Canada Spends Logo

Explorez notre analyse du budget 2026

Grant: Discovery Grants Program - Individual

Carleton University

$36,000.00

Vector-symbolic implementation of the common model of cognition

EF 2024-25

Résumé
Humans acquire new skills throughout our lifetime. What is it about human minds/brains that allows for continuous lifelong learning? My research is in computational models of memory that use Hebbian learning, an algorithm that models how neurons in the brain wire together to encode new skills or knowledge. Conversely, the artificial neural networks used in most artificial intelligence applications, such as machine translation and face recognition, learn using back-propagation. Back-propagation networks suffer from catastrophic interference: if a network is trained on a new task, it will forget all prior tasks (if the AI learns to play the piano, it will forget how to ride a bicycle). For a back-propagation network to learn multiple tasks, it needs to learn in parallel (the AI learns to play the piano while riding a bicycle). To avoid this problem, I specialize in models that use Hebbian learning, which do not suffer from catastrophic interference and can scale to continuous, lifelong learning. My long-term objective is to build a new cognitive architecture, a model of the mind's sensory, motor, memory, and learning systems, that can sequentially learn new tasks while retaining skill at prior tasks. In designing my architecture, I use the Common Model of Cognition as a blueprint. The Common Model of Cognition is a high-level description of mind/brain function, an emergent consensus from decades of research. My short-term objectives, which allow my architecture to avoid catastrophic interference, are to model four aspects of learning in brains: (1) how the brain learns associations through interacting patterns of neural activity, (2) how the brain makes and corrects predictions, (3) how neurogenesis in the hippocampus supports the encoding of new memories, and (4) how cortical regions self-organize to allocate resources to different tasks (e.g., piano players have more brain dedicated to their fingers). I have published results from a prototype of my architecture on a maze-learning task, where it performs on par with or better than back-propagation networks. As detailed in this proposal, the next steps are to fully develop the architecture and to evaluate it on benchmark tasks for AI (e.g., simple video games). My goal is not to outperform the state-of-the-art on any given task but to show a proof-of-concept model that learns continuously across tasks. Advances in neural networks made available on open source platforms are widely adopted, and thus my work will influence the development of AI systems. The proposed architecture will be useful for training students in machine learning, as a tool for collaboration between researchers, and as a platform for a new generation of AI. Development of an architecture that is immune to catastrophic interference has the potential to be both a significant theoretical step in closing the gap between machine and human intelligence and a powerful machine learning tool with both academic and industrial applications.
Programme
Discovery Grants Program - Individual
Bénéficiaire
Emplacement
Ontario, Canada
Type de bénéficiaire
Academia
Type
Grant
Source
NSERC's Awards Data

Tel que publié

Application id
RGPIN-2023-05268
Name nom
Kelly, Mary
Department département
Cognitive Science
Organization id
23
Institution établissement
Carleton University
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
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
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
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
Vector-symbolic implementation of the common model of cognition
Keywords
cognitive architectures; learning; memory; artificial neural networks; catastrophic interference; vector-symbolic architectures; holographic reduced representations; reinforcement learning
Application summary
Humans acquire new skills throughout our lifetime. What is it about human minds/brains that allows for continuous lifelong learning? My research is in computational models of memory that use Hebbian learning, an algorithm that models how neurons in the brain wire together to encode new skills or knowledge. Conversely, the artificial neural networks used in most artificial intelligence applications, such as machine translation and face recognition, learn using back-propagation. Back-propagation networks suffer from catastrophic interference: if a network is trained on a new task, it will forget all prior tasks (if the AI learns to play the piano, it will forget how to ride a bicycle). For a back-propagation network to learn multiple tasks, it needs to learn in parallel (the AI learns to play the piano while riding a bicycle). To avoid this problem, I specialize in models that use Hebbian learning, which do not suffer from catastrophic interference and can scale to continuous, lifelong learning. My long-term objective is to build a new cognitive architecture, a model of the mind's sensory, motor, memory, and learning systems, that can sequentially learn new tasks while retaining skill at prior tasks. In designing my architecture, I use the Common Model of Cognition as a blueprint. The Common Model of Cognition is a high-level description of mind/brain function, an emergent consensus from decades of research. My short-term objectives, which allow my architecture to avoid catastrophic interference, are to model four aspects of learning in brains: (1) how the brain learns associations through interacting patterns of neural activity, (2) how the brain makes and corrects predictions, (3) how neurogenesis in the hippocampus supports the encoding of new memories, and (4) how cortical regions self-organize to allocate resources to different tasks (e.g., piano players have more brain dedicated to their fingers). I have published results from a prototype of my architecture on a maze-learning task, where it performs on par with or better than back-propagation networks. As detailed in this proposal, the next steps are to fully develop the architecture and to evaluate it on benchmark tasks for AI (e.g., simple video games). My goal is not to outperform the state-of-the-art on any given task but to show a proof-of-concept model that learns continuously across tasks. Advances in neural networks made available on open source platforms are widely adopted, and thus my work will influence the development of AI systems. The proposed architecture will be useful for training students in machine learning, as a tool for collaboration between researchers, and as a platform for a new generation of AI. Development of an architecture that is immune to catastrophic interference has the potential to be both a significant theoretical step in closing the gap between machine and human intelligence and a powerful machine learning tool with both academic and industrial applications.