Director, LG Corporation
3 papers at NeurIPS 2025
Patent examination remains hard for NLP since evaluation needs to consider examiners’ reasoning. PANORAMA captures this with 8,143 U.S. records and full trails (applications, prior art, rejections, allowances) split into stepwise benchmarks.
We present MLRC-Bench, a dynamic benchmark designed to rigorously assess how well language agents address ML research challenges with objective, performance-based evaluations.
Self-supervised models learn robust representation of real images under cropping and resizing, which can be applied to detect AI-generated images without training.
