About

Lawrence Rauchwerger is a Professor in the Department of Computer Science at the University of Illinois at Urbana-Champaign. He previously held the position of Eppright Professor of Computer Science and Engineering at Texas A&M University and was the co-Director of the Parasol Lab. Rauchwerger received an Engineer degree from the Polytechnic Institute Bucharest, a M.S. in Electrical Engineering from Stanford University, and a Ph.D. in Computer Science from the University of Illinois at Urbana-Champaign. His research focuses on parallel and distributed programming environments, compilers, and architectures for parallel and distributed computing. Rauchwerger's approach to auto-parallelization, thread-level speculation, and parallel code development has influenced industrial products at corporations such as IBM, Intel, and Sun. He is recognized as an AAAS Fellow and IEEE Fellow, and has received an NSF CAREER award. Rauchwerger has chaired various IEEE and ACM conferences, including serving as Program Chair of PACT 2016 and PPoPP 2017.

Research topics

• Computer Science
• Algorithm
• Mathematics
• Parallel computing
• Artificial Intelligence
• Computational science
• Mathematical optimization
• Computer hardware
• Physics
• Optics
• Geometry

Selected publications

• Joint Surrogate Learning of Objectives, Constraints, and Sensitivities for Efficient Multi-objective Optimization of Neural Dynamical Systems

  ArXiv.org · 2026-03-22

  articleOpen access

  Biophysical neural system simulations are among the most computationally demanding scientific applications, and their optimization requires navigating high-dimensional parameter spaces under numerous constraints that impose a binary feasible/infeasible partition with no gradient signal to guide the search. Here, we introduce DMOSOPT, a scalable optimization framework that leverages a unified, jointly learned surrogate model to capture the interplay between objectives, constraints, and parameter sensitivities. By learning a smooth approximation of both the objective landscape and the feasibility boundary, the joint surrogate provides a unified gradient that simultaneously steers the search toward improved objective values and greater constraint satisfaction, while its partial derivatives yield per-parameter sensitivity estimates that enable more targeted exploration. We validate the framework from single-cell dynamics to population-level network activity, spanning incremental stages of a neural circuit modeling workflow, and demonstrate efficient, effective optimization of highly constrained problems at supercomputing scale with substantially fewer problem evaluations. While motivated by and demonstrated in the context of computational neuroscience, the framework is general and applicable to constrained multi-objective optimization problems across scientific and engineering domains.

  PublisherOA PDF

• livn: A testbed for learning to interact with in vitro neural networks

  bioRxiv (Cold Spring Harbor Laboratory) · 2025-12-19

  articleOpen accessSenior author

  A bstract The investigation of cultured biological neural networks is a critical frontier in neuroscience with profound implications for the advancement of brain-machine interfaces, treatments of neurological diseases, and fundamental insights into neural computation and cognition. Advances in induced pluripotent stem cell (iPSC) technology and machine learning are converging to enable novel approaches to interrogating neural circuits in vitro . However, progress in this emerging field is hampered by the technical challenges and resource-intensive nature of acquiring datasets suitable for machine learning. Experimental recordings typically do not allow for interactive learning and generally lack ground-truth information that would enable rigorous algorithm development and validation. To overcome the limitations of experimental setups, simulated biophysical models of neurons and neuronal networks can serve as crucial accelerators. They allow for more controlled and systematic exploration than currently possible with living cultures and serve as interpretable intermediaries between abstract computational theory and complex biological reality. Using this approach, we introduce livn : an open source interactive simulation environment for learning to control in vitro neural networks. livn generates synthetic neural data with ground truth at scale, enabling the development and testing of ML models in interactive settings that mimic experimental platforms. We describe benchmark tasks that challenge ML models to exploit simulated neural dynamics and release generated synthetic datasets that mimic in vitro systems. By providing an open, extensible platform for developing and benchmarking machine learning models, livn aims to accelerate progress in both ML-driven understanding and engineering of in vitro neural systems and fundamental understanding of computation in biological neural networks.

