Abhinav Bhatele
Professor
Researches parallel software and HPC performance, including scalable frameworks for training large neural networks across many GPUs.
Abhinav Bhatele's lab at UMD works on efficient training & inference and distributed, cloud & edge systems. 38 papers since 2023, 6 at top venues, mostly NeurIPS and TVCG. 5 PhD students. 6 alumni.
At a glance
- papers, 2023–2026
- 10/ year
- at T1 venues
- 16%
- PhD students
- 5
- MS students
- 0
- alumni listed
- 6
Papers by yearfrom the lab homepage, 2023–2026
- 7 papers2026 · T1 0 · collecting
- 14 papers2025 · T1 3
- 13 papers2024 · T1 3
- 4 papers2023 · T1 0
T1 venuesOtherVenues: NeurIPS, TVCG, ICCV
Research focuspaper titles that use each topic's words
- Distributed, cloud & edge systems9 papers
- Efficient training & inference7 papers
- performance7 papers
- gpus0 papers
- frameworks0 papers
Paper list
From the lab homepage.
20267 papers
- Proceedings of the IEEE International Parallel & Distributed Processing SymposiumCharacterizing Production GPU Workloads using System-wide Telemetry Data
- Proceedings of the ACM Conference on AI and Agentic SystemsDissecting and Improving Communication Performance in Multi-Node LLM Inference
- IEEE Transactions on Parallel and Distributed SystemsPerformance-Aligned LLMs for Generating Fast HPC Code
- Proceedings of the International Conference on SupercomputingSkew-aware Adaptive All-to-allv Algorithms for Dynamic Deep Learning Workloads
- Proceedings of the ACM/IEEE International Conference for High Performance Computing, Networking, Storage and AnalysisSpCCL: A Sparsity-Aware Collective Communication Library for GPU Platforms
- Proceedings of the IEEE International Parallel & Distributed Processing SymposiumThe Big Send-off: Scalable and Performant Collectives for Deep Learning
- Proceedings of the IEEE International Parallel & Distributed Processing SymposiumThe Case of the Elusive Application Performance on Production GPU Supercomputers
202514 papers
- Proceedings of the Symposium on Experimental AlgorithmsElias-Fano Compression for Space-efficient Rank and Select Structures
- NeurIPS (T1)Gemstones: A Model Suite for Scaling Laws
- Proceedings of the ISC High Performance ConferenceHPC-Coder-v2: Studying Code LLMs Across Low-Resource Parallel Languages
- Proceedings of the First International Workshop on Foundational Large Language Models Advances for HPCLeveraging AI for Productive and Trustworthy HPC Software: Challenges and Research Directions
- Proceedings of the IEEE International Parallel & Distributed Processing SymposiumPandemics in silico: Scaling an Agent-Based Simulation on Realistic Social Contact Networks
- Proceedings of the International Conference on Parallel ProcessingParEval-Repo: A Benchmark Suite for Evaluating LLMs with Repository-level HPC Translation Tasks
- Proceedings of the ACM/IEEE International Conference for High Performance Computing, Networking, Storage and AnalysisPlexus: Taming Billion-edge Graphs with 3D Parallel GNN Training
- NeurIPS (T1)Scaling up Test-Time Compute with Latent Reasoning: A Recurrent Depth Approach
- Nature Scientific ReportsSimulating nationwide coupled disease and fear spread in an agent-based model.
- Proceedings of the International Conference on SupercomputingTaking GPU Programming Models to Task for Performance Portability
- Proceedings of the Second International Workshop on Large Language Models for CodeThe Shortcomings of Code LLMs in Modeling Code Properties
- IEEE Transactions on Parallel and Distributed SystemsVAHRM: Variation-Aware Resource Management in Heterogeneous Supercomputing Systems
202413 papers
- Proceedings of the ACM/IEEE International Conference for High Performance Computing, Networking, Storage and AnalysisA probabilistic approach to selecting build configurations in package managers
- NeurIPS (T1)Be like a Goldfish, Don't Memorize! Mitigating Memorization in Generative LLMs
- Proceedings of the International Symposium on High-Performance Parallel and Distributed ComputingCan large language models write parallel code?
- Proceedings of the ACM/IEEE International Conference for High Performance Computing, Networking, Storage and AnalysisDemocratizing AI: Open-source scalable LLM training on GPU-based supercomputers
- TVCGDesign Concerns for Integrated Scripting and Interactive Visualization in Notebook Environments
- Proceedings of the ISC High Performance ConferenceHPC-Coder: Modeling Parallel Programs using Large Language Models
- Proceedings of the Mining Software Repositories ConferenceLearning to Predict and Improve Build Successes in Package Ecosystems
- NeurIPS (T1)Loki: Low-Rank Keys for Efficient Sparse Attention
- Proceedings of the IEEE International Conference on High Performance ComputingML-based modeling to predict I/O performance on different storage sub-systems
- Proceedings of the IEEE International Parallel & Distributed Processing SymposiumPredicting Cross-Architecture Performance of Parallel Programs
- Proceedings of the Euromicro International Conference on Parallel, Distributed, and Network-Based ProcessingPredicting GPUDirect Benefits for HPC Workloads
- Proceedings of the IEEE International Workshop on Deep Analysis of Data-Driven ApplicationsRelative performance prediction using few-shot learning
20234 papers
- Proceedings of the International Conference on SupercomputingA Hybrid Tensor-Expert-Data Parallelism Approach to Optimize Mixture-of-Experts Training
- Proceedings of the IEEE International Parallel & Distributed Processing SymposiumExploiting sparsity in pruned neural networks to optimize large model training
- Proceedings of the IEEE International Parallel & Distributed Processing SymposiumPorting a computational fluid dynamics code with AMR to large-scale GPU platforms
- TVCGScalable comparative visualization of ensembles of call graphs
Where alumni went6 listed
Apply to UMDFall 2027
Not announced yet
UMD Computer Science PhD
- Round
- Single round
- Opens
- —
- GRE
- See program page
- English
- See program page
- Fee
- See program page
Not yet confirmed for this intake.
Similar labsnearest by research summary
- CNSJTU
Xingda Wei
- Efficient training & inference
- GBEdinburgh
Michael O'Boyle
- Compilers & programming languages
- USUSC
Konstantinos Psounis
- Efficient training & inference
- GBUCL
Ran Ben Basat
- Efficient training & inference
- USMIT
Manya Ghobadi
- Distributed, cloud & edge systems
- USMIT
Tim Kraska
- Large language models
Your notes
Save this lab to keep notes.
Data
Checked against the lab homepage.
- Collected
- Verified
- Papers
- The lab homepage
- Members
- Lab homepage
- Alumni
- Lab homepage
Tiers are venue classifications, not lab rankings.