Dendrite: A Real-Time Python Application for Online Brain-Computer Interface Research and Development

Jan Babi\v{c}; Niko Kroflic

REVIEW 2 major objections 5 minor 16 references

Dendrite is a single Python application that handles the full online BCI workflow—multimodal acquisition, live decoder training and hot-swapping, inference, replay, and provenance tracking—validated end-to-end on motor-imagery and P300 benc

Reviewed by Pith at T0; open to challenge. T0 means a machine referee read the full paper against a public rubric. the ladder, T0–T4 →

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

T0 review · deepseek-v4-flash

2026-08-02 01:26 UTC pith:OUHPZC2R

load-bearing objection A real, honest software-systems contribution to online BCI — the architecture is sound and the reported metrics are computed fairly — but the benchmark panels all run through replay rather than live hardware, so the real-time claims are conditional until that gap is closed. the 2 major comments →

arxiv 2607.14655 v1 pith:OUHPZC2R submitted 2026-07-16 cs.HC

Dendrite: A Real-Time Python Application for Online Brain-Computer Interface Research and Development

Niko Kroflic , Jan Babi\v{c} This is my paper

classification cs.HC

keywords brain-computer interfacereal-time processingonline decoder trainingmultimodal acquisitionprovenance trackingPythonEEGreproducible research

verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved

The pith

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

Dendrite is a Python application that claims to unify the entire online BCI workflow—multimodal streaming acquisition, real-time processing, decoder training and mid-session hot-swapping, live inference, offline replay, and traceable storage—in a single runtime. The authors argue this reduces the fragmentation that currently forces researchers to assemble separate acquisition, training, and analysis tools. They validate the system end-to-end on motor-imagery and P300 datasets, showing above-chance online decoding while retraining every ten epochs, with per-step inference in the low milliseconds and end-to-end closed-loop latency around 13 ms at the median. If correct, the contribution is a reproducible, open-source infrastructure that makes online BCI experiments easier to build, run, and audit.

Core claim

The central claim is that a single modifiable Python runtime can handle concurrent multimodal acquisition at native rates, execute multiple processing modes against one shared-memory ring buffer, train or update decoders mid-session without interrupting inference, and preserve full provenance by linking each decoder to its training configuration and source recordings. The system is validated by running the complete live pipeline—training on every ten epochs and hot-swapping the new decoder into an independently running asynchronous inference mode—across an in-house motor-imagery study, BNCI 2014-001, and BNCI 2014-009 P300, with online prequential accuracy above the dynamic no-information ra

What carries the argument

The design that carries the argument is a per-stream shared-memory ring buffer written by acquisition threads and read independently by three isolated processing subprocesses—synchronous (event-locked epochs), asynchronous (sliding-window classification), and neurofeedback (band power). Decoders are trained by a separate machine-learning subprocess that reads live-recorded epochs directly from the HDF5 file in single-writer/multi-reader mode and publishes the new model into shared state, from which running modes hot-swap it on the next step. A SQLite database records every recording, decoder, and training run, so each deployed decoder traces back to its configuration and source recording.

Load-bearing premise

The benchmark panels and motor-imagery latency figures come from replaying recorded sessions through the same code path that would handle a live amplifier; if replay understates real-hardware timing jitter, clock drift, or synchronization failures, the real-time and online-accuracy conclusions could be optimistic.

What would settle it

Run the identical motor-imagery and P300 pipelines against a live EEG amplifier, rather than a replay, and measure per-step inference latency, step cadence, and prequential accuracy; if the closed-loop latency exceeds the ~100 ms asynchronous step or accuracy drops to the no-information rate, the central claim fails.

Watch this falsifier — get emailed when new claim-graph text bears on it.

If this is right

A decoder can be retrained and replaced live while inference continues, enabling adaptive online experiments without downtime.
Every online decision can be audited, since each decoder stores its training run, configuration, and source recordings.
The same configuration drives live acquisition and recorded replay, so benchmark and live conditions share a code path.
Because the paradigm is external behind a simple event contract, task applications can be written in any language.
The reported latency suggests Python-based real-time BCI is practical, at least for decoders of the tested complexity.

