Search papers, labs, and topics across Lattice.
Institute of Software, Chinese Academy of Sciences
3
0
6
4
Ditch the deep stacks: a dual-stream architecture for learned data compression unlocks state-of-the-art compression ratios and throughput while slashing latency and memory usage.
RL agents can learn to write stronger formal specifications by using automatically generated negative tests as a reward signal for completeness, outperforming standard verification-based rewards.
Stop blindly trusting LLM-generated tests: ACES reveals which tests actually distinguish good code from bad, leading to better code evaluation without needing ground truth labels.
