Gennady Pekhimenko and Nandita Vijaykumar
U of T 海角视频 of Computer Science faculty members and have been inducted into the International Symposium on Computer Architecture (ISCA) Hall of Fame for publishing at least eight papers in the conference, a top-tier venue for computer architecture research.
Pekhimenko and Vijaykumar are both assistant professors in the 海角视频 of Computer and Mathematical Sciences at U of T Scarborough and the tri-campus graduate 海角视频 of Computer Science.
In his research, Pekhimenko focuses on efficient memory systems, systems for machine learning, stream processing, machine learning for systems, compilers, and hardware acceleration. Vijaykumar is interested in the system-level and programming challenges of accelerators, memory systems and more recently, robotics and large-scale machine learning.
A sampling of their recent ISCA papers is listed below:
鈥淣VOverlay: Enabling Efficient and Scalable High-Frequency Snapshotting to NVM鈥 ISCA 2021.
Ziqi Wang, Chul-Hwan Choo, Michael A. Kozuch, Todd C. Mowry, Gennady Pekhimenko, Vivek Seshadri, Dimitrios Skarlatos
This work presents NVOverlay, a scalable and efficient technique for capturing frequent persistent snapshots to NVM (non-volatile memory) such that they can be randomly accessed later.
鈥淓cho: Compiler-based GPU Memory Footprint Reduction for LSTM RNN Training,鈥 ISCA 2020.
Bojian Zheng, Abhishek Tiwari, Nandita Vijaykumar, Gennady Pekhimenko
This work explores significantly reducing the memory footprint of LSTM (long short-term memory) training by selectively and automatically recomputing intermediate values instead of storing them. This enables training significantly larger models and the use of bigger batch sizes for training without loss in accuracy.
鈥淪ysScale: Exploiting Multi-domain Dynamic Voltage and Frequency Scaling for Energy Efficient Mobile Processors,鈥 ISCA 2020.
Jawad Haj-Yahya, Mohammed Alser, Jeremie Kim, A. Giray Yaglikci, Nandita Vijaykumar, Efraim Rotem, and Onur Mutlu
This work proposes a novel power management mechanism that enables significant performance improvements and energy savings in mobile processors via intelligent allocation of the power budget across power domains.