Traffic signal control is an important problem in urban mobility with a significant potential of economic and environmental impact. While there is a growing interest in Reinforcement Learning (RL) for traffic control, the work so far has focussed on learning through interactions which, in practice, is costly. Instead, real experience data on traffic is available and could be exploited at minimal costs. Recent progress in offline or batch RL has enabled just that. Model-based offline RL methods, in particular, have been shown to generalize to the experience data much better than others. We build a model-based learning framework, A-DAC, which infers a Markov Decision Process (MDP) from dataset with pessimistic costs built in to deal with data uncertainties. The costs are modeled through an adaptive shaping of rewards in the MDP which provides better regularization of data compared to the prior related work. A-DAC is evaluated on a complex signalized roundabout using multiple datasets varying in size and in batch collection policy. The evaluation results show that it is possible to build high performance control policies in a data efficient manner using simplistic batch collection policies.

Conventional machine learning (ML) relies heavily on manual design from machine learning experts to decide learning tasks, data, models, optimization algorithms, and evaluation metrics, which is labor-intensive, time-consuming, and cannot learn autonomously like humans. In education science, self-directed learning, where human learners select learning tasks and materials on their own without requiring hands-on guidance, has been shown to be more effective than passive teacher-guided learning. Inspired by the concept of self-directed human learning, we introduce the principal concept of Self-directed Machine Learning (SDML) and propose a framework for SDML. Specifically, we design SDML as a self-directed learning process guided by self-awareness, including internal awareness and external awareness. Our proposed SDML process benefits from self task selection, self data selection, self model selection, self optimization strategy selection and self evaluation metric selection through self-awareness without human guidance. Meanwhile, the learning performance of the SDML process serves as feedback to further improve self-awareness. We propose a mathematical formulation for SDML based on multi-level optimization. Furthermore, we present case studies together with potential applications of SDML, followed by discussing future research directions. We expect that SDML could enable machines to conduct human-like self-directed learning and provide a new perspective towards artificial general intelligence.

Affordance refers to the perception of possible actions allowed by an object. Despite its relevance to human-computer interaction, no existing theory explains the mechanisms that underpin affordance-formation; that is, how affordances are discovered and adapted via interaction. We propose an integrative theory of affordance-formation based on the theory of reinforcement learning in cognitive sciences. The key assumption is that users learn to associate promising motor actions to percepts via experience when reinforcement signals (success/failure) are present. They also learn to categorize actions (e.g., "rotating" a dial), giving them the ability to name and reason about affordance. Upon encountering novel widgets, their ability to generalize these actions determines their ability to perceive affordances. We implement this theory in a virtual robot model, which demonstrates human-like adaptation of affordance in interactive widgets tasks. While its predictions align with trends in human data, humans are able to adapt affordances faster, suggesting the existence of additional mechanisms.

Reward design is a fundamental problem in reinforcement learning (RL). A misspecified or poorly designed reward can result in low sample efficiency and undesired behaviors. In this paper, we propose the idea of programmatic reward design, i.e. using programs to specify the reward functions in RL environments. Programs allow human engineers to express sub-goals and complex task scenarios in a structured and interpretable way. The challenge of programmatic reward design, however, is that while humans can provide the high-level structures, properly setting the low-level details, such as the right amount of reward for a specific sub-task, remains difficult. A major contribution of this paper is a probabilistic framework that can infer the best candidate programmatic reward function from expert demonstrations. Inspired by recent generative-adversarial approaches, our framework searches for the most likely programmatic reward function under which the optimally generated trajectories cannot be differentiated from the demonstrated trajectories. Experimental results show that programmatic reward functionslearned using this framework can significantly outperform those learned using existing reward learning algo-rithms, and enable RL agents to achieve state-of-the-artperformance on highly complex tasks.

The study of generalisation in deep Reinforcement Learning (RL) aims to produce RL algorithms whose policies generalise well to novel unseen situations at deployment time, avoiding overfitting to their training environments. Tackling this is vital if we are to deploy reinforcement learning algorithms in real world scenarios, where the environment will be diverse, dynamic and unpredictable. This survey is an overview of this nascent field. We provide a unifying formalism and terminology for discussing different generalisation problems, building upon previous works. We go on to categorise existing benchmarks for generalisation, as well as current methods for tackling the generalisation problem. Finally, we provide a critical discussion of the current state of the field, including recommendations for future work. Among other conclusions, we argue that taking a purely procedural content generation approach to benchmark design is not conducive to progress in generalisation, we suggest fast online adaptation and tackling RL-specific problems as some areas for future work on methods for generalisation, and we recommend building benchmarks in underexplored problem settings such as offline RL generalisation and reward-function variation.

