Brief Bio
I’m an Assistant Professor in the Department of Computer Science at Cornell University. I received my PhD in Computer Science from UC Berkeley. I work in the field of high-performance computing (HPC) for large-scale computational sciences and lead the ALPS (Algorithms for Large-scale Parallel Systems) Lab. I’m interested in developing algorithms and software infrastructures on parallel machines to speed up data processing without sacrificing programming productivity, and to make high-performance computing more accessible. I’m a big fan of sparse linear algebra and believe in sparse linear algebra as a computational abstraction for tackling large-scale computational challenges.
I received the 2026 IEEE TCHPC Early Career Researchers Award for Excellence in High Performance Computing, 2026 NSF CAREER Award, the 2024 SIAG/Supercomputing Early Career Prize, the 2023 ISSNAF Young Investigator Mario Gerla Award, and the 2020 SIGHPC Computational & Data Science Fellowship.
I’m an Affiliate Faculty in the Applied Math and Computational Sciences Division (Performance and Algorithms Group) at Lawrence Berkeley National Laboratory and a Graduate Field Faculty in the School of Electrical and Computer Engineering, Department of Computational Biology and the Center for Applied Math at Cornell.
I’m also a faculty member of the Computer Systems Laboratory (CSL) at Cornell University.
Publications, Talks, Teaching
For a complete list of publications, talks, and teaching information, please see my CV (I’m fairly good at keeping it up to date) or my Google Scholar account. The PDFs of most of my articles can be found on arXiv.
If you’re interested in course or research talk slides, please feel free to email me. I’ll add them here eventually.
In Spring 2027, I’ll be teaching CS 5220: Applied High-Performance and Parallel Computing (CS 5220 will co-meet with CS 4230).
Recent Updates
Our work, “Sparse Linear Algebra Accelerates Genotype Representation Graph Computation at Biobank Scale,” led by Yifan Li and Qingyao Sun, is now on bioRxiv. It introduces Mikado, which reformulates genotype representation graph (GRG) traversal as a sparse triangular solve built from blocked SpMVs, achieving up to 470× speedup on the core matrix-vector primitive and orders-of-magnitude end-to-end speedup for PCA and BOLT-LMM at biobank scale, including on All of Us. This work is a collaboration with Professor Wei’s Lab and Professor Musharoff’s Lab.
I’m deeply honored to receive the 2026 IEEE Computer Society TCHPC Early Career Researchers Award for Excellence in High Performance Computing.
Two ALPS Lab papers have been accepted to the IA³ workshop at SC’26. “Cheaper by the Batch: Shared Traversal for Genotype Graph Editing,” led by Aaron Li, introduces a shared-traversal method for batched edits on Genotype Representation Graphs. “SpSYRK: Half the Work in Distributed Sparse Matrix Multiplication,” led by Thomas McFarland, reduces both computation and communication in distributed sparse matrix products.
I’m thrilled to serve as Program Co-Chair for ACM ICS 2027, to be held in Atlanta next June during ACM FCRC, together with Prof. Daniel Wong (UC Riverside).
I’m honored to have been invited to give a talk at SC26 in Chicago (November 15–20, 2026), speaking in an invited session on the growing role of advanced computing and AI in biomedical research and healthcare.
Congratulations to Yifan Li (and team!) for receiving the Best Poster Award at ICS26 for his poster “Faster Biobank-Scale Analysis through Sparse Linear Algebra”. More on this work soon!
I have been awarded the NSF CAREER Award to support research on sparse linear algebra as a unifying abstraction for portable, scalable scientific computing. A huge thank you to the ALPS Lab students and collaborators, and to the mentors and colleagues who supported this proposal. Really excited for what’s ahead!
Irene Simó will present her poster, “Low-Precision SpMV and s-step SGD on Processing-in-Memory,” at IPDPS26 in New Orleans, May 25–29, as part of the IPDPS PhD Forum.
Our work “Ocean: Fast Estimation-Based Sparse General Matrix-Matrix Multiplication on GPU” led by Yifan Li has been accepted to ICS 2026. Ocean replaces the costly symbolic pass in GPU SpGEMM with fast HyperLogLog estimation.