Where Pith is reading between the lines

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

If replay fidelity to live hardware is imperfect—for example, if real amplifiers introduce more jitter or clock drift—the benchmark-accuracy numbers could overstate live performance; a direct live-versus-replay comparison would settle this.
The architecture's per-stream independent decoding means joint multimodal models are not supported; adding them would require a fusion step inside the runtime rather than after prediction.
The hot-swap mechanism could support more sophisticated online learning, such as incremental adaptation or cross-session transfer, without architectural changes.
Because the REST API exposes the same actions as the browser UI, the entire research loop—configuration sweeps, replay runs, result collection—could be automated by a script or agent, and every result would remain traceable.

Editorial analysis

A structured set of objections, weighed in public.

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

Desk Editor's Note

A real, honest software-systems contribution to online BCI — the architecture is sound and the reported metrics are computed fairly — but the benchmark panels all run through replay rather than live hardware, so the real-time claims are conditional until that gap is closed.

read the letter

Short take: this is a genuine software contribution, and the paper is honest about what it does not prove. The headline risk — replay standing in for live hardware — is real, but it is a validation gap, not a broken load-bearing wall.

What is actually new: a single Python runtime that handles multimodal acquisition at native rates, trains decoders mid-session, hot-swaps them into an asynchronous inference loop without stopping output, and tracks full provenance in SQLite/HDF5. The architecture is clean: per-stream ring buffers, isolated subprocess modes, and an external LSL event contract that keeps the paradigm language-agnostic. The evaluation matches the claims. Prequential accuracy is computed honestly with a disclosed forgetting factor, the P300 result reports kappa instead of leaning on the inflated 0.92 raw accuracy, and latency is decomposed by stage rather than presented as one magic number. The code is open-source and the design choices are well justified against the cited platforms.

Where it is soft, in order of real softness. First, all three benchmark panels and the motor-imagery latency numbers come from replaying recorded sessions through the offline replay manager. The paper asserts in Section 2.2 that the same code path runs against a live amplifier, but it never validates that side-by-side. Replay cannot exercise multi-device clock drift, dropped-sample handling, or real acquisition jitter. The only live measurement is the neurofeedback closed-loop latency in Section 3.3, which is a good piece of evidence but covers a simpler 4 Hz band-power path, not the benchmarked decoder path. Second, there is no head-to-head comparison with BCI2000, OpenViBE, or MEDUSA — the platforms the paper positions itself against. That is not fatal for a systems paper, but it would turn the fragmentation argument from a claim into a demonstration. Third, the latency figures are self-reported, though the measurement method is reasonable and the per-stage breakdown helps.

The Discussion already admits several limitations (simple retrain policy, small in-house datasets, no joint multimodal training), which makes me trust the authors. The missing replay-validation is not acknowledged there, which is the main addition I would ask for.

This paper is for BCI researchers who need a reproducible online loop without assembling their own framework. It deserves a serious referee, not a desk reject. I would send it back with a request for one live-acquisition benchmark panel (at least on the motor-imagery path) or a direct timing comparison between replay and live, plus a commit hash for v0.10 and ideally a short comparison table with the existing tools. The central architecture is sound, and the honest evaluation makes it worth engaging.

Referee Report

2 major / 5 minor

Summary. Dendrite is an open-source Python application for online BCI research that combines multimodal LSL acquisition at native rates, three processing modes (synchronous event-locked, asynchronous sliding-window, and neurofeedback band-power), mid-session decoder training and hot-swapping, HDF5/SQLite provenance, a browser-based interface, and an external task-application contract. The claimed contributions are per-stream acquisition/decoding, unification of online and offline analysis in one runtime, and low-friction reproducible analysis. The evaluation replays recorded sessions through the live backend for an in-house motor-imagery dataset, BNCI 2014-001, and BNCI 2014-009, reporting prequential accuracy and Cohen's kappa, and measures per-step inference latency (1-4 ms) plus live closed-loop neurofeedback latency (p50 ~13 ms, p99 ~66 ms).