Offline policy evaluation (OPE) is an indispensable component of the offline reinforcement learning (RL), which is a variant of reinforcement learning with special emphasis on cost-sensitive real-life applications (Levine et al., 2020), such as autonomous vehicles, finance, healthcare and molecular discovery. Almost all the offline RL algorithms involve their own hyperparameters. For example, if we employ neural networks in policy learning, we have to at least decide the network topology (e.g., number of neurons and layers, to use the residual connection or not, to use the dense connection or the convolution), the activation functions, regularization weights and the optimizers (e.g., SGD or Adam with their own hyperparameter choices). The choice of the models such as neural network is also considered to be a hyperparameters. OPE allows us to optimize or validate the choices over these hyperparameters based only on offline datasets, i.e., without access to environment simulators.

We study reward design strategies for incentivizing a reinforcement learning agent to adopt a policy from a set of admissible policies. The goal of the reward designer is to modify the underlying reward function cost-efficiently while ensuring that any approximately optimal deterministic policy under the new reward function is admissible and performs well under the original reward function. This problem can be viewed as a dual to the problem of optimal reward poisoning attacks: instead of forcing an agent to adopt a specific policy, the reward designer incentivizes an agent to avoid taking actions that are inadmissible in certain states. Perhaps surprisingly, and in contrast to the problem of optimal reward poisoning attacks, we first show that the reward design problem for admissible policy teaching is computationally challenging, and it is NP-hard to find an approximately optimal reward modification. We then proceed by formulating a surrogate problem whose optimal solution approximates the optimal solution to the reward design problem in our setting, but is more amenable to optimization techniques and analysis. For this surrogate problem, we present characterization results that provide bounds on the value of the optimal solution. Finally, we design a local search algorithm to solve the surrogate problem and showcase its utility using simulation-based experiments.

In recent years, voice-based virtual assistants such as Google Assistant and Amazon Alexa have grown popular. This has presented both potential and challenges for natural language understanding (NLU) systems. These devices' production systems are often trained by supervised learning and rely significantly on annotated data. But, data annotation is costly and time-consuming. Furthermore, model updates using offline supervised learning can take long and miss trending requests. In the underlying architecture of voice-based virtual assistants, the NLU model often categorizes user requests into hypotheses for downstream applications to fulfill.

We consider the joint design and control of discrete-time stochastic dynamical systems over a finite time horizon. We formulate the problem as a multi-step optimization problem under uncertainty seeking to identify a system design and a control policy that jointly maximize the expected sum of rewards collected over the time horizon considered. The transition function, the reward function and the policy are all parametrized, assumed known and differentiable with respect to their parameters. We then introduce a deep reinforcement learning algorithm combining policy gradient methods with model-based optimization techniques to solve this problem. In essence, our algorithm iteratively approximates the gradient of the expected return via Monte-Carlo sampling and automatic differentiation and takes projected gradient ascent steps in the space of environment and policy parameters. This algorithm is referred to as Direct Environment and Policy Search (DEPS). We assess the performance of our algorithm in three environments concerned with the design and control of a mass-spring-damper system, a small-scale off-grid power system and a drone, respectively. In addition, our algorithm is benchmarked against a state-of-the-art deep reinforcement learning algorithm used to tackle joint design and control problems. We show that DEPS performs at least as well or better in all three environments, consistently yielding solutions with higher returns in fewer iterations. Finally, solutions produced by our algorithm are also compared with solutions produced by an algorithm that does not jointly optimize environment and policy parameters, highlighting the fact that higher returns can be achieved when joint optimization is performed.

We suggest a simple practical method to combine the human and artificial intelligence to both learn best investment practices of fund managers, and provide recommendations to improve them. Our approach is based on a combination of Inverse Reinforcement Learning (IRL) and RL. First, the IRL component learns the intent of fund managers as suggested by their trading history, and recovers their implied reward function. At the second step, this reward function is used by a direct RL algorithm to optimize asset allocation decisions. We show that our method is able to improve over the performance of individual fund managers.