Juan-David Guerrero-Balaguera
Papers
9
Total Citations
75
H-Index
5
About
Juan-David Guerrero-Balaguera is a leading researcher at the intersection of hardware reliability and artificial intelligence, specializing in the dependability of Graphics Processing Units (GPUs) used for Deep Neural Networks (DNNs) and Convolutional Neural Networks (CNNs). His work addresses a critical challenge: ensuring that GPUs—widely deployed in safety-critical domains like automotive, robotics, and healthcare—remain reliable over their expected ten-year lifetimes despite permanent hardware faults. Guerrero-Balaguera pioneered a multi-level framework for fault injection and reliability assessment, enabling systematic evaluation of how GPU permanent faults impact CNN accuracy. His key contributions include exploring the structural resilience of Tensor Core Units (TCUs) under different real-number representations and developing compaction methods for GPU in-field testing. He has also advanced the emerging paradigm of Split Computing for IoT, evaluating how supervised compression and partitioning strategies affect DNN reliability on mobile devices. With over 75 citations across his most-cited works, his research has been published in top venues and is instrumental for designing fault-tolerant AI accelerators. His notable achievements include proposing novel scheduling policies to enhance GPU reliability and developing methods to trade off performance and accuracy in large DNN models, making him a pivotal figure in dependable AI hardware.
Research Focus
Key Achievements
Top Papers
- 1
- 2
- 3Evaluating the Reliability of Supervised Compression for Split Computing8 citations · 2024
- 4A Compaction Method for STLs for GPU in-field test8 citations · 2022
- 5
- 6
- 7
- 8
- 9Enhancing the Reliability of Split Computing Deep Neural Networks2 citations · 2024