PyTorchKR Papers

최신 논문

총 1,872편
태그·기간 등 상세 필터

카테고리 탐색

로그인 후 이 태그 구독

arXiv:2109.07137

Optimal Cycling of a Heterogenous Battery Bank via Reinforcement Learning

We consider the problem of optimal charging/discharging of a bank of heterogenous battery units, driven by stochastic electricity generation and demand processes. The batteries in the battery bank may differ with respect to their capacities, ramp constraints, losses, as well as cycling costs. The goal is to minimize the degradation costs associated with battery cycling in the long run; this is posed formally as a Markov decision process. We propose a linear function approximation based Q-learning algorithm for learning the optimal solution, using a specially designed class of kernel functions that approximate the structure of the value functions associated with the MDP. The proposed algorithm is validated via an extensive case study.

arXiv:2109.07054

Convergence of a Human-in-the-Loop Policy-Gradient Algorithm With Eligibility Trace Under Reward, Policy, and Advantage Feedback

Fluid human-agent communication is essential for the future of human-in-the-loop reinforcement learning. An agent must respond appropriately to feedback from its human trainer even before they have significant experience working together. Therefore, it is important that learning agents respond well to various feedback schemes human trainers are likely to provide. This work analyzes the COnvergent Actor-Critic by Humans (COACH) algorithm under three different types of feedback-policy feedback, reward feedback, and advantage feedback. For these three feedback types, we find that COACH can behave sub-optimally. We propose a variant of COACH, episodic COACH (E-COACH), which we prove converges for all three types. We compare our COACH variant with two other reinforcement-learning algorithms: Q-learning and TAMER.

arXiv:2109.06449

Deep hierarchical reinforcement agents for automated penetration testing

Penetration testing the organised attack of a computer system in order to test existing defences has been used extensively to evaluate network security. This is a time consuming process and requires in-depth knowledge for the establishment of a strategy that resembles a real cyber-attack. This paper presents a novel deep reinforcement learning architecture with hierarchically structured agents called HA-DRL, which employs an algebraic action decomposition strategy to address the large discrete action space of an autonomous penetration testing simulator where the number of actions is exponentially increased with the complexity of the designed cybersecurity network. The proposed architecture is shown to find the optimal attacking policy faster and more stably than a conventional deep Q-learning agent which is commonly used as a method to apply artificial intelligence in automatic penetration testing.

arXiv:2109.04504

Bootstrapped Meta-Learning

Meta-learning empowers artificial intelligence to increase its efficiency by learning how to learn. Unlocking this potential involves overcoming a challenging meta-optimisation problem. We propose an algorithm that tackles this problem by letting the meta-learner teach itself. The algorithm first bootstraps a target from the meta-learner, then optimises the meta-learner by minimising the distance to that target under a chosen (pseudo-)metric. Focusing on meta-learning with gradients, we establish conditions that guarantee performance improvements and show that the metric can control meta-optimisation. Meanwhile, the bootstrapping mechanism can extend the effective meta-learning horizon without requiring backpropagation through all updates. We achieve a new state-of-the art for model-free agents on the Atari ALE benchmark and demonstrate that it yields both performance and efficiency gains in multi-task meta-learning. Finally, we explore how bootstrapping opens up new possibilities and find that it can meta-learn efficient exploration in an epsilon-greedy Q-learning agent, without backpropagating through the update rule.

arXiv:2109.04155공식 코드 · Papers with Code

Deep Active Inference for Pixel-Based Discrete Control: Evaluation on the Car Racing Problem

Despite the potential of active inference for visual-based control, learning the model and the preferences (priors) while interacting with the environment is challenging. Here, we study the performance of a deep active inference (dAIF) agent on OpenAI's car racing benchmark, where there is no access to the car's state. The agent learns to encode the world's state from high-dimensional input through unsupervised representation learning. State inference and control are learned end-to-end by optimizing the expected free energy. Results show that our model achieves comparable performance to deep Q-learning. However, vanilla dAIF does not reach state-of-the-art performance compared to other world model approaches. Hence, we discuss the current model implementation's limitations and potential architectures to overcome them.

