Bristol Benjamin Meaker Distinguished Visiting Professor Yakov Ben-Haim, Technion-Israel Institute of Technology, Israel

Info-gap theory for managing deep uncertainty in distributed learning and collective decision making by robot swarms

Visit dates to be confirmed for 2021/22

Biography

Professor Yakov Ben-Haim initiated and developed info-gap decision theory for modelling and managing severe uncertainty. Info-gap theory is a decision-support tool, providing a methodology for assisting in assessment and selection of policy, strategy, action, or decision in a wide range of disciplines. Info-gap theory has impacted the fundamental understanding of uncertainty in human affairs, and is applied in decision making by scholars and practitioners around the world in engineering, biological conservation, economics, project management, climate change, natural hazard response, national security, medicine, and other areas (see info-gap.com). He has been a visiting scholar in many countries and has lectured at universities, technological and medical research institutions, public utilities and central banks. He has published more than 100 articles and 6 books. He is a professor of mechanical engineering and holds the Yitzhak Moda'i Chair in Technology and Economics at the Technion - Israel Institute of Technology.

Summary

Distributed learning and collective decision making by interacting agents occur in many human and artificial systems. Such systems face deep uncertainties that arise from diverse sources, including variable reliability of individual agents in the collective, faulty communication, incomplete data or understanding, changing dynamics of the environment, strategic behavior by agents or adversaries, etc.

The basic idea of info-gap robust satisficing is to first identify outcomes that are essential – goals that must be achieved – and then to choose the option that will achieve those critical outcomes over the greatest range of future surprise. In this context knowledge is used in two ways. First, to assess the putative desirability of the alternative options, and second, to evaluate the vulnerability of those options to surprising future developments.

The robust-satisficing strategy is the one with maximal robustness against deep uncertainty while satisfying the critical requirements. In other words, what is optimized is not the predicted quality of the outcome, but rather the immunity to error and surprise. The outcome will be satisfactory, though not necessarily optimal, over the greatest range of future deviations from our current understanding. What constitutes a satisfactory outcome can be as modest or as demanding as one wants, though the robustness decreases as the demands increase.

We focus on the following questions.

Uncertainty. What are the dominant uncertainties in learning and decision making, and how to formulate info-gap models of uncertainty?

Robustness. We study the robustness – to uncertainty – of the learning and decision processes. How is the info-gap robustness assessed and what trade-offs are observed?

Learning and Decision-making. How is the concept of robustness used to evaluate and prioritize algorithms for distributed learning and decision-making?

Swarm design. The previous questions provide the basis for overall design of the swarm.

Professor Ben-Haim is hosted by Professor Jonathan Lowry, Engineering Mathematics.

Planned events include:

Public Lecture
Info-Gap Theory: An Overview

Departmental Lecture
Strategic Planning for Acute Adverse Multi-Site Events: An Info-Gap Analysis

Postgraduate Seminar
Uncertainty, probability, and ignorance: A methodological discussion