magazine , volume=. 2019 , publisher=

  30. [38]

    Frontiers in neuroscience , volume=

    Spatio-temporal backpropagation for training high-performance spiking neural networks , author=. Frontiers in neuroscience , volume=. 2018 , publisher=

  31. [39]

    Advances in neural information processing systems , volume=

    Long short-term memory and learning-to-learn in networks of spiking neurons , author=. Advances in neural information processing systems , volume=

  32. [40]

    Advances in neural information processing systems , volume=

    Slayer: Spike layer error reassignment in time , author=. Advances in neural information processing systems , volume=

  33. [41]

    Frontiers in neuroscience , volume=

    Training deep spiking neural networks using backpropagation , author=. Frontiers in neuroscience , volume=. 2016 , publisher=

  34. [42]

    arXiv preprint arXiv:2005.01807 , year=

    Enabling deep spiking neural networks with hybrid conversion and spike timing dependent backpropagation , author=. arXiv preprint arXiv:2005.01807 , year=

  35. [43]

    Frontiers in neuroscience , volume=

    Going deeper in spiking neural networks: VGG and residual architectures , author=. Frontiers in neuroscience , volume=. 2019 , publisher=

  36. [44]

    Frontiers in neuroscience , volume=

    Conversion of continuous-valued deep networks to efficient event-driven networks for image classification , author=. Frontiers in neuroscience , volume=. 2017 , publisher=

  37. [45]

    Ieee micro , volume=

    Loihi: A neuromorphic manycore processor with on-chip learning , author=. Ieee micro , volume=. 2018 , publisher=

  38. [46]

    Science , volume=

    A million spiking-neuron integrated circuit with a scalable communication network and interface , author=. Science , volume=. 2014 , publisher=

  39. [47]

    2009 , publisher=

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

  40. [48]

    2009 IEEE conference on computer vision and pattern recognition , pages=

    Imagenet: A large-scale hierarchical image database , author=. 2009 IEEE conference on computer vision and pattern recognition , pages=. 2009 , organization=

  41. [49]

    Frontiers in neuroscience , volume=

    Cifar10-dvs: an event-stream dataset for object classification , author=. Frontiers in neuroscience , volume=. 2017 , publisher=

  42. [50]

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

    A low power, fully event-based gesture recognition system , author=. Proceedings of the IEEE conference on computer vision and pattern recognition , pages=

  43. [51]

    IEEE transactions on pattern analysis and machine intelligence , volume=

    Event-based vision: A survey , author=. IEEE transactions on pattern analysis and machine intelligence , volume=. 2020 , publisher=

  44. [52]

    International conference on machine learning , pages=

    Similarity of neural network representations revisited , author=. International conference on machine learning , pages=. 2019 , organization=

  45. [53]

    arXiv preprint arXiv:1503.02531 , year=

    Distilling the knowledge in a neural network , author=. arXiv preprint arXiv:1503.02531 , year=

  46. [54]

    arXiv preprint arXiv:1602.02830 , year=

    Binarized neural networks: Training deep neural networks with weights and activations constrained to+ 1 or-1 , author=. arXiv preprint arXiv:1602.02830 , year=

  47. [55]

    European conference on computer vision , pages=

    Xnor-net: Imagenet classification using binary convolutional neural networks , author=. European conference on computer vision , pages=. 2016 , organization=

  48. [56]

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

    Quantization and training of neural networks for efficient integer-arithmetic-only inference , author=. Proceedings of the IEEE conference on computer vision and pattern recognition , pages=

  49. [57]

    arXiv preprint arXiv:1412.7024 , year=

    Training deep neural networks with low precision multiplications , author=. arXiv preprint arXiv:1412.7024 , year=

  50. [58]

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

    Recdis-snn: Rectifying membrane potential distribution for directly training spiking neural networks , author=. Proceedings of the IEEE/CVF conference on computer vision and pattern recognition , pages=

  51. [59]

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

    A hybrid ANN-SNN architecture for low-power and low-latency visual perception , author=. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages=

  52. [60]

    arXiv preprint arXiv:2303.04347 , year=

    Optimal ANN-SNN conversion for high-accuracy and ultra-low-latency spiking neural networks , author=. arXiv preprint arXiv:2303.04347 , year=

  53. [61]

    Advances in neural information processing systems , volume=

    Deep residual learning in spiking neural networks , author=. Advances in neural information processing systems , volume=

