Martha White

Optimal reverse prediction: a unified perspective on supervised, unsupervised and semi-supervised learning

Training principles for unsupervised learning are often derived from motivations that appear to be independent of supervised learning. In this paper we present a simple unification of several supervised and unsupervised training principles through the concept of optimal reverse prediction: predict the inputs from the target labels, optimizing both over model parameters and any missing labels. In particular, we show how supervised least squares, principal components analysis, k-means clustering and normalized graph-cut can all be expressed as instances of the same training principle. Natural forms of semi-supervised regression and classification are then automatically derived, yielding semi-supervised learning algorithms for regression and classification that, surprisingly, are novel and refine the state of the art. These algorithms can all be combined with standard regularizers and made non-linear via kernels.

Partition Tree Weighting

This paper introduces the Partition Tree Weighting technique, an efficient meta-algorithm for piecewise stationary sources. The technique works by performing Bayesian model averaging over a large class of possible partitions of the data into locally stationary segments. It uses a prior, closely related to the Context Tree Weighting technique of Willems, that is well suited to data compression applications. Our technique can be applied to any coding distribution at an additional time and space cost only logarithmic in the sequence length. We provide a competitive analysis of the redundancy of our method, and explore its application in a variety of settings. The order of the redundancy and the complexity of our algorithm matches those of the best competitors available in the literature, and the new algorithm exhibits a superior complexity-performance trade-off in our experiments.

Scalable metric learning for co-embedding

We present a general formulation of metric learning for co-embedding, where the goal is to relate objects from different sets. The framework allows metric learning to be applied to a wide range of problems—including link prediction, relation learning, multi-label tagging and ranking—while allowing training to be reformulated as convex optimization. For training we provide a fast iterative algorithm that improves the scalability of existing metric learning approaches. Empirically, we demonstrate that the proposed method converges to a global optimum efficiently, and achieves competitive results in a variety of co-embedding problems such as multi-label classification and multi-relational prediction.

A general framework for reducing variance in agent evaluation

In this work, we present a unified, general approach to variance reduction in agent evaluation using machine learning to minimize variance. Evaluating an agent’s performance in a stochastic setting is necessary for agent development, scientific evaluation, and competitions. Traditionally, evaluation is done using Monte Carlo estimation (sample averages); the magnitude of the stochasticity in the domain or the high cost of sampling, however, can often prevent the approach from resulting in statistically significant conclusions. Recently, an advantage sum technique based on control variates has been proposed for constructing unbiased, low variance estimates of agent performance. The technique requires an expert to define a value function over states of the system, essentially a guess of the state’s unknown value. In this work, we propose learning this value function from past interactions between agents in some target population. Our learned value functions have two key advantages: they can be applied in domains where no expert value function is available and they can result in tuned evaluation for a specific population of agents (e.g., novice versus advanced agents). This work has three main contributions. First, we consolidate previous work in using control variates for variance reduction into one unified, general framework and summarize the connections between this previous work. Second, our framework makes variance reduction practically possible in any sequential decision making task where designing the expert value function is time-consuming, difficult or essentially impossible. We prove the optimality of our approach and extend the theoretical understanding of advantage sum estimators. In addition, we significantly extend the applicability of advantage sum estimators and discuss practical methods for using our framework in real-world scenarios. Finally, we provide low-variance estimators for three poker domains previously without variance reduction and improve strategy selection in the expert-level University of Alberta poker bot. This work is an elaboration of published work [White and Bowling, 2009].

Organizing experience: a deeper look at replay mechanisms for sample-based planning in continuous state domains

Model-based strategies for control are critical to obtain sample efficient learning. Dyna is a planning paradigm that naturally interleaves learning and planning, by simulating one-step experience to update the action-value function. This elegant planning strategy has been mostly explored in the tabular setting. The aim of this paper is to revisit sample-based planning, in stochastic and continuous domains with learned models. We first highlight the flexibility afforded by a model over Experience Replay (ER). Replay-based methods can be seen as stochastic planning methods that repeatedly sample from a buffer of recent agent-environment interactions and perform updates to improve data efficiency. We show that a model, as opposed to a replay buffer, is particularly useful for specifying which states to sample from during planning, such as predecessor states that propagate information in reverse from a state more quickly. We introduce a semi-parametric model learning approach, called Reweighted Experience Models (REMs), that makes it simple to sample next states or predecessors. We demonstrate that REM-Dyna exhibits similar advantages over replay-based methods in learning in continuous state problems, and that the performance gap grows when moving to stochastic domains, of increasing size.

Learning Sparse Representations in Reinforcement Learning with Sparse Coding

A variety of representation learning approaches have been investigated for reinforcement learning; much less attention, however, has been given to investigating the utility of sparse coding. Outside of reinforcement learning, sparse coding representations have been widely used, with non-convex objectives that result in discriminative representations. In this work, we develop a supervised sparse coding objective for policy evaluation. Despite the non-convexity of this objective, we prove that all local minima are global minima, making the approach amenable to simple optimization strategies. We empirically show that it is key to use a supervised objective, rather than the more straightforward unsupervised sparse coding approach. We compare the learned representations to a canonical fixed sparse representation, called tile-coding, demonstrating that the sparse coding representation outperforms a wide variety of tilecoding representations.

Erratum to: Scalable metric learning for co-embedding

Erratum to: Chapter 39 in: A. Appice et al. (Eds.) Machine Learning and Knowledge Discovery in Databases DOI: 10.1007/978-3-319-23528-8_39