CoGR: Co-Evolving Generative Retrieval Trained with Reinforcement Learning
_akhaliq · x · 2026-09-04
CoGR (Co-evolving Generative Retrieval) trains LLMs to generate retrieval keywords for both user queries and apps.
- Co-evolving framework alternately optimizes one side while fixing the other, letting query and app keywords adapt to each other
- Iterative training yields steady improvements in retrieval performance
Related event: Apple Open-Sources CoGR for Co-Evolved Generative Retrieval(3 posts)→
More from Research
- 10 Claude agents spend 15 hours devising and Lean-proving a faster shortest-path algorithm, C-HD — ctjlewis · 2026-09-23
- Perplexity's hint-guided self-distillation cuts its agent's tool-call failures by 21.2% — perplexity_ai · 2026-09-23
- Kundaje updates ChromBPNet preprint to refute seq2PRINT's 'better bias correction' claim — anshulkundaje · 2026-09-23
- KostasVisualizations grows to 22 interactive explainers on the math behind computer vision — CSProfKGD · 2026-09-23
- Anshul Kundaje accuses seq2PRINT authors of dodging prior work on bias correction — anshulkundaje · 2026-09-23
- UCLA Releases ACLArena: A Framework for Agent Continual Learning in Multi-Stage Post-Training — UCLA-SCAI · 2026-09-23