pith. sign in

arxiv: 2603.27492 · v1 · submitted 2026-03-29 · 💻 cs.RO · cs.AI· cs.HC· cs.LG

Copilot-Assisted Second-Thought Framework for Brain-to-Robot Hand Motion Decoding

Yizhe Li (1) , Shixiao Wang (1) , Jian K. Liu (1) ((1) University of Birmingham , Birmingham , United Kingdom) This is my paper

Pith reviewed 2026-05-14 22:22 UTC · model grok-4.3

classification 💻 cs.RO cs.AIcs.HCcs.LG
keywords EEG decodingbrain-computer interfacehand kinematicsfinite-state machinemotion predictionrobot controlmultimodal decodingPearson correlation
0
0 comments X

The pith

A finite-state machine critic filters low-confidence EEG hand-motion predictions to reach 0.93 PCC while excluding under 20 percent of points.

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

The paper builds a CNN-attention hybrid model that decodes hand kinematics from EEG during grasp-and-lift tasks and reports strong within-subject Pearson correlations. It then adds a copilot post-processing layer that runs a motion-state-aware critic inside a finite-state machine to remove uncertain decoded points. This step lifts overall within-subject EEG-only PCC to 0.93 while discarding fewer than 20 percent of the data. The filtered trajectories are shown to drive a simulated Franka Panda arm. Readers would care because the method turns noisy brain signals into usable robot commands with limited data loss.

Core claim

A CNN-attention model first predicts hand trajectories from EEG (and from combined EEG-EMG), achieving within-subject PCCs of 0.9854, 0.9946, and 0.9065 on the X, Y, and Z axes of the thumb-index midpoint. These decoded paths are then passed through a copilot framework whose motion-state-aware critic, embedded in a finite-state machine, identifies and removes low-confidence points; the result is an overall within-subject PCC of 0.93 while fewer than 20 percent of points are excluded, yielding trajectories suitable for controlling a Franka Panda arm in MuJoCo simulation.

What carries the argument

The motion-state-aware critic inside a finite-state machine that flags and filters low-confidence points in the decoded kinematic trajectory.

If this is right

  • EEG-only decoding becomes reliable enough for simulated robotic arm control after the filtering step.
  • Multimodal EEG-EMG decoding reaches higher axis-wise PCCs than EEG alone.
  • Cross-subject performance stays lower, especially on the Z axis (0.5852).
  • The method works for both within-subject and cross-subject settings while keeping data exclusion modest.

Where Pith is reading between the lines

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

  • The same critic could be adapted to other motor tasks such as reaching or walking without retraining the entire decoder.
  • Running the finite-state machine in real time on streaming EEG might support continuous rather than offline robot assistance.
  • Combining the filter with existing artifact-rejection pipelines could further reduce the fraction of points excluded.

Load-bearing premise

The critic inside the finite-state machine correctly identifies only low-confidence points without removing valid motion segments or introducing selection bias that inflates the reported correlation.

What would settle it

Applying the same copilot filter to a fresh set of grasp-and-lift EEG recordings and finding that PCC remains below 0.9 or that more than 20 percent of points must be dropped to reach 0.93 would falsify the claimed improvement.

Figures

Figures reproduced from arXiv: 2603.27492 by Birmingham, Jian K. Liu (1) ((1) University of Birmingham, Shixiao Wang (1), United Kingdom), Yizhe Li (1).

Figure 1
Figure 1. Figure 1: Structure of the decoding model. B. Data Preprocessing EEG preprocessing was performed using MNE-Python. Raw data were band-pass filtered between 0.1 and 40 Hz with an infinite impulse response (IIR) filter to remove low￾frequency drifts and high-frequency noise. Common average referencing (CAR) was then applied across all electrodes to reduce spatially distributed artifacts and enhance localized neural ac… view at source ↗
Figure 3
Figure 3. Figure 3: Structure of the copilot and decoding point filtering [PITH_FULL_IMAGE:figures/full_fig_p003_3.png] view at source ↗
Figure 2
Figure 2. Figure 2: Structure of the multi-convolution block (top). Structure [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗
Figure 5
Figure 5. Figure 5: (Left) Model performance with varying input window [PITH_FULL_IMAGE:figures/full_fig_p004_5.png] view at source ↗
Figure 7
Figure 7. Figure 7: Performance of the test dataset for participant 4 after [PITH_FULL_IMAGE:figures/full_fig_p005_7.png] view at source ↗
read the original abstract

