Multitask Learning is an approach to inductive transfer that improves generalization by using the domain information contained in the training signals of related tasks as an inductive bias. It does this by learning tasks in parallel while using a shared representation; what is learned for each task can help other tasks be learned better. This paper reviews prior work on MTL, presents new evidence that MTL in backprop nets discovers task relatedness without the need of supervisory signals, and presents new results for MTL with k-nearest neighbor and kernel regression. In this paper we demonstrate multitask learning in three domains. We explain how multitask learning works, and show that there are many opportunities for multitask learning in real domains. We present an algorithm and results for multitask learning with case-based methods like k-nearest neighbor and kernel regression, and sketch an algorithm for multitask learning in decision trees. Because multitask learning works, can be applied to many different kinds of domains, and can be used with different learning algorithms, we conjecture there will be many opportunities for its use on real-world problems.

There are many applications in which it is desirable to order rather than classify instances. Here we consider the problem of learning how to order, given feedback in the form of preference judgments, i.e., statements to the effect that one instance should be ranked ahead of another. We outline a two-stage approach in which one first learns by conventional means a preference function, of the form PREF(u; v), which indicates whether it is advisable to rank u before v. New instances are then ordered so as to maximize agreements with the learned preference function. We show that the problem of finding the ordering that agrees best with a preference function is NP-complete, even under very restrictive assumptions. Nevertheless, we describe a simple greedy algorithm that is guaranteed to find a good approximation. We then discuss an on-line learning algorithm, based on the "Hedge" algorithm, for finding a good linear combination of ranking "experts." We use the ordering algorith...

This paper identifies two fundamentally different kinds of training information for learning search control in terms of an evaluation function. Each kind of training information suggests its own set of methods for learning an evaluation function. The paper shows that one can integrate the methods and learn simultaneously from both kinds of information.

A patient visits the doctor; the doctor reviews the patient's history, asks questions, makes basic measurements (blood pressure, ...), and prescribes tests or treatment. The prescribed course of action is based on an assessment of patient risk---patients at higher risk are given more and faster attention. It is also sequential---it is too expensive to immediately order all tests which might later be of value. This paper presents two methods that together improve the accuracy of backprop nets on a pneumonia risk assessment problem by 10-50%. Rankprop improves on backpropagation with sum of squares error in ranking patients by risk. Multitask learning takes advantage of future lab tests available in the training set, but not available in practice when predictions must be made. Both methods are broadly applicable. 1 Background There are 3,000,000 cases of pneumonia each year in the U.S., 900,000 of which are admitted to the hospital for treatment and testing. Most pneumonia patients rec...

In many optimization and decision problems the objective function can be expressed as a linear combination of competing criteria, the weights of which specify the relative importance of the criteria for the user. We consider the problem of learning such a "subjective" function from preference judgments collected from traces of user interactions. We propose a new algorithm for that task based on the theory of Support Vector Machines. One advantage of the algorithm is that prior knowledge about the domain can easily be included to constrain the solution. We demonstrate the algorithm in a route recommendation system that adapts to the driver's route preferences. We present experimental results on real users that show that the algorithm performs well in practice. 1.

Decision support systems often require knowledge of users' preferences. However, preferences may vary among individual users or be difficult for users to articulate. This paper describes how user preferences can be acquired in the form of preference predicates by a learning apprentice system and proposes two new instance-based algorithms for preference predicate acquisition: 1ARC and Compositional Instance-Based Learning (CIBL). An empirical evaluation using simulated preference behavior indicated that the instance-based approaches are preferable to decision-tree induction and perceptrons as the learning component of a learning apprentice system if representation of the relevant characteristics of problem-solving states requires a large number of attributes, if attributes interact in a complex fashion, or if there are very few training instances. Conversely, decision-tree induction or perceptron learning is preferable if there are a

The size and complexity of software and hardware systems have significantly increased in the past years. As a result, it is harder to guarantee their correct behavior. One of the most successful methods for automated verification of finite-state systems is model checking. Most of the current model-checking systems use binary decision diagrams (BDDs) for the representation of the tested model and in the verification process of its properties.

In this paper we present an algorithm for learning a function able to assess objects. We assume that our teachers can provide a collection of pairwise comparisons but encounter certain difficulties in assigning a number to the qualities of the objects considered. This is a typical situation when dealing with food products, where it is very interesting to have repeatable, reliable mechanisms that are as objective as possible to evaluate quality in order to provide markets with products of a uniform quality. The same problem arises when we are trying to learn user preferences in an information retrieval system or in configuring a complex device. The algorithm is implemented using a growing variant of Kohonen's Self-Organizing Maps (growing neural gas), and is tested with a variety of data sets to demonstrate the capabilities of our approach.

This paper describes Zenith, a discovery system that performs constructive induction. The system is able to generate and extend new features for concept learning using agenda-based heuristic search. The search is guided by feature worth (a composite measure of discriminability and cost). Zenith is distinguished from existing constructive induction systems by its interaction with a performance system, and its ability to extend its knowledge base by creating new domain classes. Zenith is able to discover known useful features for the Othello board game. Feature Discovery for Inductive Concept Learning 1 1 Introduction One of the central concerns of machine learning is that of inductive concept learning from examples, in which a system is given a set of examples and produces a characterization of them. Many induction algorithms have been devised that are able to inductively generalize in different formalisms, using learning rules appropriate for that formalism. For example, decision t...

. Knowledge of the preferences of individual users is essential for intelligent systems whose performance is tailored for individual users, such as agents that interact with human users, instructional environments, and learning apprentice systems. Various memory-based, instance-based, and case-based systems have been developed for preference modeling, but these system have generally not addressed the task of selecting examples to use as queries to the user. This paper describes UGAMA, an approach to learning preference criteria through active exploration. Under this approach, Unit Gradient Approximations (UGAs) of the underlying quality function are obtained at a set of reference points through a series of queries to the user. Equivalence sets of UGAs are then merged and aligned (MA) with the apparent boundaries between linear regions. In an empirical evaluation with artificial data, use of UGAs as training data for an instance-based ranking algorithm (1ARC) led to more a...