Scholar
Hao Kang
Google Scholar ID: xTG6Nn0AAAAJ
Georgia Institute of Technology
Citations & Impact
Citations
177
H-index
4
i10-index
3
Publications
8
Co-authors
7
list available
Publications
Resume (English only)
Academic Achievements
- Published multiple papers, including:
- - GEAR: An Efficient KV Cache Compression Recipe for Near-Lossless Generative Inference of LLM, which achieves near-lossless compression ratio, 2x speedup, and 2x peak memory saving on inference time.
- - KV Cache Optimizations for Large Language Model Inference (under review for Mlsys2024)
- - Towards Sustainable Learning: Coresets for Data-efficient Deep Learning (ICML2023), which presents a dataset distillation algorithm based on submodular function and batch SGD.
- Additionally, developed THOP: PyTorch-OpCounter, a Python third-party library that counts FLOPs of models.
Research Experience
- Involved in several research projects, including torchanalyse, a model profiling tool based on TVM and Maestro, and Epipe, a research project that reduces activation transfer bandwidth during cloud-based training using compression algorithms.
Education
- Received a B.Eng. in Computer Science from Zhejiang University in 2023; currently a PhD student at Georgia Institute of Technology, advised by Prof. Tushar Krishna; previously worked with Prof. Baharan at UCLA on efficient machine learning from massive datasets; and collaborated with Prof. Song Han at MIT on efficient machine learning on edge devices.
Background
- Interested in efficient machine learning and systems, with experience at the intersection of both fields. Aims to use low-rank approximation and compression algorithms to accelerate machine learning models, especially LLMs. Also designs efficient systems like inference/fine-tuning schedulers to speed up the training/inference process.
Miscellany
- No personal interests mentioned.
