Canada Spends Logo

Explore our Budget 2026 analysis

Grant: Discovery Grants Program - Individual

University of Manitoba

$36,000.00

Event-Driven Signal/System Analysis: Nonparametric Inference and Learning

FY 2024-25

Summary
In recent years, a growing number of industrial and social processes have been automated through Machine Learning (ML) algorithms. State-of-the-art ML procedures achieve the high recognition and prediction accuracy for problems that can be characterized by large data sets defined in terms of high-dimensional feature vectors. However, numerous applications in science and engineering can be characterized by observations in the form of events. Such data consists of a sequence of the occurrence times of events (spike trains) scattered in time. Designing learning algorithms based on spike trains is one of the most promising paradigm in ML. This is due to the inherent sparseness and asynchrony of spike trains yielding ML algorithms with the reduced power consumption while achieving the high prediction performance. The existing ML algorithms for spike trains have been mostly designed in the form of the so-called spiking neural networks (SNN). Training of SNN is commonly performed by converting the conventional neural networks to spiking neurons via some encoding strategy. However, converting methods are inefficient and direct training of SNN is difficult. Moreover, the performance of the proposed algorithms for SNN is evaluated only on benchmark datasets and there is the lack of fundamental bounds on the prediction accuracy. There is, however, the well established probabilistic theory of point processes (PP) that provides powerful tool for handling spiking data. This methodology offers models that are interpretable and less data demanding. A point process is characterized by a local intensity function that may also incorporate the internal history. The probabilistic modelling of PP enables the direct formulation of learning rules expressed in terms of intensity functions. While statistical inference for PP is well developed other machine learning tasks are surprisingly poorly addressed. This includes such learning schemes as supervised classification, semi-supervised learning, transfer learning, and learning from weak supervision. The principal objective of this proposed program is to examine the alternative strategy for learning spiking data based on the theory of point processes. We will focus mostly on the supervised classification problem for point processes ranging from stochastic memoryless models (Cox processes) to models with long memory (Hawkes processes). The long-term objective of this proposal is to explore the combined models where the expressivity and interpretability of PP models will be enhanced by SNN that provide higher capacity to process massive event data. This would allow for fast processing of complex data utilizing the neuromorphic hardware. This proposal addresses timely issues of practical importance and aims at improving Canada's competitiveness in the field of ML. Over this 5-year funding cycle I propose to train 5 Ph.D. and 2 M.Sc. students.
Program
Discovery Grants Program - Individual
Location
Manitoba, Canada
Recipient Type
Academia
Type
Grant
Source
NSERC's Awards Data

As published

Application id
RGPIN-2024-05463
Name nom
Pawlak, Miroslaw
Department département
Electrical & Computer Engineering
Organization id
19
Institution établissement
University of Manitoba
Province en
Manitoba
Province fr
Manitoba
Country en
CANADA
Country fr
CANADA
Fiscal year exercice financier
2024
Competition year année de concours
2024
Award amount
36000
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
1200
Area of application group en
Advancement of knowledge
Area of application group fr
Avancement des connaissances
Area of application en
Advancement of knowledge
Area of application fr
Avancement des connaissances
Research subject code
2500
Research subject group en
Electrical and electronic engineering
Research subject group fr
Génie électrique et électronique
Research subject en
Electrical and electronic engineering
Research subject fr
Génie électrique et électronique
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
Event-Driven Signal/System Analysis: Nonparametric Inference and Learning
Keywords
event-driven data analysis; nonparametric inference; level-crossings signal analysis; nonuniform sampling; temporal point processes; spike trains data; supervised learning; functional data analysis; ultra-low power consumption learning; event-triggered learning
Application summary
In recent years, a growing number of industrial and social processes have been automated through Machine Learning (ML) algorithms. State-of-the-art ML procedures achieve the high recognition and prediction accuracy for problems that can be characterized by large data sets defined in terms of high-dimensional feature vectors. However, numerous applications in science and engineering can be characterized by observations in the form of events. Such data consists of a sequence of the occurrence times of events (spike trains) scattered in time. Designing learning algorithms based on spike trains is one of the most promising paradigm in ML. This is due to the inherent sparseness and asynchrony of spike trains yielding ML algorithms with the reduced power consumption while achieving the high prediction performance. The existing ML algorithms for spike trains have been mostly designed in the form of the so-called spiking neural networks (SNN). Training of SNN is commonly performed by converting the conventional neural networks to spiking neurons via some encoding strategy. However, converting methods are inefficient and direct training of SNN is difficult. Moreover, the performance of the proposed algorithms for SNN is evaluated only on benchmark datasets and there is the lack of fundamental bounds on the prediction accuracy. There is, however, the well established probabilistic theory of point processes (PP) that provides powerful tool for handling spiking data. This methodology offers models that are interpretable and less data demanding. A point process is characterized by a local intensity function that may also incorporate the internal history. The probabilistic modelling of PP enables the direct formulation of learning rules expressed in terms of intensity functions. While statistical inference for PP is well developed other machine learning tasks are surprisingly poorly addressed. This includes such learning schemes as supervised classification, semi-supervised learning, transfer learning, and learning from weak supervision. The principal objective of this proposed program is to examine the alternative strategy for learning spiking data based on the theory of point processes. We will focus mostly on the supervised classification problem for point processes ranging from stochastic memoryless models (Cox processes) to models with long memory (Hawkes processes). The long-term objective of this proposal is to explore the combined models where the expressivity and interpretability of PP models will be enhanced by SNN that provide higher capacity to process massive event data. This would allow for fast processing of complex data utilizing the neuromorphic hardware. This proposal addresses timely issues of practical importance and aims at improving Canada's competitiveness in the field of ML. Over this 5-year funding cycle I propose to train 5 Ph.D. and 2 M.Sc. students.
University of Manitoba - $36,000 Grant (2024-25) | Canada Spends