The reinforcement learning paradigm is a popular way to address problems that have only limited environmental feedback, rather than correctly labeled examples, as is common in other machine learning contexts. While significant progress has been made to improve learning in a single task, the idea of transfer learning has only recently been applied to reinforcement learning tasks. The core idea of transfer is that experience gained in learning to perform one task can help improve learning performance in a related, but different, task. In this article we present a framework that classifies transfer learning methods in terms of their capabilities and goals, and then use it to survey the existing literature, as well as to suggest future directions for transfer learning work.

We propose the use of statistical relational learning, and in particular the formalism of Markov Logic Networks, for transfer in reinforcement learning. Our goal is to extract relational knowledge from a source task and use it to speed up learning in a related target task. We do so by learning a Markov Logic Network that describes the source-task Q-function, and then using it for decision making in the early learning stages of the target task. Through experiments in the RoboCup simulated-soccer domain, we show that this approach can provide a substantial performance benefit in the target task.

Abstract. Transfer learning is the improvement of learning in a new task through the transfer of knowledge from a related task that has already been learned. While most machine learning algorithms are designed to address single tasks, the development of algorithms that facilitate transfer learning is a topic of ongoing interest in the machine-learning community. This chapter provides an introduction to the goals, formulations, and challenges of transfer learning. It surveys current research in this area, giving an overview of the state of the art and outlining the open problems. The survey covers transfer in both inductive learning and reinforcement learning, and discusses the issues of negative transfer and task mapping in depth.

A central goal of transfer learning is to enable learning when training data from the domain of interest is limited. Yet, work on transfer across relational domains has so far focused on the case where there is a significant amount of target data. This paper bridges this gap by studying transfer when the amount of target data is minimal and consists of information about just a handful of entities. In the extreme case, only a single entity is known. We present the SR2LR algorithm that finds an effective mapping of predicates from a source model to the target domain in this setting and thus renders preexisting knowledge useful to the target task. We demonstrate SR2LR’s effectiveness in three benchmark relational domains on social interactions and study its behavior as information about an increasing number of entities becomes available.

A Relational Instance-Based Policy can be defined as an action policy described following a relational instancebased learning approach. The policy is represented with a set of state-goal-action tuples in some form of predicate logic and a distance metric: whenever the planner is in a state trying to reach a goal, the next action to execute is computed as the action associated to the closest state-goal pair in that set. In this work, the representation language is relational, following the ideas of Relational Reinforcement Learning. The policy to transfer (the set of state-goal-action tuples) is generated with a planning system solving optimally simple source problems. The target problems are defined in the same planning domain, have different initial and goal states to the source problems, and could be much more complex. We show that the transferred policy can solve similar problems to the ones used to learn it, but also more complex problems. In fact, the policy learned outperforms the planning system used to generate the initial state-action pairs in two ways: it is faster and scales up better.

Imitation learning refers to the problem of learning how to behave by observing a teacher in action. We consider imitation learning in relational domains, in which there is a varying number of objects and relations among them. In prior work, simple relational policies are learned by viewing imitation learning as supervised learning of a function from states to actions. For propositional worlds, functional gradient methods have been proved to be beneficial. They are simpler to implement than most existing methods, more efficient, more naturally satisfy common constraints on the cost function, and better represent our prior beliefs about the form of the function. Building on recent generalizations of functional gradient boosting to relational representations, we implement a functional gradient boosting approach to imitation learning in relational domains. In particular, given a set of traces from the human teacher, our system learns a policy in the form of a set of relational regression trees that additively approximate the functional gradients. The use of multiple additive trees combined with relational representation allows for learning more expressive policies than what has been done before. We demonstrate the usefulness of our approach in several different domains. 1

Abstract Transfer learning seeks to leverage previously learned tasks to achieve faster learning in a new task. In this paper, we consider transfer learning in the context of related but distinct Reinforcement Learning (RL) problems. In particular, our RL problems are derived from Semi-Markov Decision Processes (SMDPs) that share the same transition dynamics but have different reward functions that are linear in a set of reward features. We formally define the transfer learning problem in the context of RL as learning an efficient algorithm to solve any SMDP drawn from a fixed distribution after experiencing a finite number of them. Furthermore, we introduce an online algorithm to solve this problem, Variable-Reward Reinforcement Learning (VRRL), that compactly stores the optimal value functions for several SMDPs, and uses them to optimally initialize the value function for a new SMDP. We generalize our method to a hierarchical RL setting where the different SMDPs share the same task hierarchy. Our experimental results in a simplified real-time strategy domain show that significant transfer learning occurs in both flat and hierarchical settings. Transfer is especially effective in the hierarchical setting where the overall value functions are decomposed into subtask value functions which are more widely amenable to transfer across different SMDPs. Keywords Hierarchical reinforcement learning · Transfer learning · Average-reward learning · Multi-criteria learning

Reinforcement learning (RL) is a machine learning technique with strong links to natural learning. However, it shares several “unnatural ” limitations with many other successful machine learning algorithms. RL agents are not typically able to take advice or to adjust to new situations beyond the specific problem they are asked to learn. Due to limitations like these, RL remains slower and less adaptable than natural learning. Our recent work focuses on extending RL to include the naturally inspired abilities of advice taking and transfer learning. Through experiments in the RoboCup domain, we show that doing so can make RL faster and more adaptable.

Abstract not found

Maclin, to my committee members, and to the Machine Learning Group at the University of Wisconsin-Madison.