Search papers, labs, and topics across Lattice.
School of Computer Science, Peking University
1
0
2
0
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.
