Samet Oymak

Zero-shot in-context inference trained strictly on synthetic causal graphs can outperform fully tuned AutoML pipelines across diverse real-world tabular benchmarks.

LLM-guided program discovery gets a serious boost: evolving programs across related tasks isn't just faster, it also beats overfitting, especially when data is scarce.

Attention bottlenecks in long-context decoding? SANTA slashes memory bandwidth demands by stochastically sampling value vectors, achieving 1.5x speedups without sacrificing accuracy.

Transformers can directly solve quadratic programs and leverage covariance matrices for superior decision-making, outperforming traditional "predict-then-optimize" methods in portfolio construction.