Grant: Discovery Grants Program - Individual
University of Alberta
$29,000.00
Machine Learning and Creativity
FY 2020-21
Summary
There are a number of human cognitive abilities that, if they were integrated into artificial intelligence (AI) and machine learning (ML) approaches, would expand the types of problems to which these approaches could apply. I identify two in particular: (1) the ability to learn well from small amounts of data and (2) the ability to work effectively with non-experts. There are a variety of arguments for the first, including efficiency and the application of ML to low training data problems. In the field of ML this is known as transfer learning or few-shot learning, adapting knowledge from one problem domain to a problem domain with less training data. As for the second, the majority of potential users of highly technical ML systems are non-experts, and so understanding how to make these systems usable by non-experts is the key to many future ML applications. This is known as human-AI interaction. Both of these problems are longstanding, with significant research put towards solving them.
Computational creativity represents the relatively new field of study into computationally replicating human cognitive processes related to creativity. While it may seem at first unrelated to the above two points, I argue that a creative person can demonstrate the behaviors outlined above: creating novel work from few prior examples and being able to work with other creatives. Thus it follows that one way to work towards these criteria is by studying the crossover between computational creativity and machine learning.
I present two claims related to these points. First, (C1) by building algorithms and systems that more fully reflect human creativity, we can make ML transfer, and adaptation more efficient and effective. Second, (C2) by taking a human-centered approach to the design of co-creative interfaces and backend AI agents, we can derive general best practices for interactions between creative AI and non-expert human users. If successful, this work will expand the application domains in which machine learning can be applied, and the success of these applications.
Toward this goal I propose three particular aims: (A1) applying computational creativity to reinforcement learning to leverage learned rules of the environment, (A2) applying computational creativity as a form of transfer learning in deep neural networks, and (A3) developing novel AI approaches and frameworks to improve human-AI interaction in the creation of video games. Aims 1-2 are motivated by C1; A3 by C2. I discuss these in greater detail in the proposal.
Program
Discovery Grants Program - Individual
Recipient
Location
Alberta, Canada
Recipient Type
Academia
Type
Grant
Source
NSERC's Awards Data
As published
Application id
RGPIN-2020-04852
Name nom
Guzdial, Matthew
Department département
Computer Science
Organization id
9
Institution établissement
University of Alberta
Province en
Alberta
Province fr
Alberta
Country en
CANADA
Country fr
CANADA
Fiscal year exercice financier
2020
Competition year année de concours
2020
Award amount
29000
Program id
RGPIN
Program name 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
1507
Committee name en
Computer Science
Committee name fr
Sciences informatiques
Area of application code
801
Area of application group en
Information and communication services
Area of application group fr
Services d'information et de télécommunications
Area of application en
Computer software
Area of application fr
Logiciel
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
Installment
inst:1-5
Partie
5
Nb partie
5
Application title
Machine Learning and Creativity
Keyword
NSERC
Application summary
There are a number of human cognitive abilities that, if they were integrated into artificial intelligence (AI) and machine learning (ML) approaches, would expand the types of problems to which these approaches could apply. I identify two in particular: (1) the ability to learn well from small amounts of data and (2) the ability to work effectively with non-experts. There are a variety of arguments for the first, including efficiency and the application of ML to low training data problems. In the field of ML this is known as transfer learning or few-shot learning, adapting knowledge from one problem domain to a problem domain with less training data. As for the second, the majority of potential users of highly technical ML systems are non-experts, and so understanding how to make these systems usable by non-experts is the key to many future ML applications. This is known as human-AI interaction. Both of these problems are longstanding, with significant research put towards solving them.
Computational creativity represents the relatively new field of study into computationally replicating human cognitive processes related to creativity. While it may seem at first unrelated to the above two points, I argue that a creative person can demonstrate the behaviors outlined above: creating novel work from few prior examples and being able to work with other creatives. Thus it follows that one way to work towards these criteria is by studying the crossover between computational creativity and machine learning.
I present two claims related to these points. First, (C1) by building algorithms and systems that more fully reflect human creativity, we can make ML transfer, and adaptation more efficient and effective. Second, (C2) by taking a human-centered approach to the design of co-creative interfaces and backend AI agents, we can derive general best practices for interactions between creative AI and non-expert human users. If successful, this work will expand the application domains in which machine learning can be applied, and the success of these applications.
Toward this goal I propose three particular aims: (A1) applying computational creativity to reinforcement learning to leverage learned rules of the environment, (A2) applying computational creativity as a form of transfer learning in deep neural networks, and (A3) developing novel AI approaches and frameworks to improve human-AI interaction in the creation of video games. Aims 1-2 are motivated by C1; A3 by C2. I discuss these in greater detail in the proposal.