Q-learning has long been one of the most popular reinforcement learning algorithms, and theoretical analysis of Q-learning has been an active research topic for decades. Although researches on asymptotic convergence analysis of Q-learning have a long tradition, non-asymptotic convergence has only recently come under active study. The main goal of this paper is to investigate new finite-time analysis of asynchronous Q-learning under Markovian observation models via a control system viewpoint. In particular, we introduce a discrete-time time-varying switching system model of Q-learning with diminishing step-sizes for our analysis, which significantly improves recent development of the switching system analysis with constant step-sizes, and leads to \(\mathcal{O}\left( \sqrt{\frac{\log k}{k}} \right)\) convergence rate that is comparable to or better than most of the state of the art results in the literature. In the mean while, a technique using the similarly transformation is newly applied to avoid the difficulty in the analysis posed by diminishing step-sizes. The proposed analysis brings in additional insights, covers different scenarios, and provides new simplified templates for analysis to deepen our understanding on Q-learning via its unique connection to discrete-time switching systems.
With the development of artificial intelligence,more and more financial practitioners apply deep reinforcement learning to financial trading strategies.However,It is difficult to extract accurate features due to the characteristics of considerable noise,highly non-stationary,and non-linearity of single-scale time series,which makes it hard to obtain high returns.In this paper,we extract a multi-scale feature matrix on multiple time scales of financial time series,according to the classic financial theory-Chan Theory,and put forward to an approach of multi-scale stroke deep deterministic policy gradient reinforcement learning model(MSSDDPG)to search for the optimal trading strategy.We carried out experiments on the datasets of the Dow Jones,S&P 500 of U.S. stocks, and China's CSI 300,SSE Composite,evaluate the performance of our approach compared with turtle trading strategy, Deep Q-learning(DQN)reinforcement learning strategy,and deep deterministic policy gradient (DDPG) reinforcement learning strategy.The result shows that our approach gets the best performance in China CSI 300,SSE Composite,and get an outstanding result in Dow Jones,S&P 500 of U.S.
Federated Learning (FL) allows for collaboratively aggregating learned information across several computing devices and sharing the same amongst them, thereby tackling issues of privacy and the need of huge bandwidth. FL techniques generally use a central server or cloud for aggregating the models received from the devices. Such centralized FL techniques suffer from inherent problems such as failure of the central node and bottlenecks in channel bandwidth. When FL is used in conjunction with connected robots serving as devices, a failure of the central controlling entity can lead to a chaotic situation. This paper describes a mobile agent based paradigm to decentralize FL in multi-robot scenarios. Using Webots, a popular free open-source robot simulator, and Tartarus, a mobile agent platform, we present a methodology to decentralize federated learning in a set of connected robots. With Webots running on different connected computing systems, we show how mobile agents can perform the task of Decentralized Federated Reinforcement Learning (dFRL). Results obtained from experiments carried out using Q-learning and SARSA by aggregating their corresponding Q-tables, show the viability of using decentralized FL in the domain of robotics. Since the proposed work can be used in conjunction with other learning algorithms and also real robots, it can act as a vital tool for the study of decentralized FL using heterogeneous learning algorithms concurrently in multi-robot scenarios.
We consider a multi-source hybrid automatic repeat request (HARQ) based system, where a transmitter sends status update packets of random arrival (i.e., uncontrollable sampling) and generate-atwill (i.e., controllable sampling) sources to a destination through an error-prone channel. We develop transmission scheduling policies to minimize the average number of transmissions subject to an average age of information (AoI) constraint. First, we consider known environment (i.e., known system statistics) and develop a near-optimal deterministic transmission policy and a low-complexity dynamic transmission (LC-DT) policy. The former policy is derived by casting the main problem into a constrained Markov decision process (CMDP) problem, which is then solved using the Lagrangian relaxation, relative value iteration algorithm, and bisection. The LC-DT policy is developed via the drift-plus-penalty (DPP) method by transforming the main problem into a sequence of per-slot problems. Finally, we consider unknown environment and devise a learning-based transmission policy by relaxing the CMDP problem into an MDP problem using the DPP method and then adopting the deep Q-learning algorithm. Numerical results show that the proposed policies achieve near-optimal performance and illustrate the benefits of HARQ in status updating.
This paper proposes new, end-to-end deep reinforcement learning algorithms for learning two-player zero-sum Markov games. Different from prior efforts on training agents to beat a fixed set of opponents, our objective is to find the Nash equilibrium policies that are free from exploitation by even the adversarial opponents. We propose (a) Nash-DQN algorithm, which integrates the deep learning techniques from single DQN into the classic Nash Q-learning algorithm for solving tabular Markov games; (b) Nash-DQN-Exploiter algorithm, which additionally adopts an exploiter to guide the exploration of the main agent. We conduct experimental evaluation on tabular examples as well as various two-player Atari games. Our empirical results demonstrate that (i) the policies found by many existing methods including Neural Fictitious Self Play and Policy Space Response Oracle can be prone to exploitation by adversarial opponents; (ii) the output policies of our algorithms are robust to exploitation, and thus outperform existing methods.
