Publications

Learning-augmented online algorithm that uses workload predictions and adaptive budget allocation to approach offline utility for private analytics.

Combines efficient syntactic filtering with precise semantic verification for scalable similar-passage search over large text corpora.

Distributed local edge-DP k-core and triangle counting with input-sensitive accuracy, within 3× error for k-core and orders of magnitude lower triangle error.

Approximate aggregate query answering over noisy, incomplete open knowledge.

Exact top-k semantic set search via bipartite-matching bounds and filter-verification, 5.5× faster than prior methods.

Information-theoretic pruning for approximate acyclic schema discovery from relations using multivalued dependencies.

Recursive neural architecture for compositional generalization in mathematical reasoning, beating Transformer/Tree-LSTM baselines.