Research map / Multi-agent systems and games / Human-AI and goal-conditioned
Relax, it doesn’t matter how you get there: A new self-supervised approach for multi-timescale behavior analysis
Mehdi Azabou, Michael Jacob Mendelson, Nauman Ahad, Maks Sorokin, Shantanu Thakoor, Carolina Urzay, Eva L Dyer
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| Add the Histogram of Actions prediction objective (ablation) | BAMS · MABe Mouse Triplets | Average sequence-level MSE: 0.093100 -> 0.090415 | Improved |
| Add the Histogram of Actions prediction objective (ablation) | BAMS · MABe Mouse Triplets | Average sequence-level F1: 68.30 -> 80.12 | Improved |
| Add the Histogram of Actions prediction objective (ablation) | BAMS · MABe Mouse Triplets | Average frame-level F1: 19.46 -> 19.85 | Improved |
| Add multi-timescale bootstrapping (ablation) | BAMS · MABe Mouse Triplets | Average sequence-level F1: 78.11 -> 80.12 | Improved |
| Add multi-timescale bootstrapping (ablation) | BAMS · MABe Mouse Triplets | Average frame-level F1: 19.45 -> 19.85 | Improved |
| Separate the short- and long-term representations instead of using one space (ablation) | BAMS · MABe Mouse Triplets | Average sequence-level F1: 73.37 -> 80.12 | Improved |
| Separate the short- and long-term representations instead of using one space (ablation) | BAMS · MABe Mouse Triplets | Average frame-level F1: 19.52 -> 19.85 | Improved |
| Use temporal contrastive loss in TS2Vec rather than instance contrastive loss (ablation) | TS2Vec · MABe Mouse Triplets — Strain | F1-score: 57.12 -> 45.82 | Hurt |
| Use temporal contrastive loss in TS2Vec rather than instance contrastive loss (ablation) | TS2Vec · MABe Mouse Triplets — Huddle | F1-score: 24.74 -> 9.38 | Hurt |
| Pre-train on the full dataset rather than the training split only | BAMS · MABe Mouse Triplets — Strain | F1-score: 83.44 -> 88.23 | Improved |
| Use HoA future-distribution prediction instead of sequential future prediction (ablation) | BAMS · MABe Mouse Triplets | Maximum prediction horizon: 10 -> 30 | Improved |
Closest papers by meaning
- MABe22: A Multi-Species Multi-Task Benchmark for Learned Representations of Behavior (ICML 2023 · similarity 0.91)
- The Multi-Agent Behavior Dataset: Mouse Dyadic Social Interactions (NeurIPS 2021 · similarity 0.89)
- BehaVERT: A Transformer-Based Motion Language Model for Decoding Behavioral Semantics in Mice (IJCV 2026 · similarity 0.89)
- Learning Reusable Motor Motifs for Continuous Animal Behavior Modeling (NeurIPS 2026 · similarity 0.88)
- Agent-to-Sim: Learning Interactive Behavior Models from Casual Longitudinal Videos (ICLR 2025 · similarity 0.86)
- Behavioral Exploration: Learning to Explore via In-Context Adaptation (ICML 2025 · similarity 0.86)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
