For this week’s speaker event we will be hosting Pedro Ortega as he gives his talk on “An Introduction to Information-Theoretic Bounded Rationality”. Come down to the statistics department at 6pm on Tuesday the 7th to hear what he has to say - all followed by a free pizza social immediately afterwards!

Bounded rationality, the art of decision-making with limited resources, is an important challenge in AI, neuroscience, and economics. This talk reviews an information-theoretic formalization that offers a sharp characterization.

The theory champions the free energy functional as the go-to objective function. This functional excels for three reasons: it narrows the solutionspace, enables Monte Carlo planners that are precise without exhaustivesearch, and encapsulates model uncertainty arising from insufficient evidence or interactions with agents of uncertain motives.

We delve into the realm of single-step decision-making and demonstrate its expansion to sequential decisions through equivalence transformations. This broadens our scope to a vast array of decision problems, capturing both classical decision rules (like EXPECTIMAX and MINIMAX) and more nuanced, trust- and risk-sensitive planning approaches."