Significance. If the real-time claims hold, Dendrite would be a useful open-source contribution to BCI software: it provides a Python-native runtime with hot-swappable decoders, an external paradigm contract, and full provenance from acquisition to deployed decoder. The evaluation has real strengths: it uses independent public datasets via MOABB, computes prequential accuracy honestly with a disclosed 0.95 forgetting factor, reports kappa rather than imbalance-inflated raw accuracy for P300, and decomposes latency by stage. The central weakness is that the benchmark real-time evidence comes from replay rather than live hardware, so the strong real-time conclusions in the Abstract and Discussion are conditional on an unverified replay-equivalence assumption.

major comments (2)

[§3.1 / §2.2] All three benchmark panels and the motor-imagery latency figures come from the offline replay manager (Section 3.1); only the neurofeedback latency (Section 3.3, Figures 3b/5) was measured on live hardware, and that path exercises band-power features rather than the benchmarked decoders. Section 2.2 asserts "the same code path runs against a saved session or a live amplifier," but no side-by-side validation is reported, and the Discussion's limitation list does not mention this. Replay pacing, clock behavior, and synchronization failure modes can differ from live multi-device acquisition; replay of one file cannot exercise multi-device drift or dropped-sample events. Because the Abstract and Section 4 conclude that the system trains and updates decoders "in real time," this assumption is load-bearing. Please add a live-vs-replay validation, document and validate the replay manager's paci
[§3.3] The claim that "cadence and inference latency are unaffected by training events" is presented as an architectural consequence, but no measurement is shown that isolates inference steps during retrain/hot-swap windows (e.g., a per-step latency trace with retrain start/end shaded). The same applies to the Discussion statement that "a training event never perturbed cadence." This is central to the hot-swap contribution. Please provide a timing trace from a replay or live run showing per-step latency across the entire session with retrain events marked, or soften the claim to "retraining executes off the inference path" without asserting measured non-perturbation.

minor comments (5)

[§2.2] The cross-reference to the "offline replay manager (Section 2.6)" is inaccurate: Section 2.6 is titled "Web interface." The replay manager's pacing and timing behavior are never specified. Fix the cross-reference and add a brief description of the replay path.
[Figure 4] The caption says series are defined in the figure key, but no key is visible. Add a legend or define the shaded band and dashed NIR line in the caption.
[Figure 3b] The y-axis label appears to read "relative -power"; it should read "relative α-power" or the Greek letter should be rendered.
[Figure 5a] Please reconcile the subject counts: the text says five live neurofeedback subjects, while the figure annotation mentions "6 subjects." Clarify whether the motor-imagery subject is included and which series is which.
[§3.1] The text says "the same harness ran against four datasets," but the exoskeleton recording is only an acquisition showcase and does not contribute to the online-learning panels. Consider saying "three datasets for online learning plus one acquisition showcase" to avoid ambiguity.

Circularity Check

0 steps flagged

No circularity: the online-learning results are genuine prequential evaluations on external public datasets, and the remaining concerns are external-validity assumptions, not circular reductions.

full rationale

This is a software-systems paper whose claims are validated empirically, not derived from a mathematical model. The online-learning evaluations are prequential: 'Each trial is predicted before its label is seen and then folded into an incrementally retrained decoder under 0.95 forgetting' (Section 3.2), so the accuracy numbers are not fitted inputs renamed as predictions. The benchmark panels use external public datasets loaded through MOABB (BNCI 2014-001 and BNCI 2014-009), which are independent of the authors, plus an in-house motor-imagery set that is used to exercise the runtime rather than to establish a novel algorithmic result. The paper explicitly disclaims algorithmic novelty: 'Dendrite does not require a specific decoder family and does not claim algorithmic novelty in the classifiers themselves' (Section 2.4). No load-bearing self-citation appears; the reference list contains no works by Kroflic or Babič. The replay-to-live equivalence ('the same code path runs against a saved session or a live amplifier', Section 2.2) is an asserted assumption about the test harness, and the paper does report limitations ('The present evaluation has several limitations', Section 4), including that the neurofeedback latency was measured with one external application and that the in-house datasets are modest. These are external-validity or correctness risks, not circularity: no quantity is defined in terms of the quantity it is used to predict, and no prediction is statistically forced by construction. The central real-time and online-decoding claims therefore rest on independent empirical evaluation rather than on circular reasoning.

