Upcoming Seminars

Speaker

Isaac Stone, PhD student at the University of Alberta, supervised by Dr. Nathan Sturtevant and Dr. Jonathan Schaeffer

Title

How Computers will Master the Game of Bridge

Abstract

The card game Bridge has remained a challenge for computers despite algorithmic advances in game-playing AI and advances in computational hardware over recent decades. By the early 2000s, computer Bridge programs achieved performance comparable to strong human players. Since that time, progress toward expert-level play has been somewhat limited.

This talk examines the limitations of traditional approaches to computer Bridge, which treat the game primarily as a challenge of action selection. I present an alternative framing based on comprehensive strategy selection — that is, constructing comprehensive strategies that consider only information available to all players. This diverges from traditional approaches, which incorporate private information (i.e., the cards a player holds) to reduce the search space. I will describe algorithms we have developed to make strategy construction computationally tractable, present recent results, and discuss the challenges involved in scaling this work to allow computers to finally master the game of Bridge.

Bio

Isaac Stone is a current PhD student in the department, supervised by AMII fellow Nathan R. Sturtevant and Prof. Emeritus Jonathan Schaeffer.  His research focuses on search algorithms, with a keen interest in understanding the challenges that — to-date — have prevented computers from achieving performance comparable to top human players in the card game Bridge.

Timing & Location

UComm Seminar Room 2-108

pizza from 11:30, seminar from noon to 1

Add event to calendar

Add event

Speaker

Owen Randal, PhD candidate at the University of Alberta, supervised by Dr. Martin Muller

Title

Astrus: AlphaGo-Inspired Search and Learning for Physics-Aware Chip Design

Abstract

AMS (analog mixed-signal) layout is the art of arranging transistors and wires for a chip’s most performance-critical parts. Every placement and routing decision can profoundly impact performance, which is why engineers spend weeks hand-tuning transistor placement and routing, through intuition and painstaking trial and error to meet performance requirements in circuit simulation. Although AMS circuits occupy only a small fraction of a sophisticated chip like a GPU, they can account for a disproportionate share of the manual design effort and cost.

Like AlphaGo, we use search as the engine for discovering novel configurations. A physics-based evaluation engine measures the physical properties of each layout, allowing us to trade off among many competing design objectives. Predictive neural network models approximate complex routing and computationally expensive evaluation to accelerate search, while generative models distill the results to guide future exploration. Together, these techniques allow us to discover layouts that satisfy stringent physical constraints and satisfy user-specified performance objectives.

Bio

Owen Randall is a current PhD candidate at the University of Alberta and former CMPUT 455 instructor, supervised by AMII fellow Martin Müller. His research is primarily concerned with search algorithms and combinatorial optimization. He is doing research at Astrus, a Toronto based startup using AI and search to automate and optimize analog chip layouts.

Website

Astrus AI

Timing & Location IN-PERSON ONLY

UComm Seminar Room 2-108

pizza from 11:30, seminar from noon to 1

Add event to calendar

Add event

Speaker

TBA

Speaker

TBA

Speaker

TBA

NO SEMINARS - Winter Closure