Timezone: »

 
Poster
End-to-end Algorithm Synthesis with Recurrent Networks: Extrapolation without Overthinking
Arpit Bansal · Avi Schwarzschild · Eitan Borgnia · Zeyad Emam · Furong Huang · Micah Goldblum · Tom Goldstein

Tue Nov 29 02:00 PM -- 04:00 PM (PST) @ Hall J #716

Machine learning systems perform well on pattern matching tasks, but their ability to perform algorithmic or logical reasoning is not well understood. One important reasoning capability is algorithmic extrapolation, in which models trained only on small/simple reasoning problems can synthesize complex strategies for large/complex problems at test time. Algorithmic extrapolation can be achieved through recurrent systems, which can be iterated many times to solve difficult reasoning problems. We observe that this approach fails to scale to highly complex problems because behavior degenerates when many iterations are applied -- an issue we refer to as "overthinking." We propose a recall architecture that keeps an explicit copy of the problem instance in memory so that it cannot be forgotten. We also employ a progressive training routine that prevents the model from learning behaviors that are specific to iteration number and instead pushes it to learn behaviors that can be repeated indefinitely. These innovations prevent the overthinking problem, and enable recurrent systems to solve extremely hard extrapolation tasks.

Author Information

Arpit Bansal (University of Maryland, College Park)
Avi Schwarzschild (University of Maryland)
Eitan Borgnia (University of Maryland)
Zeyad Emam (University of Maryland, College Park)
Furong Huang (University of Maryland)
Micah Goldblum (University of Maryland)
Tom Goldstein (University of Maryland)

More from the Same Authors