LLM Agents Refactor Software for High-Level Synthesis: A CMU-UCLA Collaboration⭐9

LLM agents from CMU and UCLA automate software refactoring for high-level synthesis, reducing manual HLS effort by 70%.

Contrastive Reflection for Iterative Prompt Optimization⭐9

Contrastive Reflection improves agentic IR prompts by comparing failed vs. successful traces, boosting HotpotQA accuracy from 51.4% to 60.4%.

Recursive Self-Evolving Agents via Held-Out Selection⭐8

RSEA enables LLM agents to self-improve via recursive context evolution without weight updates, using held-out selection to prevent regression and ensure safe t...