We investigate Nash equilibrium learning in a competitive Markov Game (MG) environment, where multiple agents compete, and multiple Nash equilibria can exist. In particular, for an oligopolistic dynamic pricing environment, exact Nash equilibria are difficult to obtain due to the curse-of-dimensionality. We develop a new model-free method to find approximate Nash equilibria. Gradient-free black box optimization is then applied to estimate $\epsilon$, the maximum reward advantage of an agent unilaterally deviating from any joint policy, and to also estimate the $\epsilon$-minimizing policy for any given state. The policy-$\epsilon$ correspondence and the state to $\epsilon$-minimizing policy are represented by neural networks, the latter being the Nash Policy Net. During batch update, we perform Nash Q learning on the system, by adjusting the action probabilities using the Nash Policy Net. We demonstrate that an approximate Nash equilibrium can be learned, particularly in the dynamic pricing domain where exact solutions are often intractable.
This paper defines the problem of optimizing the downlink multi-user multiple input, single output (MU-MISO) sum-rate for ground users served by an aerial reconfigurable intelligent surface (ARIS) that acts as a relay to the terrestrial base station. The deep deterministic policy gradient (DDPG) is proposed to calculate the optimal active beamforming matrix at the base station and the phase shifts of the reflecting elements at the ARIS to maximize the data rate. Simulation results show the superiority of the proposed scheme when compared to deep Q-learning (DQL) and baseline approaches.
In lifelong learning, an agent learns throughout its entire life without resets, in a constantly changing environment, as we humans do. Consequently, lifelong learning comes with a plethora of research problems such as continual domain shifts, which result in non-stationary rewards and environment dynamics. These non-stationarities are difficult to detect and cope with due to their continuous nature. Therefore, exploration strategies and learning methods are required that are capable of tracking the steady domain shifts, and adapting to them. We propose Reactive Exploration to track and react to continual domain shifts in lifelong reinforcement learning, and to update the policy correspondingly. To this end, we conduct experiments in order to investigate different exploration strategies. We empirically show that representatives of the policy-gradient family are better suited for lifelong learning, as they adapt more quickly to distribution shifts than Q-learning. Thereby, policy-gradient methods profit the most from Reactive Exploration and show good results in lifelong learning with continual domain shifts. Our code is available at: https://github.com/ml-jku/reactive-exploration.
TSP is a classical NP-hard combinatorial optimization problem with many practical variants. LKH is one of the state-of-the-art local search algorithms for the TSP. LKH-3 is a powerful extension of LKH that can solve many TSP variants. Both LKH and LKH-3 associate a candidate set to each city to improve the efficiency, and have two different methods, $\alpha$-measure and POPMUSIC, to decide the candidate sets. In this work, we first propose a Variable Strategy Reinforced LKH (VSR-LKH) algorithm, which incorporates three reinforcement learning methods (Q-learning, Sarsa, Monte Carlo) with LKH, for the TSP. We further propose a new algorithm called VSR-LKH-3 that combines the variable strategy reinforcement learning method with LKH-3 for typical TSP variants, including the TSP with time windows (TSPTW) and Colored TSP (CTSP). The proposed algorithms replace the inflexible traversal operations in LKH and LKH-3 and let the algorithms learn to make a choice at each search step by reinforcement learning. Both LKH and LKH-3, with either $\alpha$-measure or POPMUSIC, can be significantly improved by our methods. Extensive experiments on 236 widely-used TSP benchmarks with up to 85,900 cities demonstrate the excellent performance of VSR-LKH. VSR-LKH-3 also significantly outperforms the state-of-the-art heuristics for TSPTW and CTSP.
Mobile notification systems play a major role in a variety of applications to communicate, send alerts and reminders to the users to inform them about news, events or messages. In this paper, we formulate the near-real-time notification decision problem as a Markov Decision Process where we optimize for multiple objectives in the rewards. We propose an end-to-end offline reinforcement learning framework to optimize sequential notification decisions. We address the challenge of offline learning using a Double Deep Q-network method based on Conservative Q-learning that mitigates the distributional shift problem and Q-value overestimation. We illustrate our fully-deployed system and demonstrate the performance and benefits of the proposed approach through both offline and online experiments.
