Results for “bullet-points”

Benchmarks zero-shot in-context learning on BIG-Bench Hard multiple-choice tasks, comparing self-generated demonstrations against direct prompting and chain-of-thought baselines, and reports accuracy.

Train, evaluate, export, prune, and run inference for PointPillars 3D object detection models from LiDAR point clouds using NVIDIA TAO.

Detect poisoned training data and backdoored models across the ML pipeline using statistical analysis, activation clustering, and spectral signatures.