Prospective Learning: Principled Extrapolation to the Future
Abstract
Learning is a process which can update decision rules, based on past experience, such that future performance improves. Traditionally, machine learning is often evaluated under the assumption that the future will be identical to the past in distribution or change adversarially. But these assumptions can be either too optimistic or pessimistic for many problems in the real world. Real world scenarios evolve over multiple spatiotemporal scales with partially predictable dynamics. Here we reformulate the learning problem to one that centers around this idea of dynamic futures that are partially learnable. We conjecture that certain sequences of tasks are not retrospectively learnable (in which the data distribution is fixed), but are prospectively learnable (in which distributions may be dynamic), suggesting that prospective learning is more difficult in kind than retrospective learning. We argue that prospective learning more accurately characterizes many real world problems that (1) currently stymie existing artificial intelligence solutions and/or (2) lack adequate explanations for how natural intelligences solve them. Thus, studying prospective learning will lead to deeper insights and solutions to currently vexing challenges in both natural and artificial intelligences.
Cite
@article{arxiv.2201.07372,
title = {Prospective Learning: Principled Extrapolation to the Future},
author = {Ashwin De Silva and Rahul Ramesh and Lyle Ungar and Marshall Hussain Shuler and Noah J. Cowan and Michael Platt and Chen Li and Leyla Isik and Seung-Eon Roh and Adam Charles and Archana Venkataraman and Brian Caffo and Javier J. How and Justus M Kebschull and John W. Krakauer and Maxim Bichuch and Kaleab Alemayehu Kinfu and Eva Yezerets and Dinesh Jayaraman and Jong M. Shin and Soledad Villar and Ian Phillips and Carey E. Priebe and Thomas Hartung and Michael I. Miller and Jayanta Dey and Ningyuan and Huang and Eric Eaton and Ralph Etienne-Cummings and Elizabeth L. Ogburn and Randal Burns and Onyema Osuagwu and Brett Mensh and Alysson R. Muotri and Julia Brown and Chris White and Weiwei Yang and Andrei A. Rusu and Timothy Verstynen and Konrad P. Kording and Pratik Chaudhari and Joshua T. Vogelstein},
journal= {arXiv preprint arXiv:2201.07372},
year = {2023}
}
Comments
Accepted at the 2nd Conference on Lifelong Learning Agents (CoLLAs), 2023