Grant: Discovery Grants Program - Individual
University of Waterloo
$41,000.00
Teaching machines to learn more efficiently and more generally by reading problem descriptions
FY 2024-25
Summary
Summary
How can we enable machine learning (ML) models to solve new problems in the real world without relying on large labeled datasets? For example, training a furniture assembly robot requires large amounts of human assembly demonstrations; training a computer tutor requires large quantities of tutorial conversations on specialized topics. When faced with new problems such as assembling new types of furniture or teaching new proprietary content, current expensively-trained ML models cannot generalize to the new problem. In many cases, the collection of labeled data is expensive and requires domain experts, but the collection of language descriptions of what the problem is and what solutions are possible is relatively cheap and can be performed by non-experts.
This research on Reading to Learn (R2L) seeks to enable effective and generalizable ML by reading the information that people refer to when they face a new problem. For instance, a robotic arm assembles new types of furniture after reading assembly manuals; a computer tutor teaches employees how to use proprietary software by reading an internal knowledge base. The objectives of this proposed program are: 1) Develop tests to evaluate generalization to new grounded environments in real-world applications such as web applications, robotics, and interactive teaching. 2) Build efficient generalist R2L agents that reason over long-context descriptions and open-domain manuals. 3) Enable automatic curriculum discovery from background knowledge to enable agents that automatically decompose goals into solvable subgoals. 4) Develop open-source software to facilitate the study of R2L in academia and in industry.
The long-term goal of this research is to produce ML models that are not constrained by data availability, but empowered by the ability to read to learn. The impact of this research includes: 1) Adaptable ML that generalize to new complex problems using expanding repositories of textual knowledge to ensure continued relevance and applicability in a rapidly changing world. 2) Collaborative ML agents that understand and learn from human-generated content to assist humans in solving complex problems. 3) Systems that autonomously acquire knowledge from textual sources to tackle new problems with minimal human intervention. These capabilities will accelerate the adoption of efficient and generalizable ML across industries such as technology, health-care, and finance both in Canada and globally. Essential to the program, students will learn the necessary skills to formulate and execute research plans, scientific writing, critical thinking, and coding. By the end of the proposed program, students will become machine learning and natural language processing experts well-positioned for industrial research scientist positions, industrial engineering positions, and academic research positions at top universities worldwide.
Program
Discovery Grants Program - Individual
Recipient
Location
Ontario, Canada
Recipient Type
Academia
Type
Grant
Source
NSERC's Awards Data
As published
Application id
RGPIN-2024-04040
Name nom
Zhong, Victor
Department département
David R. Cheriton School of Computer Science
Organization id
33
Institution établissement
University of Waterloo
Province en
Ontario
Province fr
Ontario
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
2801
Research subject group en
Artificial intelligence
Research subject group fr
Intelligence artificielle
Research subject en
Natural language and speech understanding
Research subject fr
Langage naturel et reconnaissance de la parole
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
Teaching machines to learn more efficiently and more generally by reading problem descriptions
Keywords
machine learning; natural language processing; deep learning; generalization; few-shot learning; reinforcement learning; in-context learning; interactive learning; active learning; adaptive systems
Application summary
Summary
How can we enable machine learning (ML) models to solve new problems in the real world without relying on large labeled datasets? For example, training a furniture assembly robot requires large amounts of human assembly demonstrations; training a computer tutor requires large quantities of tutorial conversations on specialized topics. When faced with new problems such as assembling new types of furniture or teaching new proprietary content, current expensively-trained ML models cannot generalize to the new problem. In many cases, the collection of labeled data is expensive and requires domain experts, but the collection of language descriptions of what the problem is and what solutions are possible is relatively cheap and can be performed by non-experts.
This research on Reading to Learn (R2L) seeks to enable effective and generalizable ML by reading the information that people refer to when they face a new problem. For instance, a robotic arm assembles new types of furniture after reading assembly manuals; a computer tutor teaches employees how to use proprietary software by reading an internal knowledge base. The objectives of this proposed program are: 1) Develop tests to evaluate generalization to new grounded environments in real-world applications such as web applications, robotics, and interactive teaching. 2) Build efficient generalist R2L agents that reason over long-context descriptions and open-domain manuals. 3) Enable automatic curriculum discovery from background knowledge to enable agents that automatically decompose goals into solvable subgoals. 4) Develop open-source software to facilitate the study of R2L in academia and in industry.
The long-term goal of this research is to produce ML models that are not constrained by data availability, but empowered by the ability to read to learn. The impact of this research includes: 1) Adaptable ML that generalize to new complex problems using expanding repositories of textual knowledge to ensure continued relevance and applicability in a rapidly changing world. 2) Collaborative ML agents that understand and learn from human-generated content to assist humans in solving complex problems. 3) Systems that autonomously acquire knowledge from textual sources to tackle new problems with minimal human intervention. These capabilities will accelerate the adoption of efficient and generalizable ML across industries such as technology, health-care, and finance both in Canada and globally. Essential to the program, students will learn the necessary skills to formulate and execute research plans, scientific writing, critical thinking, and coding. By the end of the proposed program, students will become machine learning and natural language processing experts well-positioned for industrial research scientist positions, industrial engineering positions, and academic research positions at top universities worldwide.