We are excited to announce the Multi-Agent Behavior (MABe) Challenge, a competition to build improved tools for the detection of multi-agent behaviors in neuroscience data! What: A machine learning competition to build supervised behavior classification tools, hosted on AIcrowd. When: The Challenge runs from March 8 2021 - April 30 2021 Where: https://www.aicrowd.com/challenges/multi-agent-behavior-representation-model... Why: To understand the brain, we must understand how it controls behavior. The study of that behavior has classically relied on manual annotation of large volumes of videos. Automating the behavior annotation process would allow researchers to increase the speed, volume, and objectivity of their experiments, and to investigate new behaviors they otherwise would not have time to study. What's in it for me: the three challenge tasks share a $9000 prize pool sponsored by Northwestern University and Amazon SageMarker. The first 50 submissions to exceed any task baseline by at least 5% will receive $200 in Amazon SageMaker credits (max 2/team). Top performing teams will also be invited to present their work at the Multi-Agent Behavior Workshop at CVPR 2021 in June (workshop details coming soon). The challenge consists of three tasks: 1) develop methods to classify human-defined behaviors from tracked pose trajectories in a large dataset of videos of socially interacting mice, 2) fine-tune classifications to different annotator styles, and 3) learn to recognize new behaviors of interest from limited training examples. Relevant techniques include supervised learning, many-to-one time series modeling, style transfer/transfer learning, few-shot learning, domain-sensitive data augmentation, and unsupervised feature learning/clustering. Organized by Ann Kennedy (Northwestern University), Jennifer J. Sun (Caltech), Yisong Yue (Caltech), and Pietro Perona (Caltech), with data contributed by Tomomi Karigo and other members of the David Anderson lab (Caltech). Happy classifying!