Axiom & Free-Parameter Ledger

4 free parameters · 4 axioms · 0 invented entities

The paper introduces no new physical or mathematical entities. Its free parameters are all explicit configuration choices (forgetting factor, retrain interval, step grid, feature update rate) that shape the reported metrics, and its load-bearing premises are the LSL synchronization fidelity, the replay-live equivalence, and label reliability.

free parameters (4)

0.95 exponential forgetting factor = 0.95
Hand-chosen smoothing in the prequential scoring (Section 3.1). The reported accuracy is a trailing weighted average and depends directly on this constant.
Retrain interval N = 10 epochs = 10
Hand-chosen policy used to exercise the runtime (Section 3.1). All reported accuracy and κ values are specific to this retrain cadence.
Asynchronous step grid = 100 ms (MI) / ~97.7 ms (P300)
Configuration choice in Section 3.3 against which the 1–4 ms inference latency is compared.
Neurofeedback feature update rate = 4 Hz
Configurable update rate (Section 2.3.3); the closed-loop latency comparison (p50 ≈13 ms) is interpreted relative to this cadence.

axioms (4)

domain assumption LSL clock-synchronization adequately corrects timing offsets between devices
Section 2.2 relies on LSL's synchronization protocol for cross-stream temporal alignment; incorrect offsets would corrupt epoch alignment and the latency figures.
domain assumption Replaying recorded files over LSL exercises the same acquisition code path as a live amplifier
Sections 2.2 and 3.1: all three benchmark panels and the MI latency numbers are generated through the offline replay manager; the validity of the real-time claims depends on this equivalence. This is also the paper's weakest assumption.
domain assumption Dataset labels and event markers are reliable ground truth
Section 3.1: BNCI trigger labels and in-house trial labels are taken as correct for computing prequential accuracy.
standard math Standard DSP/ML building blocks (IIR filtering, CSP, xDAWN, covariance tangent space) behave as implemented by their libraries
The paper assumes correctness of scikit-learn, MNE, and related implementations; no algorithm-level novelty is claimed (Section 2.4).

pith-pipeline@v1.3.0-alltime-deepseek · 10644 in / 9120 out tokens · 93381 ms · 2026-08-02T01:26:37.394490+00:00 · methodology

0 comments

read the original abstract

Online brain-computer interface research requires software that can acquire multimodal physiological data, train and update decoders, run live inference, and preserve the full experimental provenance in a reproducible workflow. We present Dendrite, a real-time brain-computer interface application in Python that brings signal acquisition, decoder training, and live inference together in a single, ready-to-run application that stays modifiable. Dendrite records several signal streams at once, each at its native rate, and executes multiple processing modes concurrently against them. A decoder can start from a previously trained model or be fit mid-session while the pipeline keeps running, and the same recordings feed offline training in the same application. Each recording, decoder, and training run is tracked in a database, and every decoder records the configuration and the source recordings it was trained from, so a deployed decoder traces back to what produced it. The experimental paradigm stays external, an independent program in any language that reaches Dendrite over the network, rather than a module built inside the runtime. We validate the full system end-to-end on in-house and public BCI datasets, training and updating decoders online while the pipeline runs in real time. Dendrite is open-source under the GPL-3.0 license at https://github.com/dendrite-bci/dendrite. The result is a reproducible, open-source biomedical-computing system for developing and evaluating online BCI paradigms.

Figures

Figures reproduced from arXiv: 2607.14655 by Jan Babi\v{c}, Niko Kroflic.

**Figure 2.** Figure 2: Dendrite Control view (v0.10) during a live recording. The centre panel shows live [PITH_FULL_IMAGE:figures/full_fig_p007_2.png] view at source ↗

**Figure 3.** Figure 3: Acquisition and closed-loop output from one in-house exoskeleton recording. [PITH_FULL_IMAGE:figures/full_fig_p009_3.png] view at source ↗

**Figure 4.** Figure 4: Online learning across three benchmark panels: in-house MI (CSP+LDA), BNCI 2014-001 [PITH_FULL_IMAGE:figures/full_fig_p009_4.png] view at source ↗

**Figure 5.** Figure 5: Closed-loop latency by stage and subject. [PITH_FULL_IMAGE:figures/full_fig_p011_5.png] view at source ↗

discussion (0)

Reference graph

Works this paper leans on

16 extracted references · 1 linked inside Pith

[1]

Optuna: A next-generation hyperparameter optimization framework

Takuya Akiba, Shotaro Sano, Toshihiko Yanase, Takeru Ohta, and Masanori Koyama. Optuna: A next-generation hyperparameter optimization framework. In Proceedings of the 25th ACM SIGKDD International Conference on Knowledge Discovery & Data Mining, pages 2623--2631, 2019. doi:10.1145/3292500.3330701

arXiv 2019
[2]

