Rich Caruana

Multitask Learning

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.

An empirical comparison of supervised learning algorithms

A number of supervised learning methods have been introduced in the last decade. Unfortunately, the last comprehensive empirical evaluation of supervised learning was the Statlog Project in the early 90s. We present a large-scale empirical comparison between ten supervised learning methods: SVMs, neural nets, logistic regression, naive bayes, memory-based learning, random forests, decision trees, bagged trees, boosted trees, and boosted stumps. We also examine the effect that calibrating the models via Platt Scaling and Isotonic Regression has on their performance. An important aspect of our study is the use of a variety of performance criteria to evaluate the learning methods.

Experience with a learning personal assistant

Personal software assistants that help users with tasks like finding information, scheduling calendars, or managing work-flow will require significant customization to each individual user. For example, an assistant that helps schedule a particular user’s calendar will have to know that user’s scheduling preferences. This paper explores the potential of machine learning methods to automatically create and maintain such customized knowledge for personal software assistants. We describe the design of one particular learning assistant: a calendar manager, called CAP (Calendar APprentice), that learns user scheduling preferences from experience. Results are summarized from approximately five user-years of experience, during which CAP has learned an evolving set of several thousand rules that characterize the scheduling preferences of its users. Based on this experience, we suggest that machine learning methods may play an important role in future personal software assistants.

Ensemble selection from libraries of models

We present a method for constructing ensembles from libraries of thousands of models. Model libraries are generated using different learning algorithms and parameter settings. Forward stepwise selection is used to add to the ensemble the models that maximize its performance. Ensemble selection allows ensembles to be optimized to performance metric such as accuracy, cross entropy, mean precision, or ROC Area. Experiments with seven test problems and ten metrics demonstrate the benefit of ensemble selection.

Predicting good probabilities with supervised learning

We examine the relationship between the predictions made by different learning algorithms and true posterior probabilities. We show that maximum margin methods such as boosted trees and boosted stumps push probability mass away from 0 and 1 yielding a characteristic sigmoid shaped distortion in the predicted probabilities. Models such as Naive Bayes, which make unrealistic independence assumptions, push probabilities toward 0 and 1. Other models such as neural nets and bagged trees do not have these biases and predict well calibrated probabilities. We experiment with two ways of correcting the biased probabilities predicted by some learning methods: Platt Scaling and Isotonic Regression. We qualitatively examine what kinds of distortions these calibration methods are suitable for and quantitatively examine how much data they need to be effective. The empirical results show that after calibration boosted trees, random forests, and SVMs predict the best probabilities.

An empirical evaluation of supervised learning in high dimensions

In this paper we perform an empirical evaluation of supervised learning on high-dimensional data. We evaluate performance on three metrics: accuracy, AUC, and squared loss and study the effect of increasing dimensionality on the performance of the learning algorithms. Our findings are consistent with previous studies for problems of relatively low dimension, but suggest that as dimensionality increases the relative performance of the learning algorithms changes. To our surprise, the method that performs consistently well across all dimensions is random forests, followed by neural nets, boosted trees, and SVMs.

Efficiently exploring architectural design spaces via predictive modeling

Architects use cycle-by-cycle simulation to evaluate design choices and understand tradeoffs and interactions among design parameters. Efficiently exploring exponential-size design spaces with many interacting parameters remains an open problem: the sheer number of experiments renders detailed simulation intractable. We attack this problem via an automated approach that builds accurate, confident predictive design-space models. We simulate sampled points, using the results to teach our models the function describing relationships among design parameters. The models produce highly accurate performance estimates for other points in the space, can be queried to predict performance impacts of architectural changes, and are very fast compared to simulation, enabling efficient discovery of tradeoffs among parameters in different regions. We validate our approach via sensitivity studies on memory hierarchy and CPU design spaces: our models generally predict IPC with only 1-2% error and reduce required simulation by two orders of magnitude. We also show the efficacy of our technique for exploring chip multiprocessor (CMP) design spaces: when trained on a 1% sample drawn from a CMP design space with 250K points and up to 55x performance swings among different system configurations, our models predict performance with only 4-5% error on average. Our approach combines with techniques to reduce time per simulation, achieving net time savings of three-four orders of magnitude.

Bridging the lexical chasm: statistical approaches to answer-finding

This paper investigates whether a machine can automatically learn the task of finding, within a large collection of candidate responses, the answers to questions. The learning process consists of inspecting a collection of answered questions and characterizing the relation between question and answer with a statistical model. For the purpose of learning this relation, we propose two sources of data: Usenet FAQ documents and customer service call-center dialogues from a large retail company. We will show that the task of “answer-finding” differs from both document retrieval and tradition question-answering, presenting challenges different from those found in these problems. The central aim of this work is to discover, through theoretical and empirical investigation, those statistical techniques best suited to the answer-finding problem.

Data mining in metric space: an empirical analysis of supervised learning performance criteria

Many criteria can be used to evaluate the performance of supervised learning. Different criteria are appropriate in different settings, and it is not always clear which criteria to use. A further complication is that learning methods that perform well on one criterion may not perform well on other criteria. For example, SVMs and boosting are designed to optimize accuracy, whereas neural nets typically optimize squared error or cross entropy. We conducted an empirical study using a variety of learning methods (SVMs, neural nets, k-nearest neighbor, bagged and boosted trees, and boosted stumps) to compare nine boolean classification performance metrics: Accuracy, Lift, F-Score, Area under the ROC Curve, Average Precision, Precision/Recall Break-Even Point, Squared Error, Cross Entropy, and Probability Calibration. Multidimensional scaling (MDS) shows that these metrics span a low dimensional manifold. The three metrics that are appropriate when predictions are interpreted as probabilities: squared error, cross entropy, and calibration, lay in one part of metric space far away from metrics that depend on the relative order of the predicted values: ROC area, average precision, break-even point, and lift. In between them fall two metrics that depend on comparing predictions to a threshold: accuracy and F-score. As expected, maximum margin methods such as SVMs and boosted trees have excellent performance on metrics like accuracy, but perform poorly on probability metrics such as squared error. What was not expected was that the margin methods have excellent performance on ordering metrics such as ROC area and average precision. We introduce a new metric, SAR, that combines squared error, accuracy, and ROC area into one metric. MDS and correlation analysis shows that SAR is centrally located and correlates well with other metrics, suggesting that it is a good general purpose metric to use when more specific criteria are not known.

Data-intensive Science: A New Paradigm for Biodiversity Studies

The increasing availability of massive volumes of scientific data requires new synthetic analysis techniques to explore and identify interesting patterns that are otherwise not apparent. For biodiversity studies, a “data-driven” approach is necessary because of the complexity of ecological systems, particularly when viewed at large spatial and temporal scales. Data-intensive science organizes large volumes of data from multiple sources and fields and then analyzes them using techniques tailored to the discovery of complex patterns in high-dimensional data through visualizations, simulations, and various types of model building. Through interpreting and analyzing these models, truly novel and surprising patterns that are “born from the data” can be discovered. These patterns provide valuable insight for concrete hypotheses about the underlying ecological processes that created the observed data. Data-intensive science allows scientists to analyze bigger and more complex systems efficiently, and complements more traditional scientific processes of hypothesis generation and experimental testing to refine our understanding of the natural world.