Canada Spends Logo

Explorez notre analyse du budget 2026

Grant: Discovery Grants Program - Individual

Université Laval

$25,000.00

Majority votes and duality in machine learning

EF 2011-12

Résumé
This program of research belongs to Machine Learning, a subfield of Artificial intelligence. More specifically, this work is related to the Classification theory (also called Pattern Recognition), and to Reinforcement Learning theory (a theory of sequential decision making under uncertainty). My long run objective is to contribute to mathematical fundaments of Machine Learning, using my expertise in other fields. In the last decades, Machine learning has produced spectacular advances of Computer Science. One can think of Neural Nets, Support Vector Machines, the Reinforcement Learning (RL) paradigm. Despite these advances that come from very interesting ideas, there is a lot left to do for the mathematical fundaments of Machine Learning. I think that such fundaments not only will improve our comprehension of the efficiency of actual methods, but it will provide tools that will allow us to invent more methods or improve the ones we have. In Classification, some of our new results already permit to obtain a novel explanation of state-of-the-arts algorithms, the objective now is to imagine new algorithms that could come out of these new results. The second objective of the program is to generalize those results to the RL framework. Theoretical guarantees are difficult to obtain here, because we turn from a single decision framework, as faced in classification, to a sequence of decisions. It is possible to obtain a reduction from the latter to the former if, in the RL framework, we can get rid of the notion of the states of the process at a given time. Defining a stochastic process without the notion of state involves duality theory, which says that states can be equivalently expressed as a set of properties that they possess. Some results already exist in that direction (e.g.: Predictive State Representation (PSR)). Moreover, such notions of duality exist in a variety of areas (namely in Mathematical Topology). In those areas, duality ideas have proven to be useful already. We expect to obtain a theoretical general framework of the notion of duality for probabilistic processes.
Programme
Discovery Grants Program - Individual
Bénéficiaire
Emplacement
Quebec, Canada
Type de bénéficiaire
Academia
Type
Grant
Source
NSERC's Awards Data

Tel que publié

Application id
262067-2008
Name nom
Laviolette, François
Department département
Informatique et de génie logiciel
Organization id
51
Institution établissement
Université Laval
Province en
Québec
Province fr
Québec
Country en
CANADA
Country fr
CANADA
Fiscal year exercice financier
2011
Competition year année de concours
2008
Award amount
25000
Program id
RGPIN
Program nane en
Discovery Grants Program - Individual
Program name fr
Programme de subventions à la découverte - individuelles
Group en
Discovery grants programs
Group fr
Programmes de subventions à la découverte
Committee code
331
Committee name en
Computing and Information Sciences - B
Committee name fr
Informatique - B
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
2805
Research subject group en
Artificial intelligence
Research subject group fr
Intelligence artificielle
Research subject en
Learning and inference theories
Research subject fr
Théories de l'apprentissage et l'inférence
Partie
4
Nb partie
5
Application title
Majority votes and duality in machine learning
Keyword
NSERC
Application summary
This program of research belongs to Machine Learning, a subfield of Artificial intelligence. More specifically, this work is related to the Classification theory (also called Pattern Recognition), and to Reinforcement Learning theory (a theory of sequential decision making under uncertainty). My long run objective is to contribute to mathematical fundaments of Machine Learning, using my expertise in other fields. In the last decades, Machine learning has produced spectacular advances of Computer Science. One can think of Neural Nets, Support Vector Machines, the Reinforcement Learning (RL) paradigm. Despite these advances that come from very interesting ideas, there is a lot left to do for the mathematical fundaments of Machine Learning. I think that such fundaments not only will improve our comprehension of the efficiency of actual methods, but it will provide tools that will allow us to invent more methods or improve the ones we have. In Classification, some of our new results already permit to obtain a novel explanation of state-of-the-arts algorithms, the objective now is to imagine new algorithms that could come out of these new results. The second objective of the program is to generalize those results to the RL framework. Theoretical guarantees are difficult to obtain here, because we turn from a single decision framework, as faced in classification, to a sequence of decisions. It is possible to obtain a reduction from the latter to the former if, in the RL framework, we can get rid of the notion of the states of the process at a given time. Defining a stochastic process without the notion of state involves duality theory, which says that states can be equivalently expressed as a set of properties that they possess. Some results already exist in that direction (e.g.: Predictive State Representation (PSR)). Moreover, such notions of duality exist in a variety of areas (namely in Mathematical Topology). In those areas, duality ideas have proven to be useful already. We expect to obtain a theoretical general framework of the notion of duality for probabilistic processes.