Learning to stop while learning to predict

ICML2020 9-25-2020

Motivation

  • Task-imbalanced meta learning: different tasks need different numbers of gradient steps for adaptation

  • Deep learning based algorithms usually have a fixed number of iterations in the architecture.

could we learn to stop automatically?

Overview

  • predictive model Fθ\mathcal{F}_\theta : transforms the input x to generate a path of states x1,x2,...,xTx_1, x_2, ...,x_T

  • stop policy πϕ\pi_\phi : sequentially observes the states xtx_t and determines the probability of stop at layer tt

  • variational stop time distribution qÏ•q_\phi : stop time distribution induced by stopping policy πϕ\pi_\phi

Reference

Last updated

Was this helpful?