Our work introducing “Trident partitioning” led by Julian Bellavita and Lorenzo Pichetti, a hierarchy-aware hybrid 2D–1D decomposition for distributed SpGEMM that minimizes costly inter-node communication by exploiting fast intra-node GPU interconnects on modern heterogeneous supercomputers, has been accepted to ICS 2026!
- Our NSF proposal “ACED: Fast and Scalable Whole Genome Analysis on Emerging Hardware Technologies” has been funded! This project, conducted in collaboration with Professor April Wei’s Lab, will address major computational challenges in population genetics through parallel computation, sparse linear algebra, and new hardware technologies.
Selected Publication & Software
Mikado: Sparse Linear Algebra Accelerates Genotype Representation Graph Computation at Biobank Scale
Our preprint introduces Mikado, which reformulates genotype representation graph (GRG) traversal as a blocked sparse triangular solve. Under a height-based level-set ordering, the GRG adjacency matrix becomes strictly block lower triangular, so the genotype matrix-vector product decomposes into a pipelined wavefront of SpMVs on standard sparse primitives (cuSPARSE, MKL). Mikado achieves up to 470× speedup on the core primitive and orders-of-magnitude end-to-end speedup and cost savings for PCA and BOLT-LMM at biobank scale, including All of Us, while porting across A100, V100, H100, and GH200 with no custom kernel.
Read the Mikado Paper
Mikado GitHub
Ocean: Fast Estimation-Based Sparse General Matrix-Matrix Multiplication on GPU
Ocean (ICS26) redesigned GPU-based SpGEMM by replacing the symbolic pass with lightweight HyperLogLog cardinality estimation, achieving a consistent 1.4×-2.8× speedup over the state-of-the-art across 400+ matrices on A100 and H100. To our knowledge, this is the first use of HyperLogLog to accelerate sparse linear algebra primitives.
Read the Ocean Paper
Ocean GitHub
Trident: Communication-Avoiding SpGEMM via Trident Partitioning on Hierarchical GPU Interconnect
Our ICS 2026 paper introduces trident partitioning for SpGEMM, i.e., a hierarchy-aware 2D-1D hybrid distributed SpGEMM algorithm that uses communication-avoiding techniques and asynchronous communication to exploit the hierarchical and heterogeneous architecture of modern supercomputing interconnect.
Read the Trident Paper
Trident GitHub
Vivaldi: Communcation-Avoiding Large-Scale Linear Algebraic Kernel K-Means
Our IPDPS 2026 paper introduces distributed memory algorithms for scaling Kernel K-means to million-point datasets (one to two orders of magnitude larger than previously practical). Central to our work is the design of partitioning schemes that enable communication-efficient composition of the linear algebra primitives.
Read the Vivaldi Paper
Vivaldi GitHub
Popcorn: Accelerating Kernel K-means on GPU using Sparse Linear Algebra
Our PPoPP 2025 paper introduces a new sparse-matrix formulation of Kernel K-means that enables an efficient, high-performance GPU implementation. Our open-source tool, Popcorn, achieves up to 123.8× speedup over a CPU version and 2.6× over a dense GPU implementation.
Read the Popcorn Paper
Popcorn GitHub
HySortK: High-Performance Sorting-Based K-mer Counting
Our ICPP 2024 paper describes HySortK, a new distributed-memory k-mer counting tool for genomics pipelines. Using a sorting-based approach and a flexible hybrid-parallelism layer, HySortK significantly reduces memory overhead and improves scalability, achieving 2–10× speedup over a GPU baseline on 4–8 nodes and up to 2× speedup over leading CPU tools on 16 nodes, while reducing peak memory usage by approximately 30%.
Read the HySortK Paper
HySortK GitHub
GPU-Accelerated Pangenome Graph Layout
Our SC 24 paper introduces a GPU-optimized layout tool for pangenome graphs. On 24 human whole-chromosome pangenomes, our implementation achieves up to 57.3× speedup over a multithreaded CPU baseline, reducing layout times from hours to minutes while maintaining layout quality.