Amazon's GEB grounds entity biographies for long-video memory, hits 72% on EgoLifeQA
amazon · hf · 2026-09-30
Amazon researchers introduced Grounded Entity Biographies (GEB), a long-video memory framework that resolves physical entity identity across hours-long videos.
- Problem: chronological descriptions and text-derived entities can't resolve physical identity — different objects share descriptions, and observations of the same object remain disconnected across events.
- Method: GEB groups visually grounded observations of the same physical instance across clips into retrievable biographies; at QA time, the biography is retrieved alongside episodic evidence, letting the model follow an entity via identity links built during memory construction.
- Results: improvements across four benchmarks including day- and week-long recordings; on EgoLifeQA, GEB reaches 72.0% accuracy, 4.4 points above the best published result. Ablations show grounded identity association and biography reading both contribute.
More from Research
- 16 Mathematicians Publish Leiden Declaration on AI's Role in Mathematics Research — burny_tech · 2026-09-30
- Repeated Solutions Make Reasoning Fragile After Instruction Tuning, Paper Finds — burny_tech · 2026-09-30
- AI is the discipline of building minds, and mathematics is just getting started — burny_tech · 2026-09-30
- A new auto-research loop that bootstraps the shape of the best possible result — burny_tech · 2026-09-30
- EasyPPO: just fix the critic — stable PPO for LLM post-training with zero training collapse — teortaxesTex · 2026-09-30
- Looped MoE Tuning: 2x Experts, 0.5x Looped Layers, 2x Loops, Attention Untied — burny_tech · 2026-09-30