U. Ogras

Achieving a staggering reduction in hardware footprint while maintaining near-perfect accuracy, Uni-SFU redefines the efficiency of activation function implementations in neural networks.

Miao Sun, Yucheng Huang, Mingcong Cao +3

Architecture Design (Transformers, SSMs, MoE)Distributed Systems & Hardware Inference & Quantization

Alish Kanani +2Aug 12, 2026·also Affiliation:

APEX: Adaptive Expert Prefetching for Memory-Efficient Edge MoE Inference

APEX achieves over 99% overlap accuracy in expert prefetching, slashing per-token latency by up to 26% while enhancing energy efficiency for edge MoE inference.

Alish Kanani, Layan Badawi, U. Ogras

Architecture Design (Transformers, SSMs, MoE)Inference & Quantization

Mar 16, 2026

Mar 16, 2026·also Affiliation:

DUET: Disaggregated Hybrid Mamba-Transformer LLMs with Prefill and Decode-Specific Packages

Hybrid Mamba-Transformer models can get 4x faster time to first token and 1.4x higher throughput by disaggregating prefill and decode phases onto specialized accelerator packages.

Alish Kanani, Sang-Won Lee, Sangwan Lee +5

Architecture Design (Transformers, SSMs, MoE)Distributed Systems & Hardware Inference & Quantization

Mar 16, 2026·also Affiliation:

LEXI: Lossless Exponent Coding for Efficient Inter-Chiplet Communication in Hybrid LLMs

LLMs can run up to 35% faster on chiplet architectures thanks to a new lossless exponent compression technique that slashes inter-chiplet communication overhead.

Alish Kanani, Kaushik Shroff, U. Ogras +1

Architecture Design (Transformers, SSMs, MoE)Distributed Systems & Hardware Inference & Quantization

U. Ogras

Publication activitypapers/week, last 8 weeks

Research focus

Frequent co-authors

Papers (5)

Search

U. Ogras

Publication activitypapers/week, last 8 weeks

Research focus

Frequent co-authors

Papers (5)