Search papers, labs, and topics across Lattice.
Department of Foundation Model, Tongji University
2
0
5
Dynamo achieves a remarkable 5.6% average accuracy boost in visual reasoning tasks without retraining, revolutionizing how VLMs adapt in real-time.
Ditch hard clipping: ratio-variance regularization offers a principled, "soft brake" approach to trust region policy optimization, unlocking substantial gains in sample efficiency and performance, especially for smaller LLMs.
