Grant: Discovery Grants Program - Individual
The University of British Columbia
$41,000.00
Probabilistic Models for Complex and Large-Scale Scientific Discovery
FY 2024-25
Summary
Predictive machine learning is revolutionizing the scientific process. High-throughput experimental methods, which are becoming the norm in fields ranging from astronomy to solid-state physics, are capable of collecting and producing vast quantities of data that can be mined for novel inferences. To date, powerful predictive models have made breakthroughs in long-standing scientific problems, including solving a 50-year challenge of predicting protein structure. Though no small achievement, these successes have often been buoyed by several favorable characteristics of the target problems, including:
(1) simple physical properties (e.g., invariance to translations and rotations) that can be easily incorporated into the modeling procedure,
(1) well-defined objectives that admit straightforward assessments of success (e.g., using atomic position metrics to assess protein structure accuracy), and
(3) access to vast computational resources (e.g., thousands of GPUs) that allow training powerful models.
Problems that exhibit these favorable characteristics represent only a small fraction of the important scientific challenges of today. In many significant applications, such as designing novel drugs, inferring neural connectivity patterns, or predicting the effects of climate change, data-driven approaches are often far messier and more complex.
Challenge C1: Expert domain knowledge (e.g., known brain connectivity patterns) is often difficult to incorporate into the modeling process.
Challenge C2: Desiderata (e.g., designing an efficacious drug with minimal side effects) are often impossible to codify mathematically and/or expensive to measure experimentally, requiring the use of potentially unreliable proxy metrics.
Challenge C3: Finally, the computational requirements, implementation efforts, and technical overhead of today's successful machine learning approaches are overly burdensome for many researchers.
My research aims to develop novel methods that address these challenges, unlocking the potential of machine learning for inferences and discoveries across scientific domains. I build upon the tools of probabilistic machine learning, which offers a principled framework for incorporating expert domain knowledge and reasoning under uncertainty. Towards my research aim, this proposal focuses on three broad classes of models:
(A) large-scale spatiotemporal models for inferences in the physical sciences,
(B) high-dimensional blackbox optimizers for discovery/decision-making in high-throughput experiments, and
(C) reliable neural networks for complex data modalities.
Each of the three directions builds on my prior work on Gaussian processes, Bayesian optimization, and neural networks. All methodological advances will be made accessible to the larger scientific community through interdisciplinary collaborations and open-source software.
Program
Discovery Grants Program - Individual
Recipient
Location
British Columbia, Canada
Recipient Type
Academia
Type
Grant
Source
NSERC's Awards Data
As published
Application id
RGPIN-2024-06405
Name nom
Pleiss, Geoff
Department département
Statistics
Organization id
2
Institution établissement
The University of British Columbia
Province en
British Columbia
Province fr
Colombie-Britannique
Country en
CANADA
Country fr
CANADA
Fiscal year exercice financier
2024
Competition year année de concours
2024
Award amount
41000
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
1507
Committee name en
Computer Science
Committee name fr
Sciences informatiques
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
2800
Research subject group en
Artificial intelligence
Research subject group fr
Intelligence artificielle
Research subject en
Artificial intelligence (computer vision, use2603)
Research subject fr
Intelligence artificielle (vision artificielle, utiliser 2603)
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
Probabilistic Models for Complex and Large-Scale Scientific Discovery
Keywords
machine learning; probabilistic modeling; bayesian optimization; gaussian processes; neural networks; uncertainty quantification
Application summary
Predictive machine learning is revolutionizing the scientific process. High-throughput experimental methods, which are becoming the norm in fields ranging from astronomy to solid-state physics, are capable of collecting and producing vast quantities of data that can be mined for novel inferences. To date, powerful predictive models have made breakthroughs in long-standing scientific problems, including solving a 50-year challenge of predicting protein structure. Though no small achievement, these successes have often been buoyed by several favorable characteristics of the target problems, including:
(1) simple physical properties (e.g., invariance to translations and rotations) that can be easily incorporated into the modeling procedure,
(1) well-defined objectives that admit straightforward assessments of success (e.g., using atomic position metrics to assess protein structure accuracy), and
(3) access to vast computational resources (e.g., thousands of GPUs) that allow training powerful models.
Problems that exhibit these favorable characteristics represent only a small fraction of the important scientific challenges of today. In many significant applications, such as designing novel drugs, inferring neural connectivity patterns, or predicting the effects of climate change, data-driven approaches are often far messier and more complex.
Challenge C1: Expert domain knowledge (e.g., known brain connectivity patterns) is often difficult to incorporate into the modeling process.
Challenge C2: Desiderata (e.g., designing an efficacious drug with minimal side effects) are often impossible to codify mathematically and/or expensive to measure experimentally, requiring the use of potentially unreliable proxy metrics.
Challenge C3: Finally, the computational requirements, implementation efforts, and technical overhead of today's successful machine learning approaches are overly burdensome for many researchers.
My research aims to develop novel methods that address these challenges, unlocking the potential of machine learning for inferences and discoveries across scientific domains. I build upon the tools of probabilistic machine learning, which offers a principled framework for incorporating expert domain knowledge and reasoning under uncertainty. Towards my research aim, this proposal focuses on three broad classes of models:
(A) large-scale spatiotemporal models for inferences in the physical sciences,
(B) high-dimensional blackbox optimizers for discovery/decision-making in high-throughput experiments, and
(C) reliable neural networks for complex data modalities.
Each of the three directions builds on my prior work on Gaussian processes, Bayesian optimization, and neural networks. All methodological advances will be made accessible to the larger scientific community through interdisciplinary collaborations and open-source software.