Vu Le

PhD Student · UMass Amherst · Berkeley Lab

I’m a third-year Computer Science PhD student at the University of Massachusetts Amherst (UMass), working with Prof. VP Nguyen and Prof. Deepak Ganesan. I’m also a research affiliate at Berkeley Lab, hosted by Dr. Yilun Xu, where I work on computer architecture and heterogeneous computing for real-time qubit readout under a sub-microsecond latency budget.

My research focuses on computer architecture and hardware acceleration for performance-critical AI. I study how computation, data movement, dependencies, and execution overhead shape performance, identify bottlenecks, and develop analytical performance models and hardware-aware optimizations—including kernel optimization, tiling, dataflow, spatial mapping, operator fusion, and scheduling—to minimize end-to-end latency under hardware resource constraints. I evaluate these techniques on AMD spatial AI accelerators (AI Engine and NPU), with applications in transformers and vision-language-action models. My recent work accelerates SmolVLA through analytical GEMM mapping, attention scheduling, and execution optimizations, achieving substantial end-to-end latency reductions over existing AMD implementations (see projects below). I also contribute high-performance kernels and compiler improvements to AMD’s open-source MLIR-AIR compiler. My earlier work includes edge GPU and FPGA-based AI acceleration.

I am seeking research or industry internships in the US for Summer/Fall 2027 in ML performance engineering and hardware acceleration, ML compilers and kernels, and hardware–software co-design, particularly low-latency inference for physical AI across NPUs, FPGAs, and GPUs. Feel free to reach out!

Get in touch

Personal: vule20.cs AT gmail [DOT] com
UMass: vdle AT umass [DOT] edu

CV (PDF)

CV Scholar GitHub LinkedIn Twitter / X

Lederle Graduate Research Center
740 North Pleasant Street, Amherst, MA

Active Research

Hard-Real-Time Inference for Physical AI on Heterogeneous Platforms

Experiments · method validation

Building a hardware-grounded latency model that predicts latency for arbitrary matrix sizes and sequence lengths, and selects tilings and placements that keep end-to-end latency low under a latency and resource budget. Validating on SmolVLA, a vision-language-action model, running entirely on the AMD XDNA2 NPU. Hardware-aware GEMM and FlashAttention kernels, along with compiler fixes and support such as an A-stationary GEMM lowering, have landed upstream in AMD’s MLIR-AIR compiler. With hardware-aware tiling and 2D attention mapping, vision-stage latency drops 53.0% and end-to-end latency 43.0% over AMD’s original MLIR implementation, and 42.9% and 55.3%, respectively, over AMD’s Ryzen AI Software. End-to-end latency is also 13.5% lower than FastFlowLM. See public benchmarks here; ongoing work extends these gains further

Heterogeneous computing pipeline for quantum state classification

Hardware-Aware Recurrent Inference for Sub-Microsecond Qubit Readout

Under review

Vu Le, Yilun Xu, Neel Vora, Gang Huang, Deepak Ganesan, Phuc Nguyen

arXiv

We develop a heterogeneous LSTM accelerator on AMD Versal(VCK190) for real-time superconducting qubit-state discrimination. By partitioning computation at the recurrent dependency boundary—parallel input projections on AI Engines (AIE) and stateful updates in programmable logic (PL)—our design avoids costly recurrent-state spills and preserves full-resolution readout signals under a 1 µs latency budget. On a qubit affected by mid-readout state transitions, our approach achieves 96.59% accuracy, slightly outperforming a full-resolution feed-forward network (96.37%), while using 403× fewer parameters and ~103× fewer DSPs. It reduces classification error by 68.2% over GMM, 43.6% over QubiCML, and 16.82% over MF-NN. Our heterogeneous mapping also reduces exposed tail latency by 60.3%versus AIE-only execution and DSP58 usage by 31.1× versus PL-only execution, using just 4 of 400 AIE tiles.

Selected Research

Representative papers on systems performance and efficient inference across applications. My current focus on heterogeneous acceleration for physical AI and quantum readout is under Active Research above. For a full list of publications, please visit my Scholar

2025

Detection and Tracking of Drone Swarms using LiDAR

Tasnim Azad Abir, Vu Le, Endrowednes Kuantama, Pranjol Sen Gupta, Austin Copley, Judith Dawes, Mohammad Islam, Richard Han, Phuc Nguyen

ACM MobiSys 2025 · A* conference

LiSWARM is a low-cost LiDAR system for accurate 3D tracking and recognition of drones in large swarms. Using point cloud processing, clustering, and neural networks, it achieves up to 98% accuracy and scales to 15,000 drones—enabling applications in airspace security, drone shows, and sensitive area monitoring.

paper

2024

MagicStream: Bandwidth-conserving Immersive Telepresence via Semantic Communication

Ruizhi Cheng, Nan Wu, Vu Le, Eugene Chai, Matteo Varvello, Bo Han

ACM SenSys 2024 · A* conference

MagicStream, a first-of-its-kind semantic-driven immersive telepresence system that effectively extracts and delivers compact semantic details of captured 3D representation of users, instead of traditional bit-by-bit communication of raw content.

Fast and Interpretable Face Identification using Vision Transformers

2024

Fast and Interpretable Face Identification for Out-Of-Distribution Data Using Vision Transformers

Hai Phan, Cindy Le, Vu Le, Yihui He, Anh Totti Nguyen

CVF/WACV 2024 · A conference

Using vision transformers for out-of-distribution data face identification, runs twice faster while achieving comparable performance with the state of the art DeepFace-EMD model.

paper code poster presentation

News

09/2026

My high performance AI kernels are merged into AMD's MLIR-AIR compiler, boosting performance for VLAs and LLMs inference on AMD XDNA2 NPU.

04/2026

Presented research on real-time qubit readout to LBNL and AMD in Berkeley, CA.

06/2025

Presented “Opportunities in Computer Systems Research for Quantum Computing” at ACM QSys 2025 (in conjunction with ACM MobiSys 2025).

03/2025

One paper accepted at ACM MobiSys 2025.

01/2025

I received the James Kurose Scholarship in Computer Science.

12/2024

I officially become a research affiliate with Berkeley Lab.

09/2024

My new academic website with the vule.us domain is live now.

09/2024

One paper accepted at ACM SenSys 2024.

09/2024

I joined University of Massachusetts Amherst, USA as a PhD student.

04/2024

I received some CS PhD offers in the US.

10/2023

One paper accepted at IEEE/CVF WACV.

Miscellanea

Outside of work, I'm an avid adventurer — I love road trips and have driven across the US to explore national parks and trails firsthand. Most of my hikes and adventures (Grand Canyon, Zion, Death Valley, Horseshoe Bend, Joshua Tree, the White Mountains in New Hampshire) were made possible by hitting the road. I also really enjoy lifting at the gym, jogging, brewing coffee, planting flowers, and skiing. I shoot with a Sony A6400 and a collection of Sigma and Sony lenses. Check out my photo gallery.