As the world seeks to become more sustainable, intelligent solutions are needed to increase the penetration of renewable energy. In this paper, the model-free deep reinforcement learning algorithm Rainbow Deep Q-Networks is used to control a battery in a small microgrid to perform energy arbitrage and more efficiently utilise solar and wind energy sources. The grid operates with its own demand and renewable generation based on a dataset collected at Keele University, as well as using dynamic energy pricing from a real wholesale energy market. Four scenarios are tested including using demand and price forecasting produced with local weather data. The algorithm and its subcomponents are evaluated against two continuous control benchmarks with Rainbow able to outperform all other method. This research shows the importance of using the distributional approach for reinforcement learning when working with complex environments and reward functions, as well as how it can be used to visualise and contextualise the agent's behaviour for real-world applications.

Today's advanced Reinforcement Learning algorithms produce black-box policies, that are often difficult to interpret and trust for a person. We introduce a policy distilling algorithm, building on the CN2 rule mining algorithm, that distills the policy into a rule-based decision system. At the core of our approach is the fact that an RL process does not just learn a policy, a mapping from states to actions, but also produces extra meta-information, such as action values indicating the quality of alternative actions. This meta-information can indicate whether more than one action is near-optimal for a certain state. We extend CN2 to make it able to leverage knowledge about equally-good actions to distill the policy into fewer rules, increasing its interpretability by a person. Then, to ensure that the rules explain a valid, non-degenerate policy, we introduce a refinement algorithm that fine-tunes the rules to obtain good performance when executed in the environment. We demonstrate the applicability of our algorithm on the Mario AI benchmark, a complex task that requires modern reinforcement learning algorithms including neural networks. The explanations we produce capture the learned policy in only a few rules, that allow a person to understand what the black-box agent learned.

In this paper, we propose a realistic multiple dynamic pricing approach to demand response in the retail market. First, an adaptive clustering-based customer segmentation framework is proposed to categorize customers into different groups to enable the effective identification of usage patterns. Second, customized demand models with important market constraints which capture the price-demand relationship explicitly, are developed for each group of customers to improve the model accuracy and enable meaningful pricing. Third, the multiple pricing based demand response is formulated as a profit maximization problem subject to realistic market constraints. The overall aim of the proposed scalable and practical method aims to achieve 'right' prices for 'right' customers so as to benefit various stakeholders in the system such as grid operators, customers and retailers. The proposed multiple pricing framework is evaluated via simulations based on real-world datasets.

In multi-agent reinforcement learning, the inherent non-stationarity of the environment caused by other agents' actions posed significant difficulties for an agent to learn a good policy independently. One way to deal with non-stationarity is agent modeling, by which the agent takes into consideration the influence of other agents' policies. Most existing work relies on predicting other agents' actions or goals, or discriminating between their policies. However, such modeling fails to capture the similarities and differences between policies simultaneously and thus cannot provide useful information when generalizing to unseen policies. To address this, we propose a general method to learn representations of other agents' policies via the joint-action distributions sampled in interactions. The similarities and differences between policies are naturally captured by the policy distance inferred from the joint-action distributions and deliberately reflected in the learned representations. Agents conditioned on the policy representations can well generalize to unseen agents. We empirically demonstrate that our method outperforms existing work in multi-agent tasks when facing unseen agents.

We introduce a new type of programming challenge called programming puzzles, as an objective and comprehensive evaluation of program synthesis, and release an open-source dataset of Python Programming Puzzles (P3). Each puzzle is defined by a short Python program $f$, and the goal is to find an input $x$ which makes $f$ output "True". The puzzles are objective in that each one is specified entirely by the source code of its verifier $f$, so evaluating $f(x)$ is all that is needed to test a candidate solution $x$. They do not require an answer key or input/output examples, nor do they depend on natural language understanding. The dataset is comprehensive in that it spans problems of a range of difficulties and domains, ranging from trivial string manipulation problems that are immediately obvious to human programmers (but not necessarily to AI), to classic programming puzzles (e.g., Towers of Hanoi), to interview/competitive-programming problems (e.g., dynamic programming), to longstanding open problems in algorithms and mathematics (e.g., factoring). The objective nature of P3 readily supports self-supervised bootstrapping. We develop baseline enumerative program synthesis and GPT-3 solvers that are capable of solving easy puzzles -- even without access to any reference solutions -- by learning from their own past solutions. Based on a small user study, we find puzzle difficulty to correlate between human programmers and the baseline AI solvers.

