ICML 2026 WORKSHOP

Neural Representations for Brain Data

Date: TBA
Location: Seoul, South Korea

Brain–computer interfaces (BCIs) are advancing rapidly: while invasive systems already benefit from well-established neural representations, non-invasive modalities (e.g., EEG/MEG) are reaching the scale needed for systematic benchmarking. Yet the field still lacks consensus on how brain signals should be represented for machine learning, representing a core limitation for generalization, robustness, and data efficiency.

Unlike speech or vision, which converged on standardized inputs such as mel spectrograms, BCI research remains fragmented, often defaulting to raw or minimally processed signals despite the complexity and noise of neural data. This workshop focuses on a central question: What should representation learning and feature engineering look like for brain data in the era of deep learning?

We invite ML researchers, neuroscientists, and BCI practitioners to benchmark representations across modalities, compare engineered and learned features, and assess whether the field needs universal or task-specific encodings. By extracting the principles that govern effective neural representations, the workshop aims to build a coherent framework for modeling brain signals, directly aligned with ICML's core interest in advancing machine learning. Our long-term goal is to catalyze BCIs' "mel-spectrogram moment": determining whether the field can converge on shared representations or requires task-specific schemes. By creating shared evaluation infrastructure and design principles, the workshop aims to accelerate progress toward robust, generalizable, and deployable brain-AI systems.

Tentative Schedule

Time	Session
09:00–09:10	Opening Remarks
09:10–10:00	Talk 1: Mirco Ravanelli (Concordia University/Mila)
10:00–10:50	Talk 2: Alexandre Gramfort (Meta Reality Labs)
10:50–11:10	Coffee Break & Networking
11:10–12:00	Talk 3: Natalie Voets (University of Oxford)
12:00–12:30	Spotlight Session 1: Selected papers (10 min each)
12:30–13:00	Morning Panel with Ravanelli, Gramfort, Voets, spotlight presenters
13:00–14:00	Lunch, Poster, and BCI Demo & Neural Art Sessions
14:00–14:50	Talk 4: SueYeon Chung (Harvard University)
14:50–15:40	Talk 5: Juan Helen Zhou (National University of Singapore)
15:40–16:00	Afternoon Coffee Break & Networking
16:00–16:30	Spotlight Session 2: Selected papers (10 min each)
16:30–17:15	Panel Discussion with all speakers
17:15+	Workshop Social

Invited Speakers

Mirco Ravanelli

Concordia University, Université de Montréal, Mila Quebec AI Institute

Natalie Voets

University of Oxford

Alexandre Gramfort

Meta Reality Labs

SueYeon Chung

Harvard University

Juan Helen Zhou

National University of Singapore

Yukiyasu Kamitani

Kyoto University, ATR Computational Neuroscience Laboratories

Call For Papers

We invite researchers at the intersection of machine learning, neuroscience, and related fields to submit their recent work on neural representations for BCIs. Accepted papers will be presented as spotlight talks or posters during the workshop.

Topics of Interest

Topics of interest include, but are not limited to:

Time–frequency–phase representations for EEG/MEG/ECoG
Spatial architectures for multi-channel neural signals: CNNs, graph neural networks, or attention mechanisms
Cross-spectral features (Riemannian geometry on covariance matrices)
Learned vs. engineered features for neural decoding: identifying when each approach excels and why
Foundation models and multi-task representations across speech/motor BCIs
Universal vs. task-specific neural representations: can one representation serve all BCI applications?

Submission Tracks

We offer two submission tracks:

Full Papers: 4–8 pages (excluding references and appendices), with one additional page permitted upon acceptance.
Short Papers: Up to 4 pages at submission time (excluding references and appendices). Accepted short papers may add one page for the camera-ready version.

References and appendices do not count toward the page limit. All submissions must use the ICML 2026 style. Accepted papers may add one extra page for the camera-ready version.

Reviewing Process

Submissions will undergo double-blind review on OpenReview. Each paper will receive at least two reviews evaluating novelty, clarity, technical soundness, and relevance, with reviewers encouraged to provide constructive, actionable feedback. All conflicts of interest will be handled by the program chairs to ensure a fair and unbiased review process.

Papers will be selected for oral (spotlight) or poster presentation based on quality, originality, and potential to stimulate discussion, with an emphasis on assembling a diverse and balanced program. Authors of accepted papers will submit a camera-ready version incorporating reviewer feedback. All accepted papers will be publicly available on OpenReview before the workshop.

Timeline

All deadlines are 11:59pm AoE (Anywhere on Earth).

Submission opens: 9 April 2026
Submission deadline (both tracks): 13 May 2026
Review period: 13–26 May 2026
Acceptance notification: 2 June 2026
Camera-ready deadline: 16 June 2026

Accepted papers will be presented during poster sessions, with exceptional submissions selected for spotlight oral presentations. All accepted papers will be made publicly available as non-archival reports, allowing for future submissions to archival conferences or journals.

Demos & Neural Art Track

In addition to research presentations, we introduce a Demos & Neural Art track connecting neural representations with generative models driven directly by brain activity. This track features systems from both academia and industry, including Prof. Yukiyasu Kamitani on visual imagery decoding, the OBVIOUS collective (IMAGINE mind-to-image series), and Blackrock Neurotech's brain-controlled art and music.