  PublisherDOI

• Learning to Harness In-Vitro Biological Neural Networks

  Lecture notes in computer science · 2025-01-01

  book-chapterSenior author
  PublisherDOI

• Automatic Task Parallelization of Dataflow Graphs in ML/DL Models

  2024-05-27

  articleSenior author

  Several methods exist today to accelerate Machine Learning(ML) or Deep-Learning(DL) model performance for training and inference. However, modern techniques that rely on various graph and operator parallelism methodologies rely on search space optimizations which are costly in terms of power and hardware usage. Especially in the case of inference, when the batch size is 1 and execution is on CPUs, or for power-constrained edge devices, current techniques can become costly, complicated or inapplicable. To ameliorate this, we present a Critical-Path-based Linear Clustering approach to exploit inherent parallel paths in ML dataflow graphs. Our task parallelization approach further optimizes the structure of graphs via cloning and prunes them via constant propagation and dead-code elimination. Contrary to other work, we generate readable and executable parallel Pytorch+Python code from input ML models in ONNX format via a new tool that we have built called Ramiel. This allows us to benefit from other downstream acceleration techniques like intra-op parallelism and potentially pipeline parallelism. Our preliminary results on several ML graphs demonstrate up to 1.9× speedup over serial execution and outperform some of the current mechanisms in both compile and runtimes. Lastly, our methods are lightweight and fast enough so that they can be used effectively for power and resource-constrained devices, while still enabling downstream optimizations.

  PublisherDOI

• Automatic Task Parallelization of Dataflow Graphs in ML/DL models

  arXiv (Cornell University) · 2023-08-22

  preprintOpen accessSenior author

  Several methods exist today to accelerate Machine Learning(ML) or Deep-Learning(DL) model performance for training and inference. However, modern techniques that rely on various graph and operator parallelism methodologies rely on search space optimizations which are costly in terms of power and hardware usage. Especially in the case of inference, when the batch size is 1 and execution is on CPUs or for power-constrained edge devices, current techniques can become costly, complicated or inapplicable. To ameliorate this, we present a Critical-Path-based Linear Clustering approach to exploit inherent parallel paths in ML dataflow graphs. Our task parallelization approach further optimizes the structure of graphs via cloning and prunes them via constant propagation and dead-code elimination. Contrary to other work, we generate readable and executable parallel Pytorch+Python code from input ML models in ONNX format via a new tool that we have built called {\bf Ramiel}. This allows us to benefit from other downstream acceleration techniques like intra-op parallelism and potentially pipeline parallelism. Our preliminary results on several ML graphs demonstrate up to 1.9$\times$ speedup over serial execution and outperform some of the current mechanisms in both compile and runtimes. Lastly, our methods are lightweight and fast enough so that they can be used effectively for power and resource-constrained devices, while still enabling downstream optimizations.

  PublisherOA PDFDOI

• Accelerating SARS-CoV-2 low frequency variant calling on ultra deep sequencing datasets

  arXiv (Cornell University) · 2021-05-07

  preprintOpen access

  With recent advances in sequencing technology it has become affordable and practical to sequence genomes to very high depth-of-coverage, allowing researchers to discover low-frequency variants in the genome. However, due to the errors in sequencing it is an active area of research to develop algorithms that can separate noise from the true variants. LoFreq is a state of the art algorithm for low-frequency variant detection but has a relatively long runtime compared to other tools. In addition to this, the interface for running in parallel could be simplified, allowing for multithreading as well as distributing jobs to a cluster. In this work we describe some specific contributions to LoFreq that remedy these issues.

  PublisherOA PDFDOI

• Accelerating Fourier and Number Theoretic Transforms using Tensor Cores and Warp Shuffles

  2021 · 28 citations

  • Computer Science
  • Computer Science
  • Parallel computing

  The discrete Fourier transform (DFT) and its specialized case, the number theoretic transform (NTT), are two important mathematical tools having applications in several areas of science and engineering. However, despite their usefulness and utility, their adoption continues to be a challenge as computing the DFT of a signal can be a time-consuming and expensive operation. To speed things up, fast Fourier transform (FFT) algorithms, which are reduced-complexity formulations for computing the DFT of a sequence, have been proposed and implemented for traditional processors and their corresponding instruction sets. With the rise of GPUs, NVIDIA introduced its own FFT computation library called cuFFT, which leverages the power of GPUs to compute the DFT. However, as this paper demonstrates, there is a lot of room for improvement to accelerate the FFT and NTT algorithms on modern GPUs by utilizing specialized operations and architectural advancements. In particular, we present four major types of optimizations that leverage tensor cores and the warp-shuffle instruction. Through extensive evaluations, we show that our approach consistently outperforms existing GPU-based implementations with a speedup of up to 4× for NTT and a speed of up to 1.5× for FFT.

