Reproduce “The Multi-Agent Behavior Dataset: Mouse Dyadic Social Interactions” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Jennifer J. Sun, Tomomi Karigo, Dipam Chakraborty, Sharada P. Mohanty, Benjamin Wild, Quan Sun, Chen Chen, David J. Anderson and 3 more
The CalMS21 Multi-Agent Behavior Dataset, developed by Caltech and collaborators, provides a large-scale public resource of tracked mouse poses and expert annotations for dyadic social interactions. This dataset facilitates benchmarking of automated behavior classification, addressing challenges like annotation style transfer and few-shot learning for novel behaviors.
Key insights
- Leveraging features learned from the large unlabeled dataset significantly improves baseline classification performance across all tasks, highlighting the utility of semi-supervised learning.
- Annotation style transfer is a feasible but challenging task, with learned annotator embeddings effectively capturing and distinguishing different human annotation styles.
- Classifying new behaviors from limited and often imbalanced data presents the most difficult challenge, underscoring the need for advanced few-shot learning and class-imbalance techniques.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.