Access Menu
BODYinTRANSIT @ MLSP 2025
Overview
This competition explores how the brain integrates sensory information to shape body perception. Using multimodal biosignal data collected during a sound-based body illusion experiment, participants must build robust machine learning models to predict perceived body weight changes.
The competition features two tasks:
- Primary Task: Classify whether individuals perceived their body weight as lighter, heavier, or unchanged.
- Beyond Classification Task: Investigate sensor contribution analysis or propose novel representation learning methods for a better understanding of multimodal biosignals.
Data was collected from 103 participants exposed to a sound-based body illusion, where walking sounds were manipulated to induce changes in body perception. The dataset includes 618 samples.
You can download Call for Competition from this link.
Task Description
Primary Task
Participants will classify changes in body perception by determining whether individuals perceive their body weight as lighter, heavier, or unchanged based on physiological signals (EMG, ECG, EDA, respiration) and kinematic signals (motion capture).
- The macro F1 score will be the primary evaluation metric, ensuring a fair assessment of all categories.
Beyond Classification
Participants are encouraged to explore representation learning techniques (e.g., self-supervised learning, feature embedding) or investigate sensor significance using explainable AI methods. The aim is to derive new insights from the BODYinTRANSIT dataset and enhance understanding of multimodal biosignals.
- We will evaluate this task based on their proposed methodology, interpretability, and impact on understanding multimodal biosignals.
Register your team to access the dataset using this link.
Submission Guidelines
Participants must submit their work through dedicated Google Forms based on their chosen task. The submission process ensures fairness, proper categorisation, and a structured evaluation of results.
A pytorch code has been made public in the following link https://github.com/tmcortes/MLSP_25_bit__baseline.git with the baseline implementation for the competition. The code includes utilities to correctly format the submissions.
Primary Task: Classification of Body Perception Changes
Participants working on the primary classification task must submit the following:
- Your model’s macro F1-score.
- The CSV files (in zip format) contain predicted labels for the evaluation set.
- A short report (PDF) detailing methodology, feature selection, preprocessing, and model performance.
Submission Link
Beyond Classification: Representation Learning & Sensor Contribution
Participants exploring sensor importance analysis or representation learning techniques should submit the following:
- A detailed report (PDF) explaining their approach, methodology, and key findings.
- (Optional) Code or models if they want to contribute to the research community.
Unlike the classification task, participants will not receive a separate test set since this task focuses on analysis rather than direct predictions.
Submission Link
Important dates (23:59 AoE)
- Data Access Request Deadline: April 1, 2025 (April 2, 2025 – 11:59 GMT)
- Submission Result Deadline: May 1, 2025 (May 2, 2025 – 11:59 GMT)
- Abstract Submission Deadline: May 5, 2025 (May 6, 2025 – 11:59 GMT)
- Paper Submission Deadline: May 20, 2025 (May 21, 2025 – 11:59 GMT)
- Notification of Acceptance: June 24, 2025 (June 25, 2025 – 11:59 GMT)
- Camera-Ready Paper Upload: July 15, 2025 (July 16, 2025 – 11:59 GMT)
*Please note that the deadlines will not be extended.