Jeffrey Kent

AI research & engineering

Research

  1. NeurIPS 2026 · 2nd Set AI

    SANEval: Open-Vocabulary Compositional Benchmarks with Failure-mode Diagnosis

    A benchmark for compositional accuracy in generated visual media.

    PaperNeurIPS

  2. Preprint · 2025, revised 2026

    SHARe-KAN: Post-Training Vector Quantization for Cache-Resident KAN Inference

    Compressing pre-trained Kolmogorov-Arnold networks for inference on memory-constrained hardware, without retraining.

    Paper

  3. VLDB 2020

    PyTorch Distributed: Experiences on Accelerating Data Parallel Training

    Scaling PyTorch training across GPUs while reducing communication overhead.

    Paper