Search papers, labs, and topics across Lattice.
2
2
5
16
Choosing the right privacy-preserving ML technique is more than just latency: this benchmark reveals the hidden energy and monetary costs that can make MPC cheaper than FHE in some surprising scenarios.
Listwise preference optimization for diffusion models (Diffusion-LPO) beats pairwise DPO baselines, finally unlocking the potential of richer ranked human feedback.