Gorgolewski, Tibor Auer, Vince D

Krzysztof J. Gorgolewski, Tibor Auer, Vince D. Calhoun, R. Cameron Craddock, Samir Das, Eugene P. Duff, Guillaume Flandin, Satrajit S. Ghosh, Tristan Glatard, Yaroslav O. Halchenko, Daniel A. Handwerker, Michael Hanke, David Keator, Xiangrui Li, Zachary Michael, Camille Maumet, B. Nolan Nichols, Thomas E. Nichols, John Pellman, Jean-Baptiste Poline, Ariel...

2016
[3]

a m \"a l \

Alexandre Gramfort, Martin Luessi, Eric Larson, Denis A. Engemann, Daniel Strohmeier, Christian Brodbeck, Roman Goj, Mainak Jas, Teon Brooks, Lauri Parkkonen, and Matti S. H \"a m \"a l \"a inen. MEG and EEG data analysis with MNE-Python . Frontiers in Neuroscience, 7: 0 267, 2013. ISSN 1662-453X. doi:10.3389/fnins.2013.00267

arXiv 2013
[4]

MOABB : Trustworthy algorithm benchmarking for BCI s

Vinay Jayaram and Alexandre Barachant. MOABB : Trustworthy algorithm benchmarking for BCI s. Journal of Neural Engineering, 15 0 (6): 0 066011, 2018. ISSN 1741-2552. doi:10.1088/1741-2552/aadea0

doi:10.1088/1741-2552/aadea0 2018
[5]

Grivich, Fiorenzo Artoni, Tim Mullen, Arnaud Delorme, and Scott Makeig

Christian Kothe, Seyed Yahya Shirazi, Tristan Stenner, David Medine, Chadwick Boulay, Matthew I. Grivich, Fiorenzo Artoni, Tim Mullen, Arnaud Delorme, and Scott Makeig. The lab streaming layer for synchronized multimodal recording. Imaging Neuroscience, 3: 0 IMAG.a.136, 2025. ISSN 2837-6056. doi:10.1162/IMAG.a.136

doi:10.1162/imag.a.136 2025
[6]

Lawhern, Amelia J

Vernon J. Lawhern, Amelia J. Solon, Nicholas R. Waytowich, Stephen M. Gordon, Chou P. Hung, and Brent J. Lance. EEGNet : A compact convolutional neural network for EEG -based brain-computer interfaces. Journal of Neural Engineering, 15 0 (5): 0 056013, 2018. doi:10.1088/1741-2552/aace8c

doi:10.1088/1741-2552/aace8c 2018
[7]

Robot operating system 2: Design, architecture, and uses in the wild

Steven Macenski, Tully Foote, Brian Gerkey, Chris Lalancette, and William Woodall. Robot operating system 2: Design, architecture, and uses in the wild. Science Robotics, 7 0 (66): 0 eabm6074, 2022. doi:10.1126/scirobotics.abm6074

doi:10.1126/scirobotics.abm6074 2022
[8]

MetaBCI : An open-source platform for brain-computer interfaces

Jie Mei, Ruixin Luo, Lichao Xu, Wei Zhao, Shengfu Wen, Kun Wang, Xiaolin Xiao, Jiayuan Meng, Yongzhi Huang, Jiabei Tang, Longlong Cheng, Minpeng Xu, and Dong Ming. MetaBCI : An open-source platform for brain-computer interfaces. Computers in Biology and Medicine, 168: 0 107806, 2024. ISSN 0010-4825. doi:10.1016/j.compbiomed.2023.107806

arXiv 2024
[9]

PyTorch : An imperative style, high-performance deep learning library

Adam Paszke, Sam Gross, Francisco Massa, Adam Lerer, James Bradbury, Gregory Chanan, Trevor Killeen, Zeming Lin, Natalia Gimelshein, Luca Antiga, Alban Desmaison, Andreas K \"o pf, Edward Yang, Zachary DeVito, Martin Raison, Alykhan Tejani, Sasank Chilamkurthy, Benoit Steiner, Lu Fang, Junjie Bai, and Soumith Chintala. PyTorch : An imperative style, high-...

