Canada Spends Logo

Explorez notre analyse du budget 2026

Grant: Discovery Grants Program - Individual

University of Waterloo

$53,000.00

Hardware Architectures for High-Performance and Energy-Efficient Machine Learning in Edge Computing Applications

EF 2024-25

Résumé
Recent advances in our understanding of machine learning (ML) have led to breakthroughs in many applications, including smart vehicles, cybersecurity, communication systems, and medicine. Problems that have been addressed often rely on deep ML models such as convolutional neural networks and generative models, but more traditional approaches such as support vector machines, random forests, and clustering algorithms are also routinely used. ML algorithms can be extremely compute-intensive and often must be accelerated using graphics processing units (GPUs), quite possibly run using cloud services, especially for the training phase. This can be problematic for many reasons. Available hardware may not be powerful enough to run the classes of ML algorithms required to tackle specific problems. Sensitive data may need to be transferred and processed off-site, which may not always be permissible. Finally, applications in IoT and remote sensing may need to communicate raw data under strict energy constraints. The vision that drives the proposed research program is to bring these sophisticated ML algorithms closer to the computing edge, in other words, to enable ML computations including training to happen at the sensors or at the site of interpretation. To do this, the hardware supporting ML algorithms must be compact and consume minimal energy. The proposed program will investigate inter-related circuits and systems approaches to achieve this, and will be driven by knowledge of application domains, and the impact of algorithm selection and implementation on hardware design considerations. This program will: (a) develop ultra-compact and low-power implementations of traditional ML models such as random forests and clustering algorithms; (b) explore hardware implementations of ML models based on stochastic computation and approximate logic; (c) explore efficient custom hardware realizations of convolutional neural network and generative models that include in-hardware training; (d) and explore approaches for hardware to be retrained online during routine operation as new data becomes available, and hardware-based approaches to support federated learning. Although based on field-programmable gate arrays (FPGAs), the hardware designs will be technology-independent and portable, to enable a path towards the design of application-specific integrated circuits. Although our approach to hardware design is application-independent, the program will focus on the development of hardware-enabled solutions for sophisticated deep learning and other algorithms to run at the site of medical imaging. This program will train 10 HQP ranging from the undergraduate to doctoral levels. HQP will gain necessary skills - from data processing and ML algorithms to register-transfer-level hardware design - to translate application domain knowledge into energy-efficient and compact hardware.
Programme
Discovery Grants Program - Individual
Emplacement
Ontario, Canada
Type de bénéficiaire
Academia
Type
Grant
Source
NSERC's Awards Data

Tel que publié

Application id
RGPIN-2024-04393
Name nom
Gaudet, Vincent
Department département
Electrical and Computer Engineering
Organization id
33
Institution établissement
University of Waterloo
Province en
Ontario
Province fr
Ontario
Country en
CANADA
Country fr
CANADA
Fiscal year exercice financier
2024
Competition year année de concours
2024
Award amount
53000
Program id
RGPIN
Program name en
Discovery Grants Program - Individual
Program name fr
Programme de subventions à la découverte - individuelles
Group en
DISCOVERY RESEARCH
Group fr
RECHERCHE AXÉE SUR LA DÉCOUVERTE
Committee code
1510
Committee name en
Electrical and Computer Engineering
Committee name fr
Génie électrique et informatique
Area of application code
1102
Area of application group en
Health, education and social services
Area of application group fr
Santé publique, éducation et services sociaux
Area of application en
Biomedical engineering
Area of application fr
Génie biomédical
Research subject code
2519
Research subject group en
Electrical and electronic engineering
Research subject group fr
Génie électrique et électronique
Research subject en
Microelectronics
Research subject fr
Microélectronique
Field of research list names en
Not available
Field of research list names fr
Non disponible
Socioeconomic objective list names en
Not available
Socioeconomic objective list names fr
Non disponible
Application title
Hardware Architectures for High-Performance and Energy-Efficient Machine Learning in Edge Computing Applications
Keywords
microelectronic circuits; field-programmable gate arrays; low-power design; machine learning hardware; neural networks; edge computing; medical imaging; very large scale integration; stochastic computing; approximate logic
Application summary
Recent advances in our understanding of machine learning (ML) have led to breakthroughs in many applications, including smart vehicles, cybersecurity, communication systems, and medicine. Problems that have been addressed often rely on deep ML models such as convolutional neural networks and generative models, but more traditional approaches such as support vector machines, random forests, and clustering algorithms are also routinely used. ML algorithms can be extremely compute-intensive and often must be accelerated using graphics processing units (GPUs), quite possibly run using cloud services, especially for the training phase. This can be problematic for many reasons. Available hardware may not be powerful enough to run the classes of ML algorithms required to tackle specific problems. Sensitive data may need to be transferred and processed off-site, which may not always be permissible. Finally, applications in IoT and remote sensing may need to communicate raw data under strict energy constraints. The vision that drives the proposed research program is to bring these sophisticated ML algorithms closer to the computing edge, in other words, to enable ML computations including training to happen at the sensors or at the site of interpretation. To do this, the hardware supporting ML algorithms must be compact and consume minimal energy. The proposed program will investigate inter-related circuits and systems approaches to achieve this, and will be driven by knowledge of application domains, and the impact of algorithm selection and implementation on hardware design considerations. This program will: (a) develop ultra-compact and low-power implementations of traditional ML models such as random forests and clustering algorithms; (b) explore hardware implementations of ML models based on stochastic computation and approximate logic; (c) explore efficient custom hardware realizations of convolutional neural network and generative models that include in-hardware training; (d) and explore approaches for hardware to be retrained online during routine operation as new data becomes available, and hardware-based approaches to support federated learning. Although based on field-programmable gate arrays (FPGAs), the hardware designs will be technology-independent and portable, to enable a path towards the design of application-specific integrated circuits. Although our approach to hardware design is application-independent, the program will focus on the development of hardware-enabled solutions for sophisticated deep learning and other algorithms to run at the site of medical imaging. This program will train 10 HQP ranging from the undergraduate to doctoral levels. HQP will gain necessary skills - from data processing and ML algorithms to register-transfer-level hardware design - to translate application domain knowledge into energy-efficient and compact hardware.