Nathan DeBardeleben

Papers

Total Citations

H-Index

About

Nathan DeBardeleben is a leading researcher in high-performance computing (HPC) reliability, with a focus on fault tolerance and resilience in large-scale systems. His work bridges the gap between hardware faults and software-level mitigation, particularly for emerging machine learning workloads. He is best known for developing **TensorFI**, a configurable fault injector for TensorFlow applications (2018, 63 citations), which enables systematic study of how transient hardware errors affect deep learning models—a critical contribution as ML becomes a "killer app" for next-generation supercomputers. His research has helped define best practices for injecting faults into complex software stacks, influencing both academic studies and industrial reliability testing. Beyond TensorFI, DeBardeleben has made foundational contributions to understanding silent data corruption in HPC, co-authoring widely cited papers on fault injection methodologies and resilience benchmarking. His work has been instrumental in shaping the design of fault-tolerant exascale systems, with cumulative citations reflecting his impact on the HPC and reliability communities. A key figure in the Los Alamos National Laboratory’s HPC resilience group, his research continues to guide how we protect critical scientific simulations and AI models from hardware failures.