I am an Assistant Professor in the College of Information and Computer Sciences (CICS) at the University of Massachusetts Amherst, the flagship campus of the UMass system. I received my Ph.D. in Electrical Engineering from North Carolina State University in 2020. I am a member of the Programming Language and Systems at Massachusetts (PLASMA) lab at UMass. My research lies in machine learning systems, with an emphasis on improving the speed, scalability, and reliability of Machine Learning through innovations in algorithms and programming systems (e.g., compilers, runtime).
Modern machine learning, especially deep learning, has made remarkable progress. However, its effective adoption faces a fundamental question: How can we create models that efficiently deliver reliable predictions to meet the requirements of diverse applications running on various systems? In response to this question, my group focuses on addressing the core challenges related to (1) reducing the costs of machine learning model development and (2) bringing deep learning applications to resource-constrained edge environments. The approaches we pursue draw insights from the unique properties of machine learning workloads, such as their inherent accuracy-efficient trade-offs, as well as system design principles such as composability, pipelining, and locality awareness. Ultimately, our goal is to help democratize machine learning by transforming it into an accessible commodity technology that can be applied to a wide range of real-world scenarios.
[Sept. 2023]: Thanks for the support of NSF to our project Memory-Driven Full-Stack Collaboration for Embedded Systems. With collaborators, we will bring the power of deep learning to resource-constrained embedded systems!
[Aug. 2023]: Thanks for the support of NSF to our project Deep Learning on Anomaly Detection for Human Dynamics and Hazard Response. With collaborators, we will work on graph machine learning for anomaly detection.
[Aug. 2023]: Congratulations to Juelin and Sandeep for their work on “Accelerating Subgraph Enumeration Using Auxiliary Graphs” accepted to PACT’23.
[May. 2023]: Our work on “Flash: Concept Drift Adaptation in Federated Learning” is accepted to ICML’23. It proposes a novel adaptive optimizer that simultanuously addresses both data heterogeneity and the concept drift issues in federated learning.
[May. 2023]: Our work on “Automatically marginalized MCMC in probabilistic programming” is accepted to ICML’23. It proposes automatic marginalization to make sampling process using Hamiltonian Monte Carlo more efficient.
[May. 2023]: Our work on “NUMAlloc: A Faster NUMA Memory Allocator” is accepted to ISMM’23.
[May. 2023]: Our work on “GSplit: Scaling Graph Neural Network Training on Large Graphs via Split-Parallelism” is on Arxiv.
[Apr. 2023]: Our work on Re-thinking computation offload for efficient inference on IoT devices with duty-cycled radios is accepted to MobiCom’23.
[Oct. 2022]: I’m excited to share that we have received an Amazon Research Award for our proposal “Groot: A GPU-Resident System for Efficient Graph Machine Learning” at UMass Amherst. Learn more about the program on the website.
[Sept. 2022]: Our work on AutoMTL: A Programming Framework for Automating Efficient Multi-Task Learning is accepted to NeurIPS’22. Congratulations to Lijun. The project is open-sourced
[Sept. 2022]: Thanks for the support of NSF to our project Transparently Scaling Graph Neural Network Training to Large-Scale Models and Graphs.
[Jul. 2022]: Our work on Fine-Grained Personalized Federated Learning Through Dynamic Routing is accepted to CrossFL’2022 Workshop @MLSys. Congratulations to Kunjal.
[Jul. 2022]: Our work on Improving Subgraph Representation Learning via Multi-View Augmentation is accepted to AI4Science’22 Workshop @ICML.
[May. 2022]: Welcome a new PhD student Qizheng Yang to join our lab this summer.
[Mar. 2022]: Thanks for the support of NVIDIA Academic Hardware Grant Program to the project “Multitasking-Centric Optimization for Deep Learning Applications”.
[Mar. 2022]: Our paper “Rethinking Hard-Parameter Sharing in Multi-Domain Learning” is accepted to ICME’22. Congratulations to Lijun.
[Mar. 2022]: Our paper “Enabling Near Real-Time NLU-Driven Natural Language Programming through Dynamic Grammar Graph-Based Translation” is accepted to CGO’22.
[Mar. 2022]: Our paper “COMET: A Novel Memory-Efficient Deep Learning Training Framework by Using Error-Bounded Lossy Compression” is accepted to VLDB’22.
[Oct. 2021]: Our paper “FreeLunch: Compression-based GPU Memory Management for Convolutional Neural Networks” is accepted to MCHPC’21 Workshop, in conjunction with SC’21.
[Oct. 2021]: Our paper “Recurrent Neural Networks Meet Context-Free Grammar: Two Birds with One Stone” is accepted to ICDM’21.
[June 2021]: Our paper “Scalable Graph Neural Network Training: The Case for Sampling” has appeared in the ACM SIGOPS Operating Systems Review.
[June 2021]: Our paper CoCoPIE is accepted to CACM’21.
[June 2021]: Our paper NumaPerf is accepted to ICS’21.
[May 2021]: I have received an Adobe Research Collaboration Grant on developing resource-efficient deep multi-task learning solutions.
- NCSU Electrical and Computer Engineering Outstanding Dissertation Award, 2020
- IBM PhD Fellowship, 2015-2018