Stock portfolio optimization is the process of continuous reallocation of funds to a selection of stocks. This is a particularly well-suited problem for reinforcement learning, as daily rewards are compounding and objective functions may include more than just profit, e.g., risk and sustainability. We developed a novel utility function with the Sharpe ratio representing risk and the environmental, social, and governance score (ESG) representing sustainability. We show that a state-of-the-art policy gradient method - multi-agent deep deterministic policy gradients (MADDPG) - fails to find the optimum policy due to flat policy gradients and we therefore replaced gradient descent with a genetic algorithm for parameter optimization. We show that our system outperforms MADDPG while improving on deep Q-learning approaches by allowing for continuous action spaces. Crucially, by incorporating risk and sustainability criteria in the utility function, we improve on the state-of-the-art in reinforcement learning for portfolio optimization; risk and sustainability are essential in any modern trading strategy and we propose a system that does not merely report these metrics, but that actively optimizes the portfolio to improve on them.
The practical application of learning agents requires sample efficient and interpretable algorithms. Learning from behavioral priors is a promising way to bootstrap agents with a better-than-random exploration policy or a safe-guard against the pitfalls of early learning. Existing solutions for imitation learning require a large number of expert demonstrations and rely on hard-to-interpret learning methods like Deep Q-learning. In this work we present a planning-based approach that can use these behavioral priors for effective exploration and learning in a reinforcement learning environment, and we demonstrate that curated exploration policies in the form of behavioral priors can help an agent learn faster.
Animal behavior is driven by multiple brain regions working in parallel with distinct control policies. We present a biologically plausible model of off-policy reinforcement learning in the basal ganglia, which enables learning in such an architecture. The model accounts for action-related modulation of dopamine activity that is not captured by previous models that implement on-policy algorithms. In particular, the model predicts that dopamine activity signals a combination of reward prediction error (as in classic models) and "action surprise," a measure of how unexpected an action is relative to the basal ganglia's current policy. In the presence of the action surprise term, the model implements an approximate form of Q-learning. On benchmark navigation and reaching tasks, we show empirically that this model is capable of learning from data driven completely or in part by other policies (e.g. from other brain regions). By contrast, models without the action surprise term suffer in the presence of additional policies, and are incapable of learning at all from behavior that is completely externally driven. The model provides a computational account for numerous experimental findings about dopamine activity that cannot be explained by classic models of reinforcement learning in the basal ganglia. These include differing levels of action surprise signals in dorsal and ventral striatum, decreasing amounts movement-modulated dopamine activity with practice, and representations of action initiation and kinematics in dopamine activity. It also provides further predictions that can be tested with recordings of striatal dopamine activity.
Natural language is an intuitive way for humans to communicate tasks to a robot. While natural language (NL) is ambiguous, real world tasks and their safety requirements need to be communicated unambiguously. Signal Temporal Logic (STL) is a formal logic that can serve as a versatile, expressive, and unambiguous formal language to describe robotic tasks. On one hand, existing work in using STL for the robotics domain typically requires end-users to express task specifications in STL, a challenge for non-expert users. On the other, translating from NL to STL specifications is currently restricted to specific fragments. In this work, we propose DIALOGUESTL, an interactive approach for learning correct and concise STL formulas from (often) ambiguous NL descriptions. We use a combination of semantic parsing, pre-trained transformer-based language models, and user-in-the-loop clarifications aided by a small number of user demonstrations to predict the best STL formula to encode NL task descriptions. An advantage of mapping NL to STL is that there has been considerable recent work on the use of reinforcement learning (RL) to identify control policies for robots. We show we can use Deep Q-Learning techniques to learn optimal policies from the learned STL specifications. We demonstrate that DIALOGUESTL is efficient, scalable, and robust, and has high accuracy in predicting the correct STL formula with a few number of demonstrations and a few interactions with an oracle user.
We study the continuous-time counterpart of Q-learning for reinforcement learning (RL) under the entropy-regularized, exploratory diffusion process formulation introduced by Wang et al. (2020). As the conventional (big) Q-function collapses in continuous time, we consider its first-order approximation and coin the term ``(little) q-function". This function is related to the instantaneous advantage rate function as well as the Hamiltonian. We develop a ``q-learning" theory around the q-function that is independent of time discretization. Given a stochastic policy, we jointly characterize the associated q-function and value function by martingale conditions of certain stochastic processes, in both on-policy and off-policy settings. We then apply the theory to devise different actor-critic algorithms for solving underlying RL problems, depending on whether or not the density function of the Gibbs measure generated from the q-function can be computed explicitly. One of our algorithms interprets the well-known Q-learning algorithm SARSA, and another recovers a policy gradient (PG) based continuous-time algorithm proposed in Jia and Zhou (2022b). Finally, we conduct simulation experiments to compare the performance of our algorithms with those of PG-based algorithms in Jia and Zhou (2022b) and time-discretized conventional Q-learning algorithms.
