Search papers, labs, and topics across Lattice.
4
0
6
Multi-way graph partitioning no longer degrades polynomially with the number of clusters, establishing an exponential improvement to an $O(\log^5 k \sqrt{\lambda_k})$ higher-order Cheeger bound.
Fragmented defenses against LLM backdoor attacks stem from fundamental differences in how clean and dirty-label attacks manipulate model features.
The mapping of non-maximal probabilities to GMM components significantly influences the performance of S-JEPA encoders, revealing that structure matters as much as values in representation learning.
LLMs can be backdoored to "think well but answer wrong," even while generating seemingly correct reasoning traces, making attacks far harder to detect.
