Ph.D. Proposal: Daniel Larsson

Thu Sep 02 2021 01:00 PM
MK 317
"Information-Theoretic Driven Frameworks for Resource-Aware Abstraction and Planning in Autonomous Systems"

You're invited to attend


Ph.D. Proposal


Daniel Larsson

(Advisor: Prof. Panagiotis Tsiotras)


"Information-Theoretic Driven Frameworks for Resource-Aware Abstraction and Planning in Autonomous Systems"


Thursday, September 2
1 p.m.
Montgomery Knight (MK) 317


The process of abstraction, or equivalently, the identification of relevant and irrelevant information, is a task humans perform subconsciously everyday. The ability to focus on details that are task-relevant, and abstract away those that are not, is considered cornerstone to human intelligence and information processing. Inspired by their ability to simplify problems by removing irrelevant details, researchers within the intelligent systems community have leveraged the power of abstractions to reduce the complexity of solving real-world problems in autonomous decision-making and control. However, despite their role in enabling autonomous agents to solve complex problems, the design of abstractions has been traditionally handled by system engineers, who provide heuristic, domain-specific knowledge that guides the construction of the reduced-order representations. For this reason, a growing interest in the development of frameworks that design task-relevant abstractions for autonomous agents has emerged, motivated largely by the central role of abstraction in intelligent systems.

To design task-relevant abstractions requires the preservation of relevant information through the process of compression. A formal treatment of these notions has been considered by information-theorists, who have developed a number of powerful frameworks for signal compression that rigorously capture the trade-off between relevant information retention and compression when encoding signals for transmission across capacity-limited communication channels. Of particular interest is the information-bottleneck framework, which formulates an optimization problem to design encoders that maximize compression while remaining maximally retentive regarding task-relevant information. Other frameworks, such as rate-distortion theory, capture the same fundamental compromise between compression and information quality, but do so through provided distortion functions that implicitly define the relevant aspects of a signal. In recent years, frameworks that employ rate-distortion-like and IB-like approaches in order to design latent representations for autonomous systems have been developed, but with varying degrees of success, reproducibility, and theoretical guarantees.

In this research, we propose to investigate and develop frameworks that leverage ideas from signal compression to generate and design abstractions for autonomous systems. The proposed frameworks allow for task-specific, multi-resolution, tree abstractions to be obtained that are not provided to the system a priori, instead emerging as a function of the agent's resource constraints. We present a number of frameworks that generate abstractions, enabling intelligent systems to create abstractions in order to reduce memory requirements for storing environment representations, create abstract graphs to reduce the complexity of executing graph-search algorithms for path-planning, and to consider communication constraints when building environment models remotely.

To accomplish our goal, we draw on connections between tree abstractions and signal encoders to formulate information-theoretic compression problems over the space of multi-resolution tree abstractions. We show how our problem can be tractably solved by developing a number of novel algorithms for which we provide theoretical guarantees. Furthermore, we demonstrate how an information-theoretic, multi-resolution, tree abstraction problem can be formulated as a mixed-integer linear program, and provide a number of examples to demonstrate the utility of our approach. For future work, we propose to investigate generalizations of our framework to consider cases where additional (side, or confidential) information is provided, and to scenarios where a time-correlated sequences of inter-related abstractions are designed, motivated by applications of remote sensing and estimation. We close by providing detailed problem formulations for the proposed work, and a discussion regarding interesting aspects of future research that warrant further investigation.


  • Prof. Panagiotis Tsiotras – School of Aerospace Engineering (advisor)
  • Prof. Kyriakos Vamvoudakis – School of Aerospace Engineering
  • Prof. Matthieu Bloch – School of Electrical and Computer Engineering
  • Prof. Faramarz Fekri – School of Electrical and Computer Engineering


MK 317