2019
[10]

Scikit-learn: Machine learning in P ython

Fabian Pedregosa, Ga \"e l Varoquaux, Alexandre Gramfort, Vincent Michel, Bertrand Thirion, Olivier Grisel, Mathieu Blondel, Peter Prettenhofer, Ron Weiss, Vincent Dubourg, Jake Vanderplas, Alexandre Passos, David Cournapeau, Matthieu Brucher, Matthieu Perrot, and \'E douard Duchesnay. Scikit-learn: Machine learning in P ython. Journal of Machine Learning...

2011
[11]

OpenViBE : An open-source software platform to design, test, and use brain-computer interfaces in real and virtual environments

Yann Renard, Fabien Lotte, Guillaume Gibert, Marco Congedo, Emmanuel Maby, Vincent Delannoy, Olivier Bertrand, and Anatole L \'e cuyer. OpenViBE : An open-source software platform to design, test, and use brain-computer interfaces in real and virtual environments. Presence: Teleoperators and Virtual Environments, 19 0 (1): 0 35--53, 2010. ISSN 1054-7460. ...

doi:10.1162/pres.19.1.35 2010
[12]

Attention and P300 -based BCI performance in people with amyotrophic lateral sclerosis

Angela Riccio, Luca Simione, Francesca Schettini, Alessia Pizzimenti, Maurizio Inghilleri, Marta Olivetti Belardinelli, Donatella Mattia, and Febo Cincotti. Attention and P300 -based BCI performance in people with amyotrophic lateral sclerosis. Frontiers in Human Neuroscience, 7: 0 732, 2013. ISSN 1662-5161. doi:10.3389/fnhum.2013.00732

arXiv 2013
[13]

MEDUSA : A novel P ython-based software ecosystem to accelerate brain-computer interface and cognitive neuroscience research

Eduardo Santamar \'i a-V \'a zquez, V \'i ctor Mart \'i nez-Cagigal, Diego Marcos-Mart \'i nez, V \'i ctor Rodr \'i guez-Gonz \'a lez, Sergio P \'e rez-Velasco, Selene Moreno-Calder \'o n, and Roberto Hornero. MEDUSA : A novel P ython-based software ecosystem to accelerate brain-computer interface and cognitive neuroscience research. Computer Methods and ...

arXiv 2023
[14]

McFarland, Thilo Hinterberger, Niels Birbaumer, and Jonathan R

Gerwin Schalk, Dennis J. McFarland, Thilo Hinterberger, Niels Birbaumer, and Jonathan R. Wolpaw. BCI2000 : A general-purpose brain-computer interface ( BCI ) system. IEEE Transactions on Biomedical Engineering, 51 0 (6): 0 1034--1043, 2004. doi:10.1109/TBME.2004.827072

arXiv 2004
[15]

Deep learning with convolutional neural networks for EEG decoding and visualization

Robin Tibor Schirrmeister, Jost Tobias Springenberg, Lukas Dominique Josef Fiederer, Martin Glasstetter, Katharina Eggensperger, Michael Tangermann, Frank Hutter, Wolfram Burgard, and Tonio Ball. Deep learning with convolutional neural networks for EEG decoding and visualization. Human Brain Mapping, 38 0 (11): 0 5391--5420, 2017. ISSN 1097-0193. doi:10.1...

doi:10.1002/hbm.23730 2017
[16]

u ller, Ad Aertsen, Niels Birbaumer, Christoph Braun, Clemens Brunner, Robert Leeb, Carsten Mehring, Kai J. Miller, Gernot R. M \

Michael Tangermann, Klaus-Robert M \"u ller, Ad Aertsen, Niels Birbaumer, Christoph Braun, Clemens Brunner, Robert Leeb, Carsten Mehring, Kai J. Miller, Gernot R. M \"u ller-Putz, Guido Nolte, Gert Pfurtscheller, Hubert Preissl, Gerwin Schalk, Alois Schl \"o gl, Carmen Vidaurre, Stephan Waldert, and Benjamin Blankertz. Review of the BCI competition IV . F...

