Expert Advice in Complex Environments

Channel:
United States Simons Institute for the Theory of Computing
Subscribers:
68,700
Published on ● Video Link: https://www.youtube.com/watch?v=MMaJoanFtQ8


Duration: 51:15
413 views
11

Steve Callander (Stanford University)
https://simons.berkeley.edu/talks/tbd-466
Quantifying Uncertainty: Stochastic, Adversarial, and Beyond

The standard approach of economic theory is to simplify, simplify, simplify. In many contexts, from experimentation to search to communication, this implies a state of the world that is often as simple as a single piece of information. Reality is much messier, and typically there is much that decision makers don’t know. This distinction is particularly stark in models of expert advice. The insight from classic models of cheap talk is that communication is imperfect when the expert is biased, and the outcome is both inefficient and unfavorable to the expert, whereas in practice, experts hold considerable power over uninformed decision-makers. We show that this dissonance is due to simplification of expertise in classic models. We introduce a model of expert advice in which the mapping from actions to outcomes is given by the realized path of a Brownian motion. We show in this environment that strategic communication can be efficient and highly favorable to the expert. Communication is efficient not despite the complexity, but rather because of it.



Other Videos By Simons Institute for the Theory of Computing


2022-09-14Machine Learning for Faster Optimization
2022-09-14Flow Time Scheduling with Uncertain Processing Time
2022-09-14Best Of Both Worlds: Stochastic & Adversarial Best-Arm Identification
2022-09-14Best of Both World Algorithms from I.I.D. to Adversarial Data
2022-09-14Corruption-Robust Contextual Search
2022-09-14Contextual Inverse Optimization: Offline and Online Learning
2022-09-13Causal Inference in Complex Systems: Network Interference, Strategic Agents, and Beyond
2022-09-13Inference and Interference in Marketplace Experimentation
2022-09-13Causal Matrix Completion: Applications to Offline Causal Reinforcement Learning
2022-09-13Greedy Approximation Algorithms for Active Sequential Hypothesis Testing
2022-09-13Expert Advice in Complex Environments
2022-09-12Retrospective Search: Exploration and Ambition on Uncharted Terrain
2022-09-12Dynamically Aggregating Diverse Information
2022-09-10Simplicity and Optimality in Algorithmic Economics
2022-09-10Algorithms Using Local Graph Features to Predict Epidemics
2022-09-10Beyond Worst Case Analysis in ML
2022-09-09Pseudorandom Generators and Small-Space Derandomization
2022-09-09Competitive Control
2022-09-09Machine Learning on Large-Scale Graphs
2022-09-09Adaptive Online Learning without Prior Knowledge
2022-09-09Incentive-Aware Machine Learning


Tags:
Simons Institute
theoretical computer science
UC Berkeley
Computer Science
Theory of Computation
Theory of Computing
Quantifying Uncertainty: Stochastic Adversarial and Beyond
Steve Callander