Abstract +

In this talk, I present a new automata-theoretic model of the brain, based on known psychology and neurophysiology. The intention of the model is to bring CS ideas and understanding to bear on the kinds of computations humans can and can't do in their heads. The model is designed to give insight into human computation in much the same way that finite automata theory gives insight into what processor hardware can and can't do, and Turing machines give insight into what general purpose computers can and can't do.

This work has applications to password generation, wherein humans - working entirely in their head - transform website names into (distinct random-looking) passwords.

Joint work with Santosh Vempala

Abstract +

Human involvement is still essential to Unsupervised Learning and Exploratory Data Analysis. Why do algorithms fail where humans succeed? We present a model that is intended to capture both how humans perform unsupervised learning and how computers might benefit from a more human-like approach. The model, which we call “meta-unsupervised-learning,” incorporates prior related or unrelated learning problems to inform decisions. We present simple agnostic models and algorithms illustrating how the meta approach circumvents impossibility results for novel "meta" problems such as meta-clustering, meta-outlier-removal, meta-feature-selection, and meta-embedding. We also present empirical results showing how the meta approach improves over standard techniques for problems such as outlier removal, choosing a clustering algorithm, and a number of clusters. We also train an unsupervised neural network that learns from prior supervised classification problems drawn from learning problems at openml.org.

Joint work with Vikas Garg from MIT

Abstract +

Many of the chunks of time we have in a day are too short to bother trying to use productively. Think of the time you spend waiting for a meeting to start, riding in an elevator, or standing in line. We try to defrag our time by booking meetings with ourselves, turning off our phones, and taking email vacations. But there is another way. Rather than fighting fragmentation by changing how we work, we can embrace it by changing our tasks to fit the way we actually do work. We call this microproductivity, in which large productivity tasks are broken down into a series of smaller microtasks. The component microtasks can then be completed by the task owner via selfsourcing, or by the crowd via crowdsourcing. The transformation of work into microwork will change when and how people work, and enable individuals and automated processes to efficiently and easily complete complex tasks.