Motor kinematics prediction (MKP) from electroencephalography (EEG) is an important research area for developing movement-related brain-computer interfaces (BCIs). While traditional methods often rely on convolutional neural networks (CNNs) or recurrent neural networks (RNNs), Transformer-based models have shown strong ability in modeling long sequential EEG data. In this study, we propose a CNN-attention hybrid model for decoding hand kinematics from EEG during grasp-and-lift tasks, achieving strong performance in within-subject experiments. We further extend this approach to EEG-EMG multimodal decoding, which yields substantially improved results. Within-subject tests achieve PCC values of 0.9854, 0.9946, and 0.9065 for the X, Y, and Z axes, respectively, computed on the midpoint trajectory between the thumb and index finger, while cross-subject tests result in 0.9643, 0.9795, and 0.5852. The decoded trajectories from both modalities are then used to control a Franka Panda robotic arm in a MuJoCo simulation. To enhance trajectory fidelity, we introduce a copilot framework that filters low-confidence decoded points using a motion-state-aware critic within a finite-state machine. This post-processing step improves the overall within-subject PCC of EEG-only decoding to 0.93 while excluding fewer than 20% of the data points.

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 proposes a CNN-attention hybrid model for decoding hand kinematics from EEG during grasp-and-lift tasks, reporting within-subject PCC values of 0.9854/0.9946/0.9065 (X/Y/Z) and cross-subject values of 0.9643/0.9795/0.5852 on the thumb-index midpoint trajectory. It extends the approach to EEG-EMG multimodal decoding and introduces a copilot post-processing framework that uses a motion-state-aware critic inside a finite-state machine to filter low-confidence decoded points, claiming this raises the overall within-subject EEG-only PCC to 0.93 while excluding fewer than 20% of samples. The resulting trajectories are used to control a Franka Panda arm in MuJoCo simulation.

Significance. If the copilot filter demonstrably improves fidelity without selection bias, the work could offer a practical post-processing technique for increasing reliability of EEG-based robotic control in BCIs. The reported within-subject PCC numbers are competitive with existing CNN/RNN/Transformer decoders, and the multimodal extension plus simulation deployment provide a concrete end-to-end pipeline. However, the absence of baseline numbers, critic implementation details, and bias diagnostics in the abstract makes it difficult to judge whether the 0.93 figure represents genuine denoising or post-hoc selection.

major comments (2)
  1. [Abstract] Abstract: The claim that the copilot framework raises within-subject EEG-only PCC to 0.93 while excluding <20% of points is load-bearing for the central contribution, yet no pre-filter PCC, no definition of the motion-state-aware critic (e.g., which features or thresholds it uses), no ablation of critic parameters, and no check that excluded segments are not systematically the high-error grasp phases are supplied. Without these, it is impossible to determine whether the retained-set PCC is an unbiased estimator or inflated by correlation between the critic and decoder error.
  2. [Abstract] Abstract: Specific PCC values (0.9854, 0.9946, 0.9065, etc.) are stated without any description of the CNN-attention architecture, loss function, training protocol, cross-validation scheme, or statistical tests. This prevents verification that the numbers support the performance claims and is especially problematic given that the copilot improvement is presented as the key advance.
minor comments (1)
  1. [Abstract] Abstract: The multimodal EEG-EMG results are described only as 'substantially improved' without quantitative comparison to the EEG-only baseline, making it hard to gauge the incremental benefit of the second modality.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive and detailed feedback. We have revised the manuscript to address the concerns about missing details in the abstract and supporting analyses for the copilot framework. Point-by-point responses follow.

read point-by-point responses

  1. Referee: [Abstract] Abstract: The claim that the copilot framework raises within-subject EEG-only PCC to 0.93 while excluding <20% of points is load-bearing for the central contribution, yet no pre-filter PCC, no definition of the motion-state-aware critic (e.g., which features or thresholds it uses), no ablation of critic parameters, and no check that excluded segments are not systematically the high-error grasp phases are supplied. Without these, it is impossible to determine whether the retained-set PCC is an unbiased estimator or inflated by correlation between the critic and decoder error.

    Authors: We agree that the abstract requires additional context to substantiate the copilot contribution. In the revised version we have updated the abstract to report the pre-copilot overall within-subject PCC for EEG-only decoding. We have also added a concise definition of the motion-state-aware critic (velocity-consistency check within the FSM) and its operating threshold. A new ablation subsection has been inserted in the results, varying the exclusion rate and critic threshold to demonstrate the robustness of the 0.93 figure. Finally, we have included a phase-distribution analysis (with statistical test) confirming that excluded segments are not disproportionately drawn from high-error grasp phases. These additions directly address the risk of selection bias. revision: yes

  2. Referee: [Abstract] Abstract: Specific PCC values (0.9854, 0.9946, 0.9065, etc.) are stated without any description of the CNN-attention architecture, loss function, training protocol, cross-validation scheme, or statistical tests. This prevents verification that the numbers support the performance claims and is especially problematic given that the copilot improvement is presented as the key advance.

    Authors: We accept that the abstract is too terse on methodology. The revised abstract now briefly indicates that a CNN-attention hybrid is employed and that performance is assessed via within-subject cross-validation. The full architecture (convolutional front-end plus attention layers), loss function, optimizer settings, cross-validation procedure, and statistical tests are already detailed in the methods and results sections; we have added a short methods-summary paragraph and a hyperparameter table to make these elements immediately accessible without requiring the reader to search the main text. revision: yes

Circularity Check

0 steps flagged

No circularity: purely empirical reporting of model outputs and post-processing filter

full rationale

The manuscript describes an empirical pipeline: training a CNN-attention hybrid on EEG (and EEG-EMG) data to predict hand kinematics, reporting within-subject PCC values (0.9854/0.9946/0.9065 for X/Y/Z), then applying a finite-state-machine motion-state critic to filter low-confidence points and recompute PCC on the retained set (0.93 after <20% exclusion). No equations, derivations, or first-principles results are presented. The reported numbers are direct statistical outputs of model fitting and selective evaluation on held-out or filtered data; they do not reduce to the inputs by algebraic identity or by renaming a fitted parameter as a prediction. No self-citations are invoked as load-bearing uniqueness theorems, and the critic is described as an external post-processing rule rather than a quantity defined in terms of the decoder's own error. The derivation chain is therefore self-contained empirical measurement, not a closed loop.

Axiom & Free-Parameter Ledger

1 free parameters · 0 axioms · 1 invented entities

The central performance claims rest on neural-network parameters fitted to EEG and EMG data plus an ad-hoc definition of the motion-state critic; no external benchmarks or parameter-free derivations are provided.

free parameters (1)
  • CNN-attention model weights and hyperparameters
    All network parameters are fitted to the EEG/EMG training data to produce the reported PCC values.
invented entities (1)
  • motion-state-aware critic no independent evidence
    purpose: To filter low-confidence decoded trajectory points inside the finite-state machine
    New component introduced by the authors with no independent evidence or external validation supplied in the abstract.

pith-pipeline@v0.9.0 · 5579 in / 1212 out tokens · 43315 ms · 2026-05-14T22:22:56.908361+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

39 extracted references · 39 canonical work pages · 1 internal anchor

  1. [1]

    Wolpaw and E

    J. Wolpaw and E. W. Wolpaw,Brain–Computer Interfaces: Principles and Practice. Oxford University Press, 01 2012

    work page 2012

  2. [2]

    Brodal,The central nervous system : structure and function / Per Brodal., 4th ed., 2010

    P. Brodal,The central nervous system : structure and function / Per Brodal., 4th ed., 2010

    work page 2010

  3. [3]

    Distributed neural dynamics underlie the shift from movement preparation to execution,

    Z. Yin, J. K. Liu, and K. Kornysheva, “Distributed neural dynamics underlie the shift from movement preparation to execution,”bioRxiv, Dec. 2025

    work page 2025

  4. [4]

    Brain-computer interfaces: Definitions and principles,

    J. R. Wolpaw, J. D. R. Mill ´an, and N. F. Ramsey, “Brain-computer interfaces: Definitions and principles,”Handbook of clinical neurology, vol. 168, p. 15—23, 2020

    work page 2020

  5. [5]

    Toward the Next Generation of Retinal Neuroprosthesis: Visual Com- putation with Spikes,

    Z. Yu, J. K. Liu, S. Jia, Y . Zhang, Y . Zheng, Y . Tian, and T. Huang, “Toward the Next Generation of Retinal Neuroprosthesis: Visual Com- putation with Spikes,”Engineering, vol. 6, no. 4, pp. 449–461, Apr. 2020

    work page 2020

  6. [6]

    Eeg dataset and openbmi toolbox for three bci paradigms: an investigation into bci illiteracy,

    M.-H. Lee, O.-Y . Kwon, Y .-J. Kim, H.-K. Kim, Y .-E. Lee, J. Williamson, S. Fazli, and S.-W. Lee, “Eeg dataset and openbmi toolbox for three bci paradigms: an investigation into bci illiteracy,”Gigascience, vol. 8, no. 5, 2019

    work page 2019

  7. [7]

    B. Z. Allison, S. Dunne, R. Leeb, J. D. R. Milln, and A. Nijholt,Towards Practical Brain-Computer Interfaces: Bridging the Gap from Research to Real-World Applications. Springer, 2012

    work page 2012

  8. [8]

    Mind the gap: State-of-the-art technologies and applications for eeg- based brain–computer interfaces,

    R. Portillo-Lara, B. Tahirbegi, C. Chapman, J. Goding, and R. Green, “Mind the gap: State-of-the-art technologies and applications for eeg- based brain–computer interfaces,”APL Bioengineering, vol. 5, p. 031507, 09 2021

    work page 2021

  9. [9]

    Empirical comparison of deep learning methods for eeg decoding,

    I. H. de Oliveira and A. C. Rodrigues, “Empirical comparison of deep learning methods for eeg decoding,”Frontiers in neuroscience, vol. 16, p. 1003984, 2023

    work page 2023

  10. [10]

    Deep learning with convolutional neural networks for eeg decoding and visualization,

    R. T. Schirrmeister, J. T. Springenberg, L. D. J. Fiederer, M. Glasstetter, K. Eggensperger, M. Tangermann, F. Hutter, W. Burgard, and T. Ball, “Deep learning with convolutional neural networks for eeg decoding and visualization,”Human brain mapping, vol. 38, no. 11, pp. 5391–5420, 2017

    work page 2017

  11. [11]

    Eegnet: a compact convolutional neural network for eeg-based brain-computer interfaces,

    V . J. Lawhern, A. J. Solon, N. R. Waytowich, S. M. Gordon, C. P. Hung, and B. J. Lance, “Eegnet: a compact convolutional neural network for eeg-based brain-computer interfaces,”Journal of neural engineering, vol. 15, no. 5, p. 56013, 2018

    work page 2018

  12. [12]

    Eeg-tcnet: An accurate temporal convolutional network for embedded motor-imagery brain-machine interfaces,

    T. M. Ingolfsson, M. Hersche, X. Wang, N. Kobayashi, L. Cavigelli, and L. Benini, “Eeg-tcnet: An accurate temporal convolutional network for embedded motor-imagery brain-machine interfaces,” inConference proceedings - IEEE International Conference on Systems, Man, and Cybernetics, vol. 2020-. IEEE, 2020, pp. 2958–2965

    work page 2020

  13. [13]

    Transformer-based spatial- temporal feature learning for eeg decoding,

    Y . Song, X. Jia, L. Yang, and L. Xie, “Transformer-based spatial- temporal feature learning for eeg decoding,” 2021

    work page 2021

  14. [14]

    Comparison of attention-based deep learning models for eeg classification,

    G. Cisotto, A. Zanga, J. Chlebus, I. Zoppis, S. Manzoni, and U. Markowska-Kaczmar, “Comparison of attention-based deep learning models for eeg classification,” 2020

    work page 2020

  15. [15]

    Eeg conformer: Convolutional transformer for eeg decoding and visualization,

    Y . Song, Q. Zheng, B. Liu, and X. Gao, “Eeg conformer: Convolutional transformer for eeg decoding and visualization,”IEEE Transactions on Neural Systems and Rehabilitation Engineering, vol. PP, pp. 1–1, 12 2022

    work page 2022

  16. [16]

    Smanet: A model combining sincnet, multi- branch spatial-temporal cnn, and attention mechanism for motor imagery bci,

    D. Wang and Q. Wei, “Smanet: A model combining sincnet, multi- branch spatial-temporal cnn, and attention mechanism for motor imagery bci,”IEEE transactions on neural systems and rehabilitation engineer- ing, vol. 33, pp. 1497–1508, 2025

    work page 2025

  17. [17]

    Distance- and speed-informed kinematics decoding improves m/eeg based upper-limb movement decoder accuracy,

    R. J. Kobler, A. I. Sburlea, V . Mondini, M. Hirata, and G. R. M ¨uller- Putz, “Distance- and speed-informed kinematics decoding improves m/eeg based upper-limb movement decoder accuracy,”Journal of neural engineering, vol. 17, no. 5, p. 56027, 2020

    work page 2020

  18. [18]

    Decoding semantic relatedness and prediction from eeg: A classification method comparison,

    T. Trammel, N. Khodayari, S. J. Luck, M. J. Traxler, and T. Y . Swaab, “Decoding semantic relatedness and prediction from eeg: A classification method comparison,”NeuroImage, vol. 277, p. 120268, 2023

    work page 2023

  19. [19]

    Detection and classification of movement-related cortical potentials associated with task force and speed,

    M. Jochumsen, I. K. Niazi, N. Mrachacz-Kersting, D. Farina, and K. Dremstrup, “Detection and classification of movement-related cortical potentials associated with task force and speed,”Journal of neural engineering, vol. 10, no. 5, p. 56015, 2013

    work page 2013

  20. [20]

    Eeg cortical source feature based hand kine- matics decoding using residual cnn-lstm neural network,

    A. Jain and L. Kumar, “Eeg cortical source feature based hand kine- matics decoding using residual cnn-lstm neural network,” in2023 45th Annual International Conference of the IEEE Engineering in Medicine &amp; Biology Society (EMBC). IEEE, Jul. 2023, p. 1–4

    work page 2023

  21. [21]

    Decoding repetitive finger movements with brain activity acquired via non-invasive electroencephalography,

    A. Y . Paek, H. A. Agashe, and J. L. Contreras-Vidal, “Decoding repetitive finger movements with brain activity acquired via non-invasive electroencephalography,”Frontiers in neuroengineering, vol. 7, p. 3, 2014

    work page 2014

  22. [22]

    Measurement of saccadic eye movements by electrooculography for simultaneous eeg recording,

    Y . Jia and C. W. Tyler, “Measurement of saccadic eye movements by electrooculography for simultaneous eeg recording,”Behavior research methods, vol. 51, no. 5, pp. 2139–2151, 2019

    work page 2019

  23. [23]

    Electroencephalogra- phy signal processing: A comprehensive review and analysis of methods and techniques,

    A. Chaddad, Y . Wu, R. Kateb, and A. Bouridane, “Electroencephalogra- phy signal processing: A comprehensive review and analysis of methods and techniques,”Sensors, vol. 23, no. 14, p. 6434, 2023

    work page 2023

  24. [24]

    Emotion recognition with eeg-based brain-computer interfaces: a systematic literature review,

    K. Erat, E. B. S ¸ahin, F. Do˘gan, N. Merdano ˘glu, A. Akcakaya, and P. O. Durdu, “Emotion recognition with eeg-based brain-computer interfaces: a systematic literature review,”Multimedia tools and applications, vol. 83, no. 33, pp. 79 647–79 694, 2024

    work page 2024

  25. [25]

    Reconstructing three-dimensional hand movements from noninvasive electroencephalo- graphic signals,

    T. J. Bradberry, R. J. Gentili, and J. L. Contreras-Vidal, “Reconstructing three-dimensional hand movements from noninvasive electroencephalo- graphic signals,”The Journal of neuroscience, vol. 30, no. 9, pp. 3432– 3437, 2010

    work page 2010

  26. [26]

    Upper limb movements can be decoded from the time-domain of low-frequency eeg,

    P. Ofner, A. Schwarz, J. Pereira, and G. R. M ¨uller-Putz, “Upper limb movements can be decoded from the time-domain of low-frequency eeg,”PloS one, vol. 12, no. 8, p. e0182578, 2017

    work page 2017

  27. [27]

    Bicurnet: Premovement eeg-based neural decoder for biceps curl trajectory estimation,

    M. Saini, A. Jain, S. P. Muthukrishnan, S. Bhasin, S. Roy, and L. Kumar, “Bicurnet: Premovement eeg-based neural decoder for biceps curl trajectory estimation,”IEEE transactions on instrumentation and measurement, vol. 73, pp. 1–11, 2024

    work page 2024

  28. [28]

