Grant: Alliance Grants
University of Toronto
$70,000.00
Enabling Private and Efficient Deep Learning
FY 2024-25
Summary
Deep learning (DL) has emerged as a highly effective approach with a wide range of use cases in many important domains. The remarkable performance achieved by DL models in domains like computer vision, natural language processing, and recommendation systems has immensely fueled their popularity and use. Models for ChatGPT, stable diffusion, and vision transformers, all demonstrate the potential of DL models in solving complex tasks. This has led to widespread interest in the generation and innovation of new models and DL innovations for novel and more powerful capabilities in both academia and industry. The advent in generation of these powerful models however has led to numerous concerns about privacy and safety of models and their widespread use. AI accounts for a significant amount of all private user data processing and has applications in safety-critical tasks including navigation, chat interactions with humans, education, and health. Effectively addressing privacy and safety concerns would enable a wider range of safe usage including uses in important domains such as healthcare, personal assistance, mental health, education, and robotics.
An important challenge with DNN inference and training, especially with larger models, e.g., Large Language Models (LLMs), are the significant computational latencies and expenses that they incur. Privacy mechanisms in DNN training and inference can further exacerbate the challenge of long training/inference latencies and the cost and computational requirements, potentially making several privacy mechanisms infeasible. In this proposal, we aim to investigate the additional computational overheads of privacy mechanisms in DNN processing and develop novel techniques to address these overheads, thus making privacy measures practical and inexpensive to support. The goal of this partnership is to thus develop high performance and cost effective frameworks to private DNN inference and training.
Program
Alliance Grants
Recipient
Location
Ontario, Canada
Recipient Type
Academia
Type
Grant
Source
NSERC's Awards Data
As published
Application id
590237-2023
Name nom
Vijaykumar, Nandita N
Department département
Computer and Mathematical Sciences (Scarborough)
Organization id
31
Institution établissement
University of Toronto
Province en
Ontario
Province fr
Ontario
Country en
CANADA
Country fr
CANADA
Fiscal year exercice financier
2024
Competition year année de concours
2023
Award amount
70000
Program id
ALLRP
Program name en
Alliance Grants
Program name fr
Subventions Alliance
Group en
RESEARCH PARTNERSHIPS
Group fr
PARTENARIATS DE RECHERCHE
Committee code
96
Committee name en
RPP Internal Decision Cttee
Committee name fr
Cté de décision interne - PPR
Area of application code
1207
Area of application group en
Advancement of knowledge
Area of application group fr
Avancement des connaissances
Area of application en
Information, computer and communication technologies
Area of application fr
Technologie de l'information, des ordinateurs et des réseaux de communications
Research subject code
2700
Research subject group en
Information technology
Research subject group fr
Technologie de l'information
Research subject en
Information technology
Research subject fr
Technologie de l'information
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
Enabling Private and Efficient Deep Learning
Keywords
compiler optimizations; computer systems; efficient deep learning; high performance deep learning; private deep learning; secure machine learning
Application summary
Deep learning (DL) has emerged as a highly effective approach with a wide range of use cases in many important domains. The remarkable performance achieved by DL models in domains like computer vision, natural language processing, and recommendation systems has immensely fueled their popularity and use. Models for ChatGPT, stable diffusion, and vision transformers, all demonstrate the potential of DL models in solving complex tasks. This has led to widespread interest in the generation and innovation of new models and DL innovations for novel and more powerful capabilities in both academia and industry. The advent in generation of these powerful models however has led to numerous concerns about privacy and safety of models and their widespread use. AI accounts for a significant amount of all private user data processing and has applications in safety-critical tasks including navigation, chat interactions with humans, education, and health. Effectively addressing privacy and safety concerns would enable a wider range of safe usage including uses in important domains such as healthcare, personal assistance, mental health, education, and robotics.
An important challenge with DNN inference and training, especially with larger models, e.g., Large Language Models (LLMs), are the significant computational latencies and expenses that they incur. Privacy mechanisms in DNN training and inference can further exacerbate the challenge of long training/inference latencies and the cost and computational requirements, potentially making several privacy mechanisms infeasible. In this proposal, we aim to investigate the additional computational overheads of privacy mechanisms in DNN processing and develop novel techniques to address these overheads, thus making privacy measures practical and inexpensive to support. The goal of this partnership is to thus develop high performance and cost effective frameworks to private DNN inference and training.