In supervised learning, we commonly assume that training and test data are sampled from the same distribution. However, this assumption can be violated in practice and then standard machine learning techniques perform poorly. This paper focuses on revealing and improving the performance of Bayesian estimation when the training and test distributions are different. We formally analyze the asymptotic Bayesian generalization error and establish its upper bound under a very general setting. Our important finding is that lower order terms—which can be ignored in the absence of the distribution change— play an important role under the distribution change. We also propose a novel variant of stochastic complexity which can be used for choosing an appropriate model and hyper-parameters under a particular distribution change.

Off-policy reinforcement learning is aimed at efficiently reusing data samples gathered in the past, which is an essential problem for physically grounded AI as experiments are usually prohibitively expensive. A common approach is to use importance sampling techniques for compensating for the bias caused by the difference between data-sampling policies and the target policy. However, existing off-policy methods do not often take the variance of value function estimators explicitly into account and therefore their performance tends to be unstable. To cope with this problem, we propose using an adaptive importance sampling technique which allows us to actively control the trade-off between bias and variance. We further provide a method for optimally determining the trade-off parameter based on a variant of cross-validation. We demonstrate the usefulness of the proposed approach through simulations.

In machine learning problems, labeled data are often in short supply. One of the feasible solution for this problem is transfer learning. It can make use of the labeled data from other domain to discriminate those unlabeled data in the target domain. In this paper, we propose a transfer learning framework based on similarity matrix approximation to tackle such problems. Two practical algorithms are proposed, which are the label propagation and the similarity propagation. In these methods, we build a hybrid graph based on all available data. Then the information is transferred cross domains through alternatively constructing the similarity matrix for different part of the graph. Among all related methods, similarity propagation approach can make maximum use of all available similarity information across domains. This leads to more efficient transfer and better learning result. The experiment on real world text mining applications demonstrates the promise and effectiveness of our algorithms. 1

We present and analyze an agnostic active learning algorithm that works without keeping a version space. This is unlike all previous approaches where a restricted set of candidate hypotheses is maintained throughout learning, and only hypotheses from this set are ever returned. By avoiding this version space approach, our algorithm sheds the computational burden and brittleness associated with maintaining version spaces, yet still allows for substantial improvements over supervised learning for classification. 1

Optimally designing the location of training input points (active learning) and choosing the best model (model selection) are two important components of supervised learning and have been studied extensively. However, these two issues seem to have been investigated separately as two independent problems. If training input points and models are simultaneously optimized, the generalization performance would be further improved. In this paper, we propose a new approach called ensemble active learning for solving the problems of active learning and model selection at the same time. We demonstrate by numerical experiments that the proposed method compares favorably with alternative approaches such as iteratively performing active learning and model selection in a sequential manner.

Abstract. Direct policy search is a promising reinforcement learning framework in particular for controlling in continuous, high-dimensional systems such as anthropomorphic robots. Policy search often requires a large number of samples for obtaining a stable policy update estimator due to its high flexibility. However, this is prohibitive when the sampling cost is expensive. In this paper, we extend an EM-based policy search method so that previously collected samples can be efficiently reused. The usefulness of the proposed method, called Reward-weighted Regression with sample Reuse (R 3), is demonstrated through a robot learning experiment. 1

Given sets of observations of training and test data, we consider the problem of re-weighting the training data such that its distribution more closely matches that of the test data. We achieve this goal by matching covariate distributions between training and test sets in a high dimensional feature space (specifically, a reproducing kernel Hilbert space). This approach does not require distribution estimation. Instead, the sample weights are obtained by a simple quadratic programming procedure. We provide a uniform convergence bound on the distance between the reweighted training feature mean and the test feature mean, a transductive bound on the expected loss of an algorithm trained on the reweighted data, and a connection to single class SVMs. While our method is designed to deal with the case of simple covariate shift (in the sense of Chapter??), we have also found benefits for sample selection bias on the labels. Our correction procedure yields its greatest and most consistent advantages when the learning algorithm returns a classifier/regressor that is “simpler” than the data might suggest.

Abstract—The aim of transfer learning is to improve prediction accuracy on a target task by exploiting the training examples for tasks that are related to the target one. Transfer learning has received more attention in recent years, because this technique is considered to be helpful in reducing the cost of labeling. In this paper, we propose a very simple approach to transfer learning: TrBagg, which is the extension of bagging. TrBagg is composed of two stages: Many weak classifiers are first generated as in standard bagging, and these classifiers are then filtered based on their usefulness for the target task. This simplicity makes it easy to work reasonably well without severe tuning of learning parameters. Further, our algorithm equips an algorithmic scheme to avoid negative transfer. We applied TrBagg to personalized tag prediction tasks for social bookmarks Our approach has several convenient characteristics for this task such as adaptation to multiple tasks with low computational cost. Keywords-transfer learning; bagging, ensemble learning, personalization, collaborative tagging I.

Divergence estimators based on direct approximation of density-ratios without going through separate approximation of numerator and denominator densities have been successfully applied to machine learning tasks that involve distribution comparison such as outlier detection, transfer learning, and two-sample homogeneity test. However, since density-ratio functions often possess high fluctuation, divergence estimation is still a challenging task in practice. In this paper, we propose to use relative divergences for distribution comparison, which involves approximation of relative density-ratios. Since relative density-ratios are always smoother than corresponding ordinary density-ratios, our proposed method is favorable in terms of the non-parametric convergence speed. Furthermore, we show that the proposed divergence estimator has asymptotic variance independent of the model complexity under a parametric setup, implying that the proposed estimator hardly overfits even with complex models. Through experiments, we demonstrate the usefulness of the proposed approach. 1

Abstract. One solution to the lack of label problem is to exploit transfer learning, whereby one acquires knowledge from source-domains to improvethe learning performance in the target-domain. Themain challenge is that the source and target domains may have different distributions. An open problem is how to select the available models (including algorithms and parameters) and importantly, abundance of source-domain data, through statistically reliable methods, thus making transfer learning practical and easy-to-use for real-world applications. To address this challenge, one needs to take into account the difference in both marginal and conditional distributions in the same time, but not just one of them. In this paper, we formulate a new criterion to overcome “double ” distributionshiftandpresentapracticalapproach“TransferCrossValidation” (TrCV) to select both models and data in a cross validation framework, optimized for transfer learning. The idea is to use density ratio weighting to overcome the difference in marginal distributions and propose a “reverse validation ” procedure to quantify how well a model approximates the true conditional distribution of target-domain. The usefulness of TrCV is demonstrated on different cross-domain tasks, including wine quality evaluation, web-user ranking and text categorization. The experiment results show that the proposed method outperforms both traditional cross-validation and one state-of-the-art method which only considersmarginaldistributionshift.Thesoftwareanddatasetsareavailable from the authors. 1