arXiv:2109.04083

User Tampering in Reinforcement Learning Recommender Systems

In this paper, we introduce new formal methods and provide empirical evidence to highlight a unique safety concern prevalent in reinforcement learning (RL)-based recommendation algorithms -- 'user tampering.' User tampering is a situation where an RL-based recommender system may manipulate a media user's opinions through its suggestions as part of a policy to maximize long-term user engagement. We use formal techniques from causal modeling to critically analyze prevailing solutions proposed in the literature for implementing scalable RL-based recommendation systems, and we observe that these methods do not adequately prevent user tampering. Moreover, we evaluate existing mitigation strategies for reward tampering issues, and show that these methods are insufficient in addressing the distinct phenomenon of user tampering within the context of recommendations. We further reinforce our findings with a simulation study of an RL-based recommendation system focused on the dissemination of political content. Our study shows that a Q-learning algorithm consistently learns to exploit its opportunities to polarize simulated users with its early recommendations in order to have more consistent success with subsequent recommendations that align with this induced polarization. Our findings emphasize the necessity for developing safer RL-based recommendation systems and suggest that achieving such safety would require a fundamental shift in the design away from the approaches we have seen in the recent literature.

arXiv:2109.03445

Convergence of Batch Asynchronous Stochastic Approximation With Applications to Reinforcement Learning

We begin by briefly surveying some results on the convergence of the Stochastic Gradient Descent (SGD) Method, proved in a companion paper by the present authors. These results are based on viewing SGD as a version of Stochastic Approximation (SA). Ever since its introduction in the classic paper of Robbins and Monro in 1951, SA has become a standard tool for finding a solution of an equation of the form $f(\theta) = 0$, when only noisy measurements of $f(\cdot)$ are available. In most situations, \textit{every component} of the putative solution $\theta_t$ is updated at each step $t$. In some applications in Reinforcement Learning (RL), \textit{only one component} of $\theta_t$ is updated at each $t$. This is known as \textbf{asynchronous} SA. In this paper, we study \textbf{Block Asynchronous SA (BASA)}, in which, at each step $t$, \textit{some but not necessarily all} components of $\theta_t$ are updated. The theory presented here embraces both conventional (synchronous) SA as well as asynchronous SA, and all in-between possibilities. We provide sufficient conditions for the convergence of BASA, and also prove bounds on the \textit{rate} of convergence of $\theta_t$ to the solution. For the case of conventional SGD, these results reduce to those proved in our companion paper. Then we apply these results to the problem of finding a fixed point of a map with only noisy measurements. This problem arises frequently in RL. We prove sufficient conditions for convergence as well as estimates for the rate of convergence.

arXiv:2109.02786

Deep SIMBAD: Active Landmark-based Self-localization Using Ranking -based Scene Descriptor

Landmark-based robot self-localization has recently garnered interest as a highly-compressive domain-invariant approach for performing visual place recognition (VPR) across domains (e.g., time of day, weather, and season). However, landmark-based self-localization can be an ill-posed problem for a passive observer (e.g., manual robot control), as many viewpoints may not provide an effective landmark view. In this study, we consider an active self-localization task by an active observer and present a novel reinforcement learning (RL)-based next-best-view (NBV) planner. Our contributions are as follows. (1) SIMBAD-based VPR: We formulate the problem of landmark-based compact scene description as SIMBAD (similarity-based pattern recognition) and further present its deep learning extension. (2) VPR-to-NBV knowledge transfer: We address the challenge of RL under uncertainty (i.e., active self-localization) by transferring the state recognition ability of VPR to the NBV. (3) NNQL-based NBV: We regard the available VPR as the experience database by adapting nearest-neighbor approximation of Q-learning (NNQL). The result shows an extremely compact data structure that compresses both the VPR and NBV into a single incremental inverted index. Experiments using the public NCLT dataset validated the effectiveness of the proposed approach.

arXiv:2109.02194

Learning-Based Strategy Design for Robot-Assisted Reminiscence Therapy Based on a Developed Model for People with Dementia

In this paper, the robot-assisted Reminiscence Therapy (RT) is studied as a psychosocial intervention to persons with dementia (PwDs). We aim at a conversation strategy for the robot by reinforcement learning to stimulate the PwD to talk. Specifically, to characterize the stochastic reactions of a PwD to the robot's actions, a simulation model of a PwD is developed which features the transition probabilities among different PwD states consisting of the response relevance, emotion levels and confusion conditions. A Q-learning (QL) algorithm is then designed to achieve the best conversation strategy for the robot. The objective is to stimