Pith. sign in

REVIEW 2 major objections 6 minor 57 references

Temporal Basis Function Models for Closed-Loop Neural Stimulation

T0 review · 2 major / 6 minor · reviewed 2026-08-06 · deepseek-v4-flash

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

Pith's one-line read A temporal basis function model predicts single-trial, spatiotemporal responses to optogenetic stimulation from a 20 ms runway of multichannel LFP, with held-out $R^2$ of 0.462 over a 164 ms horizon; the paper argues this makes…

desk verdict Solid forward model with real translational metrics, but the closed-loop simulation claim conflates forecasting with control because no no-stimulation counterfactual is ever generated. read the letter →

arxiv 2507.15274 v1 pith:BT4LBE53 submitted 2025-07-21 cs.LG

Matthew J. Bryan , Felix Schwock , Azadeh Yazdan-Shahmorad , Rajesh P N Rao This is my paper · ORCID
classification cs.LG
keywords temporalbasisfunctionmodelclosed-loopneuralstimulationoptogeneticslocalfieldpotentialsstate-dependentresponsenon-humanprimatebrainco-processorsample-efficientmachinelearning
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

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

The reading

This paper argues that a simple temporal basis function model can learn, from about 5,000 trials collected early in a session, a single-trial forward prediction of how paired-pulse optogenetic stimulation will reshape local field potentials across an electrode array. On held-out trials the model achieves a mean test-set $R^2$ of 0.462 over a 164 ms forecast horizon, and it beats a more complex recurrent model on most sessions and a linear state-space model on every session while training in minutes rather than hours. The point is translational: if a forward model is this cheap to build and this fast to evaluate, closed-loop stimulation can be shaped to the patient's current brain state within a realistic loop latency. The authors also present statistical evidence that the stimulation response depends on the pre-stimulus brain state, and they show that removing that state dependence from the model collapses its accuracy.

What carries the argument

The central object is the temporal basis function model (TBFM), defined by $\hat{y}_c = x_{c,r}\mathbf{1} + \sum_{i=1}^b W(X)_{c,i} B_{i,*}$: the forecast for each channel is the last measured LFP value plus a weighted sum of $b$ temporal basis functions. The basis functions are produced by a small multilayer perceptron from a stimulation descriptor (here, the timing of the two light pulses), and the weight matrix $W$ is an affine function of the Z-scored "runway"—the final 20 ms of multichannel LFP before the forecast begins. This decomposition carries the argument because the nonlinear dependence on stimulation parameters lives in the basis generator, while the trial-specific dependence on brain state lives in the linear weight estimator, and the whole horizon is produced in one forward pass instead of being unrolled through time. After training, the nonlinear generator can be compiled away for a discrete set of stimulation parameters, leaving a linear map that runs in about 0.115 ms on a desktop CPU.

What would settle it

Run the trained TBFM in a real-time closed-loop experiment on the same non-human primate preparation, with the controller instructed not to stimulate on a random subset of trials and the true unstimulated LFP recorded under each initial state; if the decision performance computed against those genuine no-stimulation outcomes does not track the reported simulated AUCs, the closed-loop claim is not supported.

Watch

Extended reading notes

Core claim

The paper's central claim is that the forward model for optogenetic stimulation does not need to be a deep recurrent network: a weighted sum of a small set of learned temporal basis functions, with weights read off affinely from the recent multichannel LFP, is enough to predict the spatiotemporal response on individual trials. The authors report a mean test-set $R^2$ of 0.462 (stdev 0.207) over the full 164 ms forecast, 0.787 over a 40 ms forecast, and a state-dependent $R^2$ of 0.878 when predictions are binned by initial state. When the model is given a sham runway, test $R^2$ falls to 0.006 (stdev 0.012), which the authors take as evidence that the model is conditioning on the pre-stimulus brain state rather than emitting a mean evoked response. They interpret the overall result as showing that the practical barriers to model-based closed-loop stimulation—sample efficiency, training time, and loop latency—can be addressed without giving up accuracy.

Load-bearing premise

The load-bearing assumption is that a trial recorded with stimulation always delivered is the correct ground-truth outcome even when the controller decides not to stimulate; if true no-stimulation trajectories differ, the simulated closed-loop results do not establish that the controller can shape brain activity.

Editorial extensions

If this is right

  • A session can be modeled from about 5,000 early trials (under 20 minutes of collection), trained in 2 to 5 minutes, and then used for stimulation decisions on the rest of the session.
  • A clinically deployed stimulation model should feed the controller the ongoing LFP rather than only the stimulation schedule, because state-agnostic models in this dataset achieve near-zero $R^2$.
  • The measured forecast latency of about 0.2 ms on a desktop CPU leaves ample room within a 20 ms control loop for sensing, filtering, and stimulation hardware.
  • If the comparison with the autoencoder-LSTM baseline generalizes, complex recurrent dynamics are not necessary for this forecasting task, and simpler, easier-to-verify models can be used in safety-critical settings.
  • Even a model with test $R^2$ around 0.46 can, in the paper's simulations, time stimulation to predicted target states and toward reference trajectories, suggesting that useful closed-loop control may not require near-perfect prediction.

Reading between the lines

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

  • One extension the paper leaves implicit is that the same architecture should generalize to continuously varying pulse parameters such as width, amplitude, or site, since the stimulation descriptor can be extended; the paper demonstrates only timing variation, so this remains untested.
  • Because the basis generator can be compiled into a fixed linear map once the stimulation set is discrete, the paper's latency figures suggest a version of the model could run on low-power embedded hardware for an implantable controller, though the authors do not test this.
  • The decisive test of the closed-loop claim would be a real-time experiment with genuine no-stimulation trials, comparing trajectories after a no-stimulate decision with unstimulated baselines matched on initial state; the reported AUCs are computed against recorded trials in which stimulation was always delivered.
Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

2 major / 6 minor

Summary. This paper proposes temporal basis function models (TBFMs) for single-trial, spatiotemporal forward prediction of local field potential (LFP) responses to optogenetic stimulation, targeting sample efficiency, short training time, and low inference latency for closed-loop neural stimulation. The authors evaluate the model on 40 sessions of paired-pulse optogenetic stimulation data from two non-human primates, reporting a mean test-set R2 of 0.462 over a 164 ms horizon on time-domain data, with improvements over a linear state-space model and accuracy comparable to or better than a nonlinear AE-LSTM baseline. They also report statistical evidence for state-dependence of the stimulation response, benchmarks of training time and latency, and two simulated closed-loop controllers that use TBFM forecasts to decide whether to stimulate, with reported AUC values of roughly 0.65–0.72.

Significance. If the forward-model claims hold, the contribution is practically important: the model appears to offer a favorable trade-off among prediction accuracy, sample efficiency (about 5k trials), training time (2–5 minutes), and inference latency (about 0.2 ms on a CPU), directly addressing translational barriers for model-based neural co-processors. Strengths of the evaluation include a time-based train/test split that tests generalization across each session, a state-agnostic ablation showing the value of conditioning on the initial state, and a cross-validation control for non-stationarity and plasticity. The authors also release a PyTorch implementation and demo code. However, the closed-loop simulation results do not validly demonstrate closed-loop control, as detailed in the major comments; this limits the scope of the central claim but does not undermine the forward-modeling contribution itself.

major comments (2)
  1. [§4.9.1, §6.3.1] The evaluation of Demonstration 1 conflates prediction with control. Because stimulation was always applied in the stimulation blocks (§3.1), no true no-stimulation trials exist in those blocks; nevertheless, Appendix 6.3.1 labels test-set trials as 'do not stimulate' trials and uses the recorded (stimulated) trajectory as the ground-truth outcome for both the stimulate and do-not-stimulate decisions. Consequently, the ROC analysis in §4.9.1 measures whether the model can forecast which already-stimulated recorded trials will cross the target range, not whether the controller's decision changes the outcome. The statement in §4.9.1 that 'These results indicate that the TBFM achieved overall controllability' does not follow from the reported analysis.
  2. [§4.9.2, §6.3.2] Demonstration 2 suffers from the same counterfactual missingness. The reference trajectories are weighted averages of stimulated test trials, and the cost function compares the TBFM forecast (which assumes stimulation) to the reference; the 'actual trajectory assuming stimulation' is the recorded stimulated trial, and no no-stimulation trajectory is ever simulated. The reported AUC values (0.721 raw time domain, 0.637 beta) therefore quantify the model's forecasting skill under a fixed always-stimulate regime, not its ability to control neural activity. The abstract and §3.6 claim that the model 'successfully allowed a closed-loop stimulator to control a neural circuit' is unsupported by the present evidence and should be withdrawn or substantially qualified.
minor comments (6)
  1. [§4.2 vs. Fig. 14 caption] The text in §4.2 says the TBFM exceeded the AE-LSTM on 'all but 6 of the 40 sessions' (i.e., 34/40), while the Figure 14 caption says 'on 32 of 40 sessions'; these numbers should be reconciled.
  2. [§4.9.2] The sentence reporting that test-set R2 was 'predictive of the AUC' gives only the p-value (p=1.01e−3); the correlation coefficient and the number of sessions should be reported.
  3. [§6.1] The phrase 'was loss-pass filtered' should read 'was low-pass filtered'.
  4. [Table 1] The column heading 'R2 mean-vs-mean' is not defined in the caption; please define it in the caption or in §4.2.
  5. [§3.5/§4.8] The FSAM stopping rule is described as adding bases until validation-set accuracy no longer increases, but no quantitative convergence threshold is given; please specify the criterion used in Figure 10.
  6. [§5.3] The code availability statement says the implementation is available under an open-source license but does not name the license; please specify it.

Circularity Check

2 steps flagged · score 6.0 of 10

The closed-loop simulations evaluate both 'stimulate' and 'do not stimulate' decisions against the recorded stimulated trajectory, so the reported AUC reduces to forecasting the training distribution rather than controlling a counterfactual no-stimulation plant.

  1. self definitional [Section 4.9.1 and Appendix 6.3.1]
    "For the rest of the trials ('do not stimulate' trials), the target range is some other randomly chosen target. ... False positive here refers to the controller applying stimulation but the trajectory does not in-fact cross into the target zone, and false negative refers to cases where the controller does not apply stimulation but the trajectory does cross into the target zone."

    Every trial in the dataset is a stimulated trial, yet the 'do not stimulate' trials are still scored against the recorded stimulated trajectory. The no-stimulation outcome is never generated or modeled, so a controller that chooses not to stimulate is evaluated as though the stimulation response still occurred. The false negative condition, 'no stimulation but the trajectory crosses the target', cannot occur in the actual counterfactual because without stimulation the trajectory would be different. Consequently, the ROC/AUC measures whether the TBFM can forecast which recorded stimulated trajectories will cross a target, not whether deciding not to stimulate avoids the target.

  2. self definitional [Section 4.9.2, Appendix 6.3.2, and Figure 4(b) caption]
    "For this demonstration we generate reference trajectories by calculating a weighted average of test set trajectories, beginning at the stimulation onset t=40 ms. ... The simulation evaluates the controller's performance by measuring the L2 distance between the actual trajectory assuming stimulation (blue) and the reference, and comparing that distance to the same threshold ϵs."

    The reference trajectories are constructed from stimulated test-set trajectories, and the 'actual trajectory' used for evaluation is always the stimulated trajectory, regardless of whether the controller chooses to stimulate or not. The controller's only decision is whether to apply stimulation, but the no-stimulation branch is never simulated. Therefore both the controller's cost function and the evaluation compare stimulated trajectories to references derived from stimulated trajectories. The reported AUC quantifies how close the TBFM's forecasts are to the stimulated data distribution, not whether the controller's decision changes the neural trajectory. This makes the second closed-loop demonstration's control claim reduce to the model's forecasting performance by construction.

full rationale

The core forward-prediction results are not circular: the TBFM is trained on early-session trials and evaluated on held-out late-session trials, it is compared against independently trained AE-LSTM and LSSM baselines, and the state-dependence ablation against a state-agnostic model is a legitimate experimental control. There is no load-bearing self-citation chain and no imported uniqueness theorem; the cited prior work supplies data and context, not the derivation. The circularity is concentrated in the simulated closed-loop demonstrations of Section 4.9. In Demonstration 1, trials labeled 'do not stimulate' are still recorded stimulated trials, and both true/false labels are defined using the stimulated trajectory, so the no-stimulation counterfactual is absent. In Demonstration 2, reference trajectories are weighted averages of stimulated test trajectories and the 'actual trajectory assuming stimulation' is used as the outcome for both controller actions. In both cases, the reported AUC therefore measures how well the model forecasts the stimulated data it was trained on, rather than whether the controller's decision to withhold stimulation would produce a different outcome. The abstract's claim that the model 'successfully allowed a closed-loop stimulator to control a neural circuit' is thus supported only in the sense that forecast accuracy transfers to the ROC; the control demonstration is partially circular by construction.

Assumptions & free parameters 8 free parameters · 5 assumptions · 0 invented entities

The central forward-model result is an empirical fit, so the main free parameters are hyperparameters and simulation thresholds. The axioms are domain assumptions about the data and the simulation. No new physical entities are introduced. The most consequential entry is the simulation ground-truth axiom, which is not valid for closed-loop evaluation.

free parameters (8)
  • Number of basis functions b = 12-15 (chosen by validation)
    Hyperparameter selected per session or globally; affects model capacity (Section 3.3).
  • Regularization coefficient lambda = 0.05
    Set via small hyperparameter search for all sessions (Appendix 6.2.3).
  • Learning rate = 2e-4
    AdamW optimizer setting used for all sessions (Appendix 6.2.3).
  • MLP hidden layers and width = 4 layers, width 4
    Architecture of the basis generator, chosen as a small default (Appendix 6.2.2).
  • Runway length = 20 ms
    Fixed pre-stimulus window used as input; assumes this window captures relevant state (Section 3.2).
  • Prediction horizon = 164 ms (default)
    Forecast length; shorter horizons give higher R2 (Section 3.3).
  • Simulation thresholds delta and epsilon = Varied over ranges to produce ROC curves
    Controller threshold parameters for the two simulated demonstrations (Sections 3.6.1, 3.6.2).
  • Training set size = 5k trials
    Chosen as the smallest training size within 1% of maximum test R2 (Section 4.5).
assumptions (5)
  • domain assumption The micro-ECoG LFP recordings reflect the neural state relevant to stimulation responses
    Widely used in the field but not verified in this paper; the entire prediction relies on it (Section 3.1).
  • domain assumption The 20 ms runway contains sufficient information to predict the state-dependent response
    The model architecture uses only this window; if longer history matters, the model cannot capture it (Section 3.2).
  • domain assumption Stimulation response statistics are stable enough across a session for early-training to generalize to late-testing
    The authors test this with cross-validation and find no significant difference on time-domain data (p=0.57), but the test may be underpowered (Appendix 6.9).
  • domain assumption Each stimulation trial can be paired with a baseline trial with similar initial state to isolate the stimulation response
    Used in the state-dependence hypothesis test; if matching is imperfect, the estimated response contains baseline mismatch (Appendix 6.5).
  • ad hoc to paper In the closed-loop simulations, recorded trajectories serve as ground truth for both stimulate and do-not-stimulate decisions
    This is the key flawed assumption: the recorded trajectory always includes stimulation, so it is not a valid outcome for a no-stimulation controller decision (Section 4.9, Appendix 6.3.1).

how reviews work

0 comments
Cite this review

Pith. "Pith review of Temporal Basis Function Models for Closed-Loop Neural Stimulation." pith.science (2026). https://pith.science/paper/BT4LBE53

@misc{pith2026250715274,
  author       = {Pith},
  title        = {Pith review of: Temporal Basis Function Models for Closed-Loop Neural Stimulation},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/BT4LBE53}},
  note         = {Machine review of arXiv:2507.15274}
}
read the original abstract

Closed-loop neural stimulation provides novel therapies for neurological diseases such as Parkinson's disease (PD), but it is not yet clear whether artificial intelligence (AI) techniques can tailor closed-loop stimulation to individual patients or identify new therapies. Progress requires us to address a number of translational issues, including sample efficiency, training time, and minimizing loop latency such that stimulation may be shaped in response to changing brain activity. We propose temporal basis function models (TBFMs) to address these difficulties, and explore this approach in the context of excitatory optogenetic stimulation. We demonstrate the ability of TBF models to provide a single-trial, spatiotemporal forward prediction of the effect of optogenetic stimulation on local field potentials (LFPs) measured in two non-human primates. We further use simulations to demonstrate the use of TBF models for closed-loop stimulation, driving neural activity towards target patterns. The simplicity of TBF models allow them to be sample efficient, rapid to train (2-4min), and low latency (0.2ms) on desktop CPUs. We demonstrate the model on 40 sessions of previously published excitatory optogenetic stimulation data. For each session, the model required 15-20min of data collection to successfully model the remainder of the session. It achieved a prediction accuracy comparable to a baseline nonlinear dynamical systems model that requires hours to train, and superior accuracy to a linear state-space model. In our simulations, it also successfully allowed a closed-loop stimulator to control a neural circuit. Our approach begins to bridge the translational gap between complex AI-based approaches to modeling dynamical systems and the vision of using such forward prediction models to develop novel, clinically useful closed-loop stimulation protocols.

Figures

Figures reproduced from arXiv: 2507.15274 by the authors.

Figure 1
Figure 1. Spatiotemporal electrocorticographic (ECoG) responses after paired pulse optical stimulation. (a) Dynamics of ECoG responses after stimulation. The three panels show spatially-smoothed time domain ECoG responses across the surface of monkey sensorimotor cortex (S1, M1), measured using an array of µECoG electrodes (circles). The precise placement of the array on the cortex varied between sessions. The ECoG responses … view at source ↗
Figure 2
Figure 2. Architecture of the temporal basis function model. The model predicts the response yc of a channel c to stimulation using a weighted sum of temporal basis functions and the last measurement xc,r made before prediction. r here is the runway length. A multilayer perceptron (MLP) generates b temporal bases from a descriptor of the stimulation parameters (known as the “stimulation descriptor”). The bases have length equ… view at source ↗
Figure 3
Figure 3. Training procedure. Training set trials are sourced from early in the session, to simulate experimental conditions. The runways are passed to the Temporal Basis Function Model (TBFM) along with the stimulation descriptors. The TBFM predicts the local field potentials (LFPs) from the latter part of the trial. Note we depict individual channels here, but the TBFM is jointly estimating all channels for a given trial; i… view at source ↗
Figures from the paper (23 more)
Figure 4
Figure 4. Figure 4: Demonstration of two closed-loop stimulators. (a) Stimulation applied to target future brain states. In this task, the controller must decide at t=20ms to apply stimulation in the future at t=40ms, but only if a target brain state is forecast to occur. The controller l…
Figure 5
Figure 5. Figure 5: Example state dependence of stimulation response. (a) Trial-averaged resting and stimulation data for N=4 binned initial states, example session and channel. Vertical dashed line at t=40ms denotes binning time. Resting data is trial-matched to stimulation trials as in …
Figure 6
Figure 6. Figure 6: Performance of the temporal basis function model (TBFM) for time domain data. (a) Training versus test R2 for time domain data. The plots show R2 computed for 3 runs per session over 40 sessions as a function of prediction horizon. Mean R2 for a horizon of 164ms is 0.5…
Figure 7
Figure 7. Figure 7: Performance of the temporal basis function model (TBFM) for beta￾bandpassed data. (a) Training versus test R2 for beta bandpassed data across all sessions. Training set R2 : 0.211 (stdev 0.051), test set: 0.114, (stdev 0.087). (b) Example of mean forecast trajectory fo…
Figure 8
Figure 8. Figure 8: Mean stimulation response as a function of initial state for time-domain data. Mean stimulation response versus prediction across binned initial states (t=20ms) (a) training set, example session and channel. State-dependent R2 : 0.94 (b) test set; state￾dependent R2 : …
Figure 9
Figure 9. Figure 9: Sample efficiencies for an example session. We vary the training set size and measure the R2 of the temporal basis function (TBF) model on both the training set and test set. The training set is sourced from the first N trials of the session. The test set is sourced fr…
Figure 10
Figure 10. Figure 10: Forward Stagewise Additive Modeling (FSAM) results on an example session. As with principal component analysis, FSAM tends to reduce L2 loss (a) and increase R2 (b) as additional basis functions are added, though with diminishing returns. The dashed plots show results…
Figure 11
Figure 11. Figure 11: Residuals in Forward Stagewise Additive Modeling (FSAM). (a) Mean responses computed from three sets of initial states: low (blue), average (orange), and high (green). (b) Mean prediction yˆ for each of the states made with a single basis function. All predictions are…
Figure 12
Figure 12. Figure 12: Results of simulated closed-loop brain stimulation (a) Demo 1, unfiltered time domain, mean AUC 0.704 (stdev 0.090); (b) Demo 1, β bandpassed, mean AUC 0.652 (stdev 0.045); (c) Demo 2, unfiltered time domain, mean AUC 0.694 (stdev 0.086); (d) Demo 2, β-bandpassed time…
Figure 13
Figure 13. Figure 13: (a) LSTM-based Dynamical Systems model (AE-LSTM) Here the + superscript refers to a single step prediction. Details in text body [PITH_FULL_IMAGE:figures/full_fig_p034_13.png]
Figure 14
Figure 14. Figure 14: Comparison of R2 on test set, reference models versus temporal basis function models (TBFMs) The TBFM model’s R2 exceeds the autoencoder with LSTM dynamics (AE-LSTM) model on 32 of 40 sessions. It exceeds the linear state space model (LSSM) on 39 of 40. While this sol…
Figure 15
Figure 15. Figure 15: Distributions of stimulation and resting trials, single session and channel (a) Trial-averaged resting (dashed lines) and stimulation (solid lines) trials. Each are binned into quartiles for graphing. Note that we do not perform binning for our statistical analysis. (…
Figure 16
Figure 16. Figure 16: Mutual information-based state dependence measure versus Hilbert￾Schmidt Independence Critereon (HSIC) state dependence measure The two measures are highly correlated: 0.96 (P ≈ 0.0), and thus appear to capture, largely, the same thing. The two methods give largely th…
Figure 17
Figure 17. Figure 17: R2 versus prediction horizon, mean of 40 sessions See [PITH_FULL_IMAGE:figures/full_fig_p038_17.png]
Figure 18
Figure 18. Figure 18: Example state-dependent predictions for bandpassed data Initial state binned at t=20ms into five states and averaged within each bin; example single channel and single session. Top row: beta bandpassed (a) Training set; (b) test set Bottom row: high gamma bandpassed (…
Figure 19
Figure 19. Figure 19: Example state-dependent predictions, time-frequency domain data Initial state binned at the 4th time step (approx t=20ms) into five states and averaged within each bin; example single channel and single session. Top row: beta band, time-frequency domain (a) Training s…
Figure 20
Figure 20. Figure 20: 1/f frequency power scaling law leads to higher model performance on lower frequencies. (a) Power spectral density (PSD) of resting state data, session, channel, and trial averaged, log-log scale. Linear trend indicates the characteristic 1/f power scaling law. Notche…
Figure 6
Figure 6. Figure 6: provides mean predictions for train and test sets across the three sessions. [PITH_FULL_IMAGE:figures/full_fig_p042_6.png]
Figure 21
Figure 21. Figure 21: Mean predicted versus actual, single temporal basis function model (TBFM) with multiple stimulation parameters Example sessions and channels, trial averaged [PITH_FULL_IMAGE:figures/full_fig_p042_21.png]
Figure 22
Figure 22. Figure 22: Cross validation results on time series data (a) R2 train versus test sets Results for 3 runs per session, 40 sessions. Mean R2 on train set 0.533 (±0.173). Mean R2 on test set 0.462 (±0.207). (b) R2 60ms forward prediction Results rise marginally if we forward predic…
Figure 23
Figure 23. Figure 23: Learned basis functions from an example model [PITH_FULL_IMAGE:figures/full_fig_p046_23.png]
Figure 24
Figure 24. Figure 24: Learned basis functions from an example additive model [PITH_FULL_IMAGE:figures/full_fig_p047_24.png]
Figure 25
Figure 25. Figure 25: Example trials from Demonstrations 1, 2 (a) Demo 1, true positive. (b) Demo 1, true negative. (c) Demo 2, true positive. (d) Demo 2, true negative [PITH_FULL_IMAGE:figures/full_fig_p048_25.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

57 extracted references · 53 canonical work pages

  1. [1]

    Niparko J K 2009Lippincott Williams and Wilkins(Oxford University Press)

    work page

  2. [2]

    Weiland J D and Humayun M S 2014IEEE transactions on bio-medical engineering61(5) 1412–1424 Retinal prosthesis

    work page

  3. [3]

    Tomlinson T and Miller L E 2016Advances in experimental medicine and biology957367–388 Toward a proprioceptive neural interface that mimics natural cortical activity

    work page

  4. [4]

    Tabot G A, Dammann J F, Berg J A, Tenore F V, Boback J L, Vogelstein R J and Bensmaia S J 2013Proceedings of the National Academy of Sciences11018279–18284 Restoring the sense of touch with a prosthetic hand through a brain interface ISSN 0027-8424 URL https://www.pnas.org/content/110/45/18279

    work page

  5. [5]

    Tyler D J 2015Current opinion in neurology28(6) 574–581 Neural interfaces for somatosensory feedback: bringing life to a prosthesis

    work page

  6. [6]

    Dadarlat M C, O’Doherty J E and Sabes P N 2015Nature neuroscience18(1) 138–144 A learning- based approach to artificial sensory feedback leads to optimal integration Temporal Basis Function Models26

    work page

  7. [7]

    Flesher S N, Collinger J L, Foldes S T, Weiss J M, Downey J E, Tyler-Kabara E C, Bensmaia S J, Schwartz A B, Boninger M L and Gaunt R A 2016Science Translational Medicine 8361ra141–361ra141 Intracortical microstimulation of human somatosensory cortex URL https://www.science.org/doi/abs/10.1126/scitranslmed.aaf8083

    work page doi:10.1126/scitranslmed.aaf8083

  8. [8]

    Cronin J A, Wu J, Collins K L, Sarma D, Rao R P N, Ojemann J G and Olson J D 2016IEEE transactions on haptics9(4) 515–522 Task-specific somatosensory feedback via cortical stimulation in humans

    work page

Show all 57 references
  1. [9]

    O’Doherty J, Lebedev M, Ifft P, Zhuang K, Shokur S, Bleuler H and Nicolelis M 2011Nature 479(7372) 228–231 Active tactile exploration using a brain–machine–brain interface URL https://doi.org/10.1038/nature10489

    doi

  2. [10]

    Bryan M J, Jiang L P and Rao R P N 2023Journal of Neural Engineering20036004 Neural co-processors for restoring brain function: results from a cortical model of grasping URL https://dx.doi.org/10.1088/1741-2552/accaa9

    doi

  3. [11]

    Bolus M, Willats A, Rozell C and Stanley G 2021Journal of neural engineering18(3) State-space optimal feedback control of optogenetically driven neural activity

  4. [12]

    Kahana M J e a 2021medRxivBiomarker-guided neuromodulation aids memory in traumatic brain injury URLhttps://www.medrxiv.org/content/early/2021/05/22/2021.05.18.21256980

    2021

  5. [13]

    Berger T W, Song D, Chan R H M, Marmarelis V Z, LaCoss J, Wills J, Hampson R E, Deadwyler S A and Granacki J J 2012IEEE transactions on neural systems and rehabilitation engineering : a publication of the IEEE Engineering in Medicine and Biology Society20(2) 198–211 A hippocam...

  6. [14]

    Tafazoli S, MacDowell C, Che Z, Letai K, Steinhardt C and Buschman T 2020Journal of Neural Engineering17056007 Learning to control the brain through adaptive closed-loop patterned stimulation URLhttps://doi.org/10.1088/1741-2552/abb860

    doi

  7. [15]

    Casta˜ no-Candamil S, Ferleger B I, Haddock A, Cooper S S, Herron J, Ko A, Chizeck H J and Tangermann M 2020Frontiers in Human Neuroscience14421 A pilot study on data- driven adaptive deep brain stimulation in chronically implanted essential tremor patients URL https://www.fro...

  8. [16]

    Little S e a 2016Journal of neurology, neurosurgery, and psychiatry87(7) 717–21 Bilateral adaptive deep brain stimulation is effective in parkinson’s disease

  9. [17]

    Bradley C, Nydam A S, Dux P E and Mattingley J B 2022Nature Reviews Neuroscience23(8) 459–475 State-dependent effects of neural stimulation on brain function and cognition

  10. [18]

    Zanos S 2019Cold Spring Harb Perspect Med.9(11) Closed-loop neuromodulation in physiological and translational research

  11. [19]

    Kabir A, Dhami P, Dussault Gomez M A, Blumberger D M, Daskalakis Z J, Moreno S and Farzan F 2024Journal of Neuroscience44Influence of large-scale brain state dynamics on the evoked response to brain stimulation ISSN 0270-6474 URL https://www.jneurosci.org/content/44/ 39/e0782242024

  12. [20]

    IEEE Engineering in Medicine and Biology Society

    Bloch J, Khateeb K, Silversmith D, O’Doherty J, Sabes P and Yazdan-Shahmorad A 2019Conference proceedings: Annual International Conference of the IEEE Engineering in Medicine and Biology Society. IEEE Engineering in Medicine and Biology Society. Conference20196446–6449 Cortica...

  13. [21]

    Zanos S, Rembado I, Chen D and Fetz E E 2018Current BiologyPhase-locked stimulation during cortical beta oscillations produces bidirectional synaptic plasticity in awake monkeys

  14. [22]

    Rao R P N 2020 Brain co-processors: Using AI to restore and augment brain functionHandbook of Neuroengineeringed Thakor N (Springer)

    2020

  15. [23]

    Rao R P N 2019Current Opinion in Neurobiology55142–151 Towards neural co-processors for the brain: Combining decoding and encoding in brain-computer interfaces

  16. [24]

    Gallego J A, Perich M G, Miller L E and Solla S A 2017Neuron94978–984 Neural manifolds for the control of movement Temporal Basis Function Models27

  17. [25]

    Furht B, Grostick D, Gluch D, Rabbat G, Parker J and McRoberts M 1991Introduction to Real-Time Computing(Boston, MA: Springer US) pp 1–35 ISBN 978-1-4615-3978-0 URL https://doi.org/10.1007/978-1-4615-3978-0_1

    doi

  18. [26]

    Huang J Q and Lewis F 2003IEEE Transactions on Neural Networks14377–389 Neural-network predictive control for nonlinear dynamic systems with time-delay

  19. [27]

    Yazdan-Shahmorad A, Silversmith D B, Kharazia V and Sabes P N 2018eLife7e31034 Targeted cortical reorganization using optogenetics in non-human primates ISSN 2050-084X URL https://doi.org/10.7554/eLife.31034

    doi

  20. [28]

    URL http://dblp.uni-trier.de/db/journals/corr/corr2406.html#abs-2406-06714

    Pan M, Schrum M, Myers V, Biyik E and Dragan A D 2024CoRRabs/2406.06714Coprocessor actor critic: A model-based reinforcement learning approach for adaptive brain stimulation. URL http://dblp.uni-trier.de/db/journals/corr/corr2406.html#abs-2406-06714

    arXiv

  21. [29]

    Chen X and Tomizuka M 2023Introduction to Modern Controls: with Illustrations in MATLAB and Python(Ind.)

  22. [30]

    Sutton R S and Barto A G 2019Reinforcement learning: An introduction, 2nd ed.(The MIT Press)

  23. [31]

    Moerland T M, Broekens J and Jonker C M 2020CoRRabs/2006.16712Model-based reinforcement learning: A survey (Preprint 2006.16712) URL https://arxiv.org/abs/2006. 16712

    2006 arXiv

  24. [32]

    Thornton C H F and M K 2019Wellcome Open ResearchThe virtual electrode recording tool for extracellular potentials (vertex) version 2.0: Modelling in vitro electrical stimulation of brain tissue URLhttps://doi.org/10.12688/wellcomeopenres.15058.1

    doi

  25. [33]

    Yang Y, Qiao S, Sani O, Sedillo J, Ferrentino B, Pesaran B and Shanechi M 2021Nature Biomedical Engineering5(4) 324–345 Modelling and prediction of the dynamic responses of large-scale brain networks during direct electrical stimulation URL https://doi.org/10.1038/ s41551-020-00666-w

  26. [34]

    Yazdan-Shahmorad A, Diaz-Botia C, Hanson T L, Kharazia V, Ledochowitsch P, Maharbiz M M and Sabes P N 2016Neuron89(5) 927–39 A large-scale interface for optogenetic stimulation and recording in nonhuman primates

  27. [35]

    Ledochowitsch P and Yazdan-Shahmorad e a 2015Journal of neuroscience methods256220–31 Strategies for optical control and simultaneous electrical readout of extended cortical circuits

  28. [36]

    IEEE Engineering in Medicine and Biology Society

    Yazdan-Shahmorad A, Silversmith D B and Sabes P N 2018Annual International Conference of the IEEE Engineering in Medicine and Biology Society. IEEE Engineering in Medicine and Biology Society. Annual International Conference20185479–5482 Novel techniques for large-scale manipu...

  29. [37]

    Yazdan-Shahmorad A, Diaz-Botia C, Hanson T, Ledochowitsch P, Maharabiz M and Sabes P 2015 Progress in Biomedical Optics and Imaging - Proceedings of SPIE9305Demonstration of a setup for chronic optogenetic stimulation and recording across cortical areas in non-human primates

    2015

  30. [38]

    Bloch J, Greaves-Tunnell A, Shea-Brown E, Harchaoui Z, Shojaie A and Yazdan-Shahmorad A 2022iScience25104285 Network structure mediates functional reorganization induced by optogenetic stimulation of non-human primate sensorimotor cortex ISSN 2589-0042 URL https://www.scienced...

  31. [39]

    Flint R D, Ethier C, Oby E R, Miller L E and Slutzky M W 2012Journal of neurophysiology108 18–24 Local field potentials allow accurate decoding of muscle activity

  32. [40]

    Busch N A, Dubois J and VanRullen R 2009Journal of Neuroscience297869–7876 The phase of ongoing eeg oscillations predicts visual perception ISSN 0270-6474 (Preprint https: //www.jneurosci.org/content/29/24/7869.full.pdf) URL https://www.jneurosci.org/ content/29/24/7869

  33. [41]

    B´ edard C, Kr¨ oger H and Destexhe A 2006Physical review letters97(11) Does the 1/f frequency scaling of brain signals reflect self-organized critical states?

  34. [42]

    Papadopoulos S, Darmet L, Szul M J, Congedo M, Bonaiuto J J and Mattout J 2024Imaging Neuroscience21–15 Surfing beta burst waveforms to improve motor imagery-based bci

  35. [43]

    1316885110

    Guggenmos D J, Azin M, Barbay S, Mahnken J D, Dunham C, Mohseni P and Nudo R J 2013 Temporal Basis Function Models28 Proceedings of the National Academy of Sciences11021177–21182 Restoration of function after brain damage using a neural prosthesis URL https://www.pnas.org/doi/...

    2013 doi

  36. [44]

    Murphy K P 2022Probabilistic Machine Learning: An introduction(MIT Press) URL probml.ai

  37. [45]

    Kim N H, Xie Z and van de Panne M 2020 Learning to correspond dynamical systems (Preprint 1912.03015) URLhttps://arxiv.org/abs/1912.03015

    2020 arXiv

  38. [46]

    Shi H and Meng M Q H 2022 Deep koopman operator with control for nonlinear systems (Preprint 2202.08004) URLhttps://arxiv.org/abs/2202.08004

    2022 arXiv

  39. [47]

    Jung M, da Costa Mendes P R, ¨Onnheim M and Gustavsson E 2023Engineering Applications of Artificial Intelligence123106226 Model predictive control when utilizing lstm as dynamic models ISSN 0952-1976 URL https://www.sciencedirect.com/science/article/ pii/S0952197623004104

    1976

  40. [48]

    Dugu´ e L, Marque P and VanRullen R 2011Journal of Neuroscience3111889–11893 The phase of ongoing oscillations mediates the causal relation between brain excitation and visual perception ISSN 0270-6474 URLhttps://www.jneurosci.org/content/31/33/11889

  41. [49]

    IEEE Engineering in Medicine and Biology Society

    Allison-Walker T J, Ann Hagan M, Chiang Price N S and Tat Wong Y 2020 Local field potential phase modulates neural responses to intracortical electrical stimulation.Annual International Conference of the IEEE Engineering in Medicine and Biology Society. IEEE Engineering in Med...

    2020

  42. [50]

    Nejatbakhsh A, Fumarola F, Esteki S, Toyoizumi T, Kiani R and Mazzucato L 2024Physical review research5Predicting the effect of micro-stimulation on macaque prefrontal activity based on spontaneous circuit dynamics

  43. [51]

    Venkatraman A, Hebert M and Bagnell J 2015Proceedings of the AAAI Conference on Artificial Intelligence29Improving multi-step prediction of learned time series models URL https: //ojs.aaai.org/index.php/AAAI/article/view/9590

  44. [52]

    Karamanakos P and Geyer T 2020IEEE Transactions on Power Electronics357434–7450 Guidelines for the design of finite control set model predictive controllers

  45. [53]

    Kraskov A, St¨ ogbauer H and Grassberger P 2004Phys. Rev. E69(6) 066138 Estimating mutual information URLhttps://link.aps.org/doi/10.1103/PhysRevE.69.066138

    doi

  46. [54]

    Efron B and Hastie T 2021Computer Age Statistical Inference, Student Edition: Algorithms, Evidence, and Data ScienceInstitute of Mathematical Statistics Monographs (Cambridge University Press)

  47. [55]

    Gretton A, Fukumizu K, Teo C, Song L, Sch¨ olkopf B and Smola A 2007 A kernel statistical test of independenceAdvances in Neural Information Processing Systemsvol 20 ed Platt J, Koller D, Singer Y and Roweis S (Curran Associates, Inc.) URL https://proceedings.neurips.cc/ paper...

    2007

  48. [56]

    Dataset and Processing Our data comes from a previously published excitatory optogenetic stimulation study [27]

    Appendix 6.1. Dataset and Processing Our data comes from a previously published excitatory optogenetic stimulation study [27]. That study involved the induction of plasticity using pairs of stimulation pulses which were spaced in time, with spacing ranging from 10-100ms. The p...

  49. [57]

    clock vector

    Their somatosensory (S1) and primary motor (M1) cortices were optogenetically photosensitized using a viral-mediated expression of the C1V1 opsin. Neural data were recorded using a µECoG array, sampled at 24kHz using a Tucker-Davis Technologies system (Florida, USA). Photosens...

Pith tools

Reviewed August 6, 2026 · model on record in the stance chip above.