Hi, I am Zeyu Liu
EE PhD StudentSkills
Work
SGDKV: Summarization Guided KV Cache Compression
LAWCAT: Efficient Distillation from Quadratic to Linear Attention with Convolution across Tokens for Long Context Modeling
AFLoRA: Adaptive Freezing of Low Rank Adaptation in Parameter Efficient Fine-Tuning of Large Models
LMUFormer: Low Complexity Yet Powerful Spiking Model With Legendre Memory Units
About Me
I am a third-year PhD student at EESSC lab in the Viterbi School of Engineering of the University of Southern California advised by Prof. Peter Beerel. My research interests lie in the field of Efficient Artificial Intelligence Systems. Currently, my recent works involve efficient fine-tuning & inference and subquadratic attention architectures.