Pith/arXiv arXiv 2012

[1] [1]

Optuna: A next-generation hyperparameter optimization framework

Takuya Akiba, Shotaro Sano, Toshihiko Yanase, Takeru Ohta, and Masanori Koyama. Optuna: A next-generation hyperparameter optimization framework. In Proceedings of the 25th ACM SIGKDD International Conference on Knowledge Discovery & Data Mining, pages 2623--2631, 2019. doi:10.1145/3292500.3330701

arXiv 2019

[2] [2]

Gorgolewski, Tibor Auer, Vince D

Krzysztof J. Gorgolewski, Tibor Auer, Vince D. Calhoun, R. Cameron Craddock, Samir Das, Eugene P. Duff, Guillaume Flandin, Satrajit S. Ghosh, Tristan Glatard, Yaroslav O. Halchenko, Daniel A. Handwerker, Michael Hanke, David Keator, Xiangrui Li, Zachary Michael, Camille Maumet, B. Nolan Nichols, Thomas E. Nichols, John Pellman, Jean-Baptiste Poline, Ariel...

2016

[3] [3]

a m \"a l \

Alexandre Gramfort, Martin Luessi, Eric Larson, Denis A. Engemann, Daniel Strohmeier, Christian Brodbeck, Roman Goj, Mainak Jas, Teon Brooks, Lauri Parkkonen, and Matti S. H \"a m \"a l \"a inen. MEG and EEG data analysis with MNE-Python . Frontiers in Neuroscience, 7: 0 267, 2013. ISSN 1662-453X. doi:10.3389/fnins.2013.00267

arXiv 2013

[4] [4]

MOABB : Trustworthy algorithm benchmarking for BCI s

Vinay Jayaram and Alexandre Barachant. MOABB : Trustworthy algorithm benchmarking for BCI s. Journal of Neural Engineering, 15 0 (6): 0 066011, 2018. ISSN 1741-2552. doi:10.1088/1741-2552/aadea0

doi:10.1088/1741-2552/aadea0 2018

[5] [5]

Grivich, Fiorenzo Artoni, Tim Mullen, Arnaud Delorme, and Scott Makeig

Christian Kothe, Seyed Yahya Shirazi, Tristan Stenner, David Medine, Chadwick Boulay, Matthew I. Grivich, Fiorenzo Artoni, Tim Mullen, Arnaud Delorme, and Scott Makeig. The lab streaming layer for synchronized multimodal recording. Imaging Neuroscience, 3: 0 IMAG.a.136, 2025. ISSN 2837-6056. doi:10.1162/IMAG.a.136

doi:10.1162/imag.a.136 2025

[6] [6]

Lawhern, Amelia J

Vernon J. Lawhern, Amelia J. Solon, Nicholas R. Waytowich, Stephen M. Gordon, Chou P. Hung, and Brent J. Lance. EEGNet : A compact convolutional neural network for EEG -based brain-computer interfaces. Journal of Neural Engineering, 15 0 (5): 0 056013, 2018. doi:10.1088/1741-2552/aace8c

doi:10.1088/1741-2552/aace8c 2018

[7] [7]

Robot operating system 2: Design, architecture, and uses in the wild

Steven Macenski, Tully Foote, Brian Gerkey, Chris Lalancette, and William Woodall. Robot operating system 2: Design, architecture, and uses in the wild. Science Robotics, 7 0 (66): 0 eabm6074, 2022. doi:10.1126/scirobotics.abm6074

doi:10.1126/scirobotics.abm6074 2022

[8] [8]

MetaBCI : An open-source platform for brain-computer interfaces

Jie Mei, Ruixin Luo, Lichao Xu, Wei Zhao, Shengfu Wen, Kun Wang, Xiaolin Xiao, Jiayuan Meng, Yongzhi Huang, Jiabei Tang, Longlong Cheng, Minpeng Xu, and Dong Ming. MetaBCI : An open-source platform for brain-computer interfaces. Computers in Biology and Medicine, 168: 0 107806, 2024. ISSN 0010-4825. doi:10.1016/j.compbiomed.2023.107806

arXiv 2024

[9] [9]

PyTorch : An imperative style, high-performance deep learning library

Adam Paszke, Sam Gross, Francisco Massa, Adam Lerer, James Bradbury, Gregory Chanan, Trevor Killeen, Zeming Lin, Natalia Gimelshein, Luca Antiga, Alban Desmaison, Andreas K \"o pf, Edward Yang, Zachary DeVito, Martin Raison, Alykhan Tejani, Sasank Chilamkurthy, Benoit Steiner, Lu Fang, Junjie Bai, and Soumith Chintala. PyTorch : An imperative style, high-...