Their proposed Decision Transformer outputs optimal actions by leveraging a causally masked transformer and can generate future actions with desired returns. Moreover, despite Decision Transformer's relative simplicity, the proposed framework matches or outperforms the performance of state-of-the-art model-free offline RL baselines on Atari, OpenAI Gym, and Key-to-Door tasks. Transformer architectures are able to efficiently model sequential data, and their self-attention mechanism allows the layer to assign "credit" by implicitly forming state-return associations via maximizing the dot product of the query and key vectors. Transformers can thus function effectively in the presence of sparse or distracting rewards. Previous studies have also shown that transformers can model a wide distribution of behaviours, enabling better generalization and transfer abilities.

Great! Check your inbox and click the link to confirm your subscription Error! Please enter a valid email address!

DeepMind, Google's artificial intelligence subsidiary, has announced the development of its new AndroidEnvtool, an open source reinforcement learning (RL) platform for the Android ecosystem. It allows RL models to interact with a wide variety of applications and services that Android users use on a daily basis thanks to an adapted interface. AndroidEnv is a tool designed by DeepMind to enable reinforcement learning agents to interact with a wide range of applications and services offered by Google. The deployed agents learn to interact with mobile applications based, unlike Deep Learning, on learning from the agent's own experience. The agents interact with the applications by using the applications as humans would: tapping, finger swiping are simulated.

Conveying complex objectives to reinforcement learning (RL) agents can often be difficult, involving meticulous design of reward functions that are sufficiently informative yet easy enough to provide. Human-in-the-loop RL methods allow practitioners to instead interactively teach agents through tailored feedback; however, such approaches have been challenging to scale since human feedback is very expensive. In this work, we aim to make this process more sample- and feedback-efficient. We present an off-policy, interactive RL algorithm that capitalizes on the strengths of both feedback and off-policy learning. Specifically, we learn a reward model by actively querying a teacher's preferences between two clips of behavior and use it to train an agent. To enable off-policy learning, we relabel all the agent's past experience when its reward model changes. We additionally show that pre-training our agents with unsupervised exploration substantially increases the mileage of its queries. We demonstrate that our approach is capable of learning tasks of higher complexity than previously considered by human-in-the-loop methods, including a variety of locomotion and robotic manipulation skills. We also show that our method is able to utilize real-time human feedback to effectively prevent reward exploitation and learn new behaviors that are difficult to specify with standard reward functions.

Recent theoretical work studies sample-efficient reinforcement learning (RL) extensively in two settings: learning interactively in the environment (online RL), or learning from an offline dataset (offline RL). However, existing algorithms and theories for learning near-optimal policies in these two settings are rather different and disconnected. Towards bridging this gap, this paper initiates the theoretical study of policy finetuning, that is, online RL where the learner has additional access to a "reference policy" $\mu$ close to the optimal policy $\pi_\star$ in a certain sense. We consider the policy finetuning problem in episodic Markov Decision Processes (MDPs) with $S$ states, $A$ actions, and horizon length $H$. We first design a sharp offline reduction algorithm -- which simply executes $\mu$ and runs offline policy optimization on the collected dataset -- that finds an $\varepsilon$ near-optimal policy within $\widetilde{O}(H^3SC^\star/\varepsilon^2)$ episodes, where $C^\star$ is the single-policy concentrability coefficient between $\mu$ and $\pi_\star$. This offline result is the first that matches the sample complexity lower bound in this setting, and resolves a recent open question in offline RL. We then establish an $\Omega(H^3S\min\{C^\star, A\}/\varepsilon^2)$ sample complexity lower bound for any policy finetuning algorithm, including those that can adaptively explore the environment. This implies that -- perhaps surprisingly -- the optimal policy finetuning algorithm is either offline reduction or a purely online RL algorithm that does not use $\mu$. Finally, we design a new hybrid offline/online algorithm for policy finetuning that achieves better sample complexity than both vanilla offline reduction and purely online RL algorithms, in a relaxed setting where $\mu$ only satisfies concentrability partially up to a certain time step.