  54. [62]

    arXiv preprint arXiv:2105.11654 , year=

    Optimal ANN-SNN conversion for fast and accurate inference in deep spiking neural networks , author=. arXiv preprint arXiv:2105.11654 , year=

  55. [63]

    Proceedings of the AAAI conference on artificial intelligence , volume=

    Reducing ann-snn conversion error through residual membrane potential , author=. Proceedings of the AAAI conference on artificial intelligence , volume=

  56. [64]

    International Conference on Machine Learning , pages=

    A unified optimization framework of ann-snn conversion: towards optimal mapping from activation values to firing rates , author=. International Conference on Machine Learning , pages=. 2023 , organization=

  57. [65]

    arXiv preprint arXiv:2506.01968 , year=

    Efficient ann-snn conversion with error compensation learning , author=. arXiv preprint arXiv:2506.01968 , year=

  58. [66]

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

    Constructing deep spiking neural networks from artificial neural networks with knowledge distillation , author=. Proceedings of the IEEE/CVF conference on computer vision and pattern recognition , pages=

  59. [67]

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

    Temporal separation with entropy regularization for knowledge distillation in spiking neural networks , author=. Proceedings of the Computer Vision and Pattern Recognition Conference , pages=

  60. [68]

    arXiv preprint arXiv:2401.11824 , year=

    Rethinking centered kernel alignment in knowledge distillation , author=. arXiv preprint arXiv:2401.11824 , year=

  61. [69]

    ACM Transactions on Internet of Things , volume=

    A survey of on-device machine learning: An algorithms and learning theory perspective , author=. ACM Transactions on Internet of Things , volume=. 2021 , publisher=

  62. [70]

    ACM SIGARCH computer architecture news , volume=

    Hitting the memory wall: Implications of the obvious , author=. ACM SIGARCH computer architecture news , volume=. 1995 , publisher=

  63. [71]

    Proceedings of the IEEE , volume=

    Efficient processing of deep neural networks: A tutorial and survey , author=. Proceedings of the IEEE , volume=. 2017 , publisher=

  64. [72]

    arXiv preprint arXiv:2202.11946 , year=

    Temporal efficient training of spiking neural network via gradient re-weighting , author=. arXiv preprint arXiv:2202.11946 , year=

  65. [73]

    Communication, Simulation, and Intelligent Agents: Implications of Personal Intelligent Machines for Medical Education

    Clancey, William J. Communication, Simulation, and Intelligent Agents: Implications of Personal Intelligent Machines for Medical Education. Proceedings of the Eighth International Joint Conference on Artificial Intelligence (IJCAI-83)

  66. [74]

    Classification Problem Solving

    Clancey, William J. Classification Problem Solving. Proceedings of the Fourth National Conference on Artificial Intelligence

  67. [75]

    , title =

    Robinson, Arthur L. , title =. 1980 , doi =. https://science.sciencemag.org/content/208/4447/1019.full.pdf , journal =

  68. [76]

    New Ways to Make Microcircuits Smaller---Duplicate Entry

    Robinson, Arthur L. New Ways to Make Microcircuits Smaller---Duplicate Entry. Science

  69. [77]

    Clancey and Glenn Rennels , abstract =

    Diane Warner Hasling and William J. Clancey and Glenn Rennels , abstract =. Strategic explanations for a diagnostic consultation system , journal =. 1984 , issn =. doi:https://doi.org/10.1016/S0020-7373(84)80003-6 , url =

  70. [78]

    and Rennels, Glenn R

    Hasling, Diane Warner and Clancey, William J. and Rennels, Glenn R. and Test, Thomas. Strategic Explanations in Consultation---Duplicate. The International Journal of Man-Machine Studies

  71. [79]

    Poligon: A System for Parallel Problem Solving

    Rice, James. Poligon: A System for Parallel Problem Solving

  72. [80]

    Transfer of Rule-Based Expertise through a Tutorial Dialogue

    Clancey, William J. Transfer of Rule-Based Expertise through a Tutorial Dialogue

  73. [81]

    The Engineering of Qualitative Models

    Clancey, William J. The Engineering of Qualitative Models

  74. [82]

    2023 , eprint=

    Attention Is All You Need , author=. 2023 , eprint=

  75. [83]

    Pluto: The 'Other' Red Planet

    NASA. Pluto: The 'Other' Red Planet

Pith tools

Reviewed August 5, 2026 · model on record in the stance chip above.