2019

[10] [10]

Scikit-learn: Machine learning in P ython

Fabian Pedregosa, Ga \"e l Varoquaux, Alexandre Gramfort, Vincent Michel, Bertrand Thirion, Olivier Grisel, Mathieu Blondel, Peter Prettenhofer, Ron Weiss, Vincent Dubourg, Jake Vanderplas, Alexandre Passos, David Cournapeau, Matthieu Brucher, Matthieu Perrot, and \'E douard Duchesnay. Scikit-learn: Machine learning in P ython. Journal of Machine Learning...

2011

[11] [11]

OpenViBE : An open-source software platform to design, test, and use brain-computer interfaces in real and virtual environments

Yann Renard, Fabien Lotte, Guillaume Gibert, Marco Congedo, Emmanuel Maby, Vincent Delannoy, Olivier Bertrand, and Anatole L \'e cuyer. OpenViBE : An open-source software platform to design, test, and use brain-computer interfaces in real and virtual environments. Presence: Teleoperators and Virtual Environments, 19 0 (1): 0 35--53, 2010. ISSN 1054-7460. ...

doi:10.1162/pres.19.1.35 2010

[12] [12]

Attention and P300 -based BCI performance in people with amyotrophic lateral sclerosis

Angela Riccio, Luca Simione, Francesca Schettini, Alessia Pizzimenti, Maurizio Inghilleri, Marta Olivetti Belardinelli, Donatella Mattia, and Febo Cincotti. Attention and P300 -based BCI performance in people with amyotrophic lateral sclerosis. Frontiers in Human Neuroscience, 7: 0 732, 2013. ISSN 1662-5161. doi:10.3389/fnhum.2013.00732

arXiv 2013

[13] [13]

MEDUSA : A novel P ython-based software ecosystem to accelerate brain-computer interface and cognitive neuroscience research

Eduardo Santamar \'i a-V \'a zquez, V \'i ctor Mart \'i nez-Cagigal, Diego Marcos-Mart \'i nez, V \'i ctor Rodr \'i guez-Gonz \'a lez, Sergio P \'e rez-Velasco, Selene Moreno-Calder \'o n, and Roberto Hornero. MEDUSA : A novel P ython-based software ecosystem to accelerate brain-computer interface and cognitive neuroscience research. Computer Methods and ...

arXiv 2023

[14] [14]

McFarland, Thilo Hinterberger, Niels Birbaumer, and Jonathan R

Gerwin Schalk, Dennis J. McFarland, Thilo Hinterberger, Niels Birbaumer, and Jonathan R. Wolpaw. BCI2000 : A general-purpose brain-computer interface ( BCI ) system. IEEE Transactions on Biomedical Engineering, 51 0 (6): 0 1034--1043, 2004. doi:10.1109/TBME.2004.827072

arXiv 2004

[15] [15]

Deep learning with convolutional neural networks for EEG decoding and visualization

Robin Tibor Schirrmeister, Jost Tobias Springenberg, Lukas Dominique Josef Fiederer, Martin Glasstetter, Katharina Eggensperger, Michael Tangermann, Frank Hutter, Wolfram Burgard, and Tonio Ball. Deep learning with convolutional neural networks for EEG decoding and visualization. Human Brain Mapping, 38 0 (11): 0 5391--5420, 2017. ISSN 1097-0193. doi:10.1...

doi:10.1002/hbm.23730 2017

[16] [16]

u ller, Ad Aertsen, Niels Birbaumer, Christoph Braun, Clemens Brunner, Robert Leeb, Carsten Mehring, Kai J. Miller, Gernot R. M \

Michael Tangermann, Klaus-Robert M \"u ller, Ad Aertsen, Niels Birbaumer, Christoph Braun, Clemens Brunner, Robert Leeb, Carsten Mehring, Kai J. Miller, Gernot R. M \"u ller-Putz, Guido Nolte, Gert Pfurtscheller, Hubert Preissl, Gerwin Schalk, Alois Schl \"o gl, Carmen Vidaurre, Stephan Waldert, and Benjamin Blankertz. Review of the BCI competition IV . F...

Pith/arXiv arXiv 2012