  PublisherDOI

• FFT blitz

  2021-02-17 · 9 citations

  articleSenior author

  The fast Fourier Transform (FFT), a reduced-complexity formulation of the Discrete Fourier Transform (DFT), is an important tool in many areas of science and engineering. FFTW is a well-known package that follows this approach and is currently one of the fastest available implementations of the FFT. NVIDIA introduced its version of FFTW called cuFFT that achieves high performance on the GPUs. In this work we present a novel way to map the FFT algorithm on the newly introduced Tensor Cores by adapting the the Cooley-Tukey recursive FFT algorithm. We present four major types of optimizations that enhance the performance of our approach for varying FFT sizes and show that the approach consistently outperforms cuFFT with a speedup of about 15% to 250% on average.

  PublisherDOI

• Accelerating SARS-CoV-2 low frequency variant calling on ultra deep\n sequencing datasets

  arXiv (Cornell University) · 2021-05-07

  preprintOpen access

  With recent advances in sequencing technology it has become affordable and\npractical to sequence genomes to very high depth-of-coverage, allowing\nresearchers to discover low-frequency variants in the genome. However, due to\nthe errors in sequencing it is an active area of research to develop algorithms\nthat can separate noise from the true variants. LoFreq is a state of the art\nalgorithm for low-frequency variant detection but has a relatively long runtime\ncompared to other tools. In addition to this, the interface for running in\nparallel could be simplified, allowing for multithreading as well as\ndistributing jobs to a cluster. In this work we describe some specific\ncontributions to LoFreq that remedy these issues.\n

  PublisherOA PDFDOI

• Accelerating SARS-CoV-2 low frequency variant calling on ultra deep sequencing datasets

  2021-06-01 · 7 citations

  preprintOpen access

  With recent advances in sequencing technology it has become affordable and practical to sequence genomes to very high depth-of-coverage, allowing researchers to discover low-frequency variants in the genome. However, due to the errors in sequencing it is an active area of research to develop algorithms that can separate noise from the true variants. LoFreq is a state of the art algorithm for low-frequency variant detection but has a relatively long runtime compared to other tools. In addition to this, the interface for running in parallel could be simplified, allowing for multithreading as well as distributing jobs to a cluster. In this work we describe some specific contributions to LoFreq that remedy these issues.

  PublisherOA PDFDOI

Recent grants

• A Compositional Approach to Scalable Parallel Software

  NSF · $1.3M · 2008–2015

• ITR/NGS: A Software Infrastructure for Computational Biology and Physics

  NSF · $428k · 2003–2008

• Collaborative Research: CRS--AES: SoftCheck: Compiler and Run-Time Technology for Efficient Fault Detection and Correction in Low nm-Scale Multicore Chips

  NSF · $120k · 2006–2009

• Collaborative Research: Next Generation Compilers for Emerging Multicore Systems

  NSF · $544k · 2007–2014

• ITR/SY: SmartApps: An Application Centric Approach to Scientific Computing

  NSF · $500k · 2001–2007

Frequent coauthors

• David A. Bader

  162 shared

• Guojing Cong

  Oak Ridge National Laboratory

  156 shared

• Srinivas Aluru

  150 shared

• Josep Torrellas

  University of Illinois Urbana-Champaign

  96 shared

• Felix Wolf

  Technical University of Darmstadt

  82 shared

• Matthias Müller

  81 shared

• Paula E. Stretz

  Virginia Tech

  81 shared

• Sandhya Dwarkadas

  81 shared

Awards & honors

• AAAS Fellow
• IEEE Fellow
• NSF CAREER award