
Geancarlo Abich
Ph.D.
Researcher
INESC-ID Lisbon
Room 308 - Rua Alves Redol, 9
1000-029 Lisbon, Portugal
Phone: +351.213.100.300 (308)
Email:
abich@ieee.org
geancarlo.abich@inesc-id.pt

Bio
Dr Geancarlo Abich is a Brazilian researcher who works on projects for performance modelling, and reliability evaluation of Machine Learning models focused on High-Performance Spacecraft Computing (HPSC). He received his bachelor's degree in computing engineering from UNISC (Universidade de Santa Cruz do Sul), Brazil, 14′, Masters degree in Computing, and Ph.D. degree in Microelectronics from UFRGS (Universidade Federal do Rio Grande do Sul), Brazil, 17′ and 22’, advised by Prof. Ricardo Reis (UFRGS) and co-advised by Prof. Luciano Ost (Loughborough University) where he received the highest grade and his thesis advances become in a Book published by Springer Nature. Dr. Abich's outstanding doctoral thesis earned him the prestigious TTTC's E. J. McCluskey Best Doctoral Thesis Award - Latin America Test Symposium (LATS) 2024. In his earlier work, he contributed to developing NoC-based MPSoC management systems, soft-error reliability evaluation tools, and parallelism exploration of specialized neural network libraries. During his PhD studies, Dr Abich collaborated mainly with Prof. Luciano Ost at Loughborough University and also with companies such as IMPERAS and Arm (University Program). In the last year, he worked as a postdoctoral fellow in extending his thesis to evaluate the performance and power costs of parallelism and fault mitigation in machine learning models aimed at resource-constrained devices. Currently, he holds a postdoctoral position at INESC-ID as a co-investigator on performance modelling strategies over the SYCLOPS project. Dr. Geancarlo Abich's research focuses on achieving efficiency at different levels of abstraction, driving both performance gains and reliability at the software and hardware level by exploring the design space for new technologies to deploy AI at high-performance and reliable computing systems. The research he has been conducting in collaboration with their partners, including universities from Portugal, the United Kingdom, and Brazil, as well as industrial partners like ARM and Imperas, led to the publication of 1 book, 4 journal papers, and 11 peer-reviewed international conferences.
Latest Publication
Power and Performance Costs of Radiation-hardened ML Inference Models Running on Edge Devices
Integrating Machine Learning (ML) inference models into edge computing devices has introduced several challenges related to improving power efficiency, performance, and reliability. As the susceptibility of these models to radiation-induced soft errors is a significant concern, applying lightweight mitigation techniques is key, mostly due to power and memory constraints inherent to edge devices. In this regard, assessing the potential power and performance penalties associated with deploying soft error mitigation techniques on customized ML inference models running in such resource-constrained devices is crucial. This paper, therefore, investigates the performance and power consumption implications of applying software-based mitigation techniques on ML inference models optimized for edge devices. The experiments demonstrate that implementing RAT technique reduced up to 3.2 x the susceptibility to the occurrence of soft errors caused by radiation with low performance and power consumption costs w.r.t. a P-TMR technique.
Geancarlo Abich, Ph.D.
HPCAS - INESC-ID
Room 308 - Rua Alves Redol, 9
1000-029 Lisbon, Portugal
abich@ieee.org
geancarlo.abich@inesc-id.pt
+351.213.100.300 (308)