Learning Multiple Tasks using Shared Hypotheses

Neural Information Processing Systems

In this work we consider a setting where we have a very large number of related tasks with few examples from each individual task. Rather than either learning each task individually (and having a large generalization error) or learning all the tasks together using a single hypothesis (and suffering a potentially large inherent error), we consider learning a small pool of {\em shared hypotheses}. Each task is then mapped to a single hypothesis in the pool (hard association). We derive VC dimension generalization bounds for our model, based on the number of tasks, shared hypothesis and the VC dimension of the hypotheses class. We conducted experiments with both synthetic problems and sentiment of reviews, which strongly support our approach.



Hypothesis Selection Biases for Incremental Learning

AAAI Conferences

Claire NEDELLEC, C line ROUVEIROL LRI, U.R.A. 410 of CNRS, Universit Paris-Sud, Bt. 490 F-95405 Orsay C dex France email: cn, celine@lrifr Abstract The incremental systems directed by hypotheses learn concepts by selecting the candidate hypothesis in advance in the search space and then testing it with examples. The selection of the "interesting" hypotheses is not based on the incoming example as in example-driven systems but on heuristics.


Rule induction from noisy examples Laura Firoiu

AAAI Conferences

This work addresses the problem of rule learning from simple robot experiences like approaching or passing an object. An experience is a sequence of predicates computed by a perceptual system. A difficult problem encountered in this domain by rule induction algorithms is that of noise, not only in the classification of the examples, but also in the facts describing them. Due to perceptual limitations and environment complexity, the descriptions of experiences may have either missing or spurious predicates. I propose a rule induction method based on generalization of clauses under subsumption which takes into consideration the frequency of predicates across examples.


Bayesian Models of Inductive Generalization

Neural Information Processing Systems

We argue that human inductive generalization is best explained in a Bayesian framework, rather than by traditional models based on similarity computations.We go beyond previous work on Bayesian concept learning by introducing an unsupervised method for constructing flexible hypothesisspaces, and we propose a version of the Bayesian Occam's razorthat trades off priors and likelihoods to prevent under-or over-generalization in these flexible spaces. We analyze two published data sets on inductive reasoning as well as the results of a new behavioral study that we have carried out.