Our world is full of asymmetries. Gravity and wind can make reaching a place easier than coming back. Social artifacts such as genealogy charts and citation graphs are inherently directed. In reinforcement learning and control, optimal goal-reaching strategies are rarely reversible (symmetrical). Distance functions supported on these asymmetrical structures are called quasimetrics. Despite their common appearance, little research has been done on the learning of quasimetrics. Our theoretical analysis reveals that a common class of learning algorithms, including unconstrained multilayer perceptrons (MLPs), provably fails to learn a quasimetric consistent with training data. In contrast, our proposed Poisson Quasimetric Embedding (PQE) is the first quasimetric learning formulation that both is learnable with gradient-based optimization and enjoys strong performance guarantees. Experiments on random graphs, social graphs, and offline Q-learning demonstrate its effectiveness over many common baselines.
As critically ill patients frequently develop anemia or coagulopathy, transfusion of blood products is a frequent intervention in the Intensive Care Units (ICU). However, inappropriate transfusion decisions made by physicians are often associated with increased risk of complications and higher hospital costs. In this work, we aim to develop a decision support tool that uses available patient information for transfusion decision-making on three common blood products (red blood cells, platelets, and fresh frozen plasma). To this end, we adopt an off-policy batch reinforcement learning (RL) algorithm, namely, discretized Batch Constrained Q-learning, to determine the best action (transfusion or not) given observed patient trajectories. Simultaneously, we consider different state representation approaches and reward design mechanisms to evaluate their impacts on policy learning. Experiments are conducted on two real-world critical care datasets: the MIMIC-III and the UCSF. Results demonstrate that policy recommendations on transfusion achieved comparable matching against true hospital policies via accuracy and weighted importance sampling evaluations on the MIMIC-III dataset. Furthermore, a combination of transfer learning (TL) and RL on the data-scarce UCSF dataset can provide up to $17.02% improvement in terms of accuracy, and up to 18.94% and 21.63% improvement in jump-start and asymptotic performance in terms of weighted importance sampling averaged over three transfusion tasks. Finally, simulations on transfusion decisions suggest that the transferred RL policy could reduce patients' estimated 28-day mortality rate by 2.74% and decreased acuity rate by 1.18% on the UCSF dataset.
Large language models distill broad knowledge from text corpora. However, they can be inconsistent when it comes to completing user specified tasks. This issue can be addressed by finetuning such models via supervised learning on curated datasets, or via reinforcement learning. In this work, we propose a novel offline RL method, implicit language Q-learning (ILQL), designed for use on language models, that combines both the flexible utility maximization framework of RL algorithms with the ability of supervised learning to leverage previously collected data, as well as its simplicity and stability. Our method employs a combination of value conservatism alongside an implicit dataset support constraint in learning value functions, which are then used to guide language model generations towards maximizing user-specified utility functions. In addition to empirically validating ILQL, we present a detailed empirical analysis of situations where offline RL can be useful in natural language generation settings, demonstrating how it can be a more effective utility optimizer than prior approaches for end-to-end dialogue, and how it can effectively optimize high variance reward functions based on subjective judgement, such as whether to label a comment as toxic or not.
In practical applications, we can rarely assume full observability of a system's environment, despite such knowledge being important for determining a reactive control system's precise interaction with its environment. Therefore, we propose an approach for reinforcement learning (RL) in partially observable environments. While assuming that the environment behaves like a partially observable Markov decision process with known discrete actions, we assume no knowledge about its structure or transition probabilities. Our approach combines Q-learning with IoAlergia, a method for learning Markov decision processes (MDP). By learning MDP models of the environment from episodes of the RL agent, we enable RL in partially observable domains without explicit, additional memory to track previous interactions for dealing with ambiguities stemming from partial observability. We instead provide RL with additional observations in the form of abstract environment states by simulating new experiences on learned environment models to track the explored states. In our evaluation, we report on the validity of our approach and its promising performance in comparison to six state-of-the-art deep RL techniques with recurrent neural networks and fixed memory.
Recursion is the fundamental paradigm to finitely describe potentially infinite objects. As state-of-the-art reinforcement learning (RL) algorithms cannot directly reason about recursion, they must rely on the practitioner's ingenuity in designing a suitable "flat" representation of the environment. The resulting manual feature constructions and approximations are cumbersome and error-prone; their lack of transparency hampers scalability. To overcome these challenges, we develop RL algorithms capable of computing optimal policies in environments described as a collection of Markov decision processes (MDPs) that can recursively invoke one another. Each constituent MDP is characterized by several entry and exit points that correspond to input and output values of these invocations. These recursive MDPs (or RMDPs) are expressively equivalent to probabilistic pushdown systems (with call-stack playing the role of the pushdown stack), and can model probabilistic programs with recursive procedural calls. We introduce Recursive Q-learning -- a model-free RL algorithm for RMDPs -- and prove that it converges for finite, single-exit and deterministic multi-exit RMDPs under mild assumptions.