    Brain-controlled robotic arm system based on multi-directional cnn-bilstm network using eeg signals,

    J.-H. Jeong, K.-H. Shim, D.-J. Kim, and S.-W. Lee, “Brain-controlled robotic arm system based on multi-directional cnn-bilstm network using eeg signals,”IEEE transactions on neural systems and rehabilitation engineering, vol. 28, no. 5, pp. 1226–1238, 2020

    work page 2020

  29. [29]

    Esi-gal: Eeg source imaging-based kinematics parameter estimation for grasp and lift task,

    A. Jain and L. Kumar, “Esi-gal: Eeg source imaging-based kinematics parameter estimation for grasp and lift task,”arXiv, 2024

    work page 2024

  30. [30]

    Eegformer: A transformer–based brain activity classification method using eeg signal,

    Z. Wan, M. Li, S. Liu, J. Huang, H. Tan, and W. Duan, “Eegformer: A transformer–based brain activity classification method using eeg signal,” Frontiers in neuroscience, vol. 17, p. 1148855, 2023

    work page 2023

  31. [31]

    Easm: An efficient attnsleep model for sleep apnea detection from eeg signals,

    M. Singh, S. Chauhan, A. K. Rajput, I. Verma, and A. K. Tiwari, “Easm: An efficient attnsleep model for sleep apnea detection from eeg signals,” Multimedia tools and applications, vol. 84, no. 4, pp. 1985–2003, 2025

    work page 1985

  32. [32]

    Eeg- tcntransformer: A temporal convolutional transformer for motor imagery brain–computer interfaces,

    A. H. P. Nguyen, O. Oyefisayo, M. A. Pfeffer, and S. H. Ling, “Eeg- tcntransformer: A temporal convolutional transformer for motor imagery brain–computer interfaces,”Signals, vol. 5, no. 3, pp. 605–632, 2024

    work page 2024

  33. [33]

    Eeg-convtransformer for single-trial eeg based visual stimuli classification,

    S. Bagchi and D. R. Bathula, “Eeg-convtransformer for single-trial eeg based visual stimuli classification,” 2021

    work page 2021

  34. [34]

    Brain–computer interface control with artificial intelligence copilots,

    J. Y . Lee, S. Lee, A. Mishra, X. Yan, B. Mcmahan, B. Gaisford, C. Kobashigawa, M. Qu, C. Xie, and J. C. Kao, “Brain–computer interface control with artificial intelligence copilots,”Nature Machine Intelligence

    work page

  35. [35]

    Way-eeg-gal: Multi-channel eeg recordings during 3,936 grasp and lift trials with varying weight and friction,

    M. David Luciw, E. Jarocka, and B. Edin, “Way-eeg-gal: Multi-channel eeg recordings during 3,936 grasp and lift trials with varying weight and friction,” Nov 2014

    work page 2014

  36. [36]

    An empirical evaluation of generic convolutional and recurrent networks for sequence modeling,

    S. Bai, J. Z. Kolter, and V . Koltun, “An empirical evaluation of generic convolutional and recurrent networks for sequence modeling,” 2018

    work page 2018

  37. [37]

    Squeeze-and-excitation networks,

    J. Hu, L. Shen, S. Albanie, G. Sun, and E. Wu, “Squeeze-and-excitation networks,”IEEE transactions on pattern analysis and machine intelli- gence, vol. 42, no. 8, pp. 2011–2023, 2020

    work page 2011

  38. [38]

    Attention Is All You Need

    A. Vaswani, N. Shazeer, N. Parmar, J. Uszkoreit, L. Jones, A. N. Gomez, L. Kaiser, and I. Polosukhin, “Attention is all you need,”arXiv, vol. abs/1706.03762, 2017

    work page internal anchor Pith review Pith/arXiv arXiv 2017

  39. [39]

    Spiking Variational Policy Gradient for Brain Inspired Reinforcement Learning,

    Z. Yang, S. Guo, Y . Fang, Z. Yu, and J. K. Liu, “Spiking Variational Policy Gradient for Brain Inspired Reinforcement Learning,”IEEE Transactions on Pattern Analysis and Machine Intelligence, vol. 47, no. 3, pp. 1975–1990, Mar. 2025

    work page 1975