Shaizeen Aga Email & Phone Number

Santa Clara, Ca, Us

Santa Clara, California, Us

Santa Clara, California, Us

-In this role I drive new and strategic research directions for my team within AMD Research with a broad goal of application-driven design of accelerators and future architectures. -My team focuses on in-depth understanding of applications of business interest such as machine learning (CNNs, BERT, RNNs, recommendation models etc.), graphics (ray tracing), gaming and high-performance computing applications and we use this understanding to guide future accelerators and architectures research. See our IISWC'22 paper for BERT characterization and our Memsys'21 paper for recommendation models optimization.-I led and contributed to several proposals for external funding opportunities along with delivering key presentations and deliverables on these funding contracts. I also served as SRC liaison for AMD.-I continue to collaborate with academia on security research. See details in our submission on Egalitarian ORAM.

Santa Clara, California, Us

-I am a key thought leader in researching architectures for Machine Learning (ML) workload acceleration. I lead a team focused on analyzing ML workloads of importance from algorithmic level to profiling on hardware, and harnessing insights from this analysis to guide future architecture design. See our projections for Transformer model's computation/communication scaling at arxiv'23 for an exemplar analysis. -I was one of the key contributors at AMD to the first-ever GPU-HBM processing-in-memory (PIM) prototype (see press release below on this front). I led the team tasked with identifying PIM ROI for wide swathe of workloads from machine learning (Transformers, recommendation models, and more), HPC (FFTs, wave simulation, and more), graphics (ray tracing). Using insights from said workload analysis, I helped guide a potential performant GPU-HBM PIM architecture along multiple dimensions (in-memory ALU design, memory interface protocol, and more). Further, the analysis my team delivered helped guide executive thinking at AMD and partner organizations. -For PIM workloads analysis, see our iMACAW@DAC 2023 workshop papers on PIM acceleration for ML data preprocessing and Ray Tracing.

Santa Clara, California, Us

-I developed analytical models and performance methodologies for accurate projections of benefits of futuristic architectural features. These projections fed into executive decision making at AMD and partner organizations on key projects.-I co-advised several summer co-ops on interesting architecture projects from architectural characterization of sequence-based neural networks, to ordering primitives for GPUs and more. See publications such as our ISPASS'20 paper for details.

Santa Clara, California, Us

-I studied the benefits of near-data computing for machine learning workloads. See publications such as our MEMSYS'19 paper for details. -Additionally, as part of AMD's PathForward Exascale research program, I studied data movement for machine learning workloads in heterogeneous systems and ways to optimize its effects. I led multiple discussions with US Department of Energy Labs as part of this effort. See our ISCA'23 paper for details.

Ann Arbor, Michigan, Us

I worked on several exciting topics including in-place computing in processor caches, building secure architectures at low cost, realizing intuitive memory models efficiently and improving performance of multi-core runtimes.

Ann Arbor, Michigan, Us

I taught Parallel Computer Architecture; a graduate course on recent advancements in parallel architectures. I co-advised graduate students on several research projects in computer architecture.

Ann Arbor, Michigan, Us

I taught Parallel Computer Architecture; a graduate course on recent advancements in parallel architectures. I co-advised graduate students on several research projects in computer architecture.

San Diego, Ca, Us

I worked here on programming mobile GPUs. I also worked on optimizing ray tracing application which is a heterogeneous benchmark harnessing both CPU and GPU.

Richland, Wa, Us

I worked here on improving the efficiency of Cilk multi-core runtime system using optimistic concurrency.

New York, Ny, Us

My work here primarily involved design and development of a Data Warehouse.

New York, Ny, Us

Santa Clara, Ca, Us

I worked on CUDA, NVIDIA's GPGPU solution. Ported a true motion estimation algorithm onto CUDA platform.

Doctor Of Philosophy - Phd, Computer Engineering

Master’S Degree, Computer Engineering

Btech, Computer Science

Education record