Search papers, labs, and topics across Lattice.
4
0
7
18
High-quality retrieval fails to guarantee correct reasoning in real-world QA systems, revealing critical vulnerabilities in their performance.
Membership inference attacks can exploit tabular ICL, but TabPATE offers a robust defense that preserves model utility without requiring public data.
WAPO leverages token-level gradient dynamics to stabilize RLVR training, achieving superior performance in reasoning tasks while addressing collapse risks.
VLMs that ace standard chart QA benchmarks can still fail spectacularly when presented with subtly altered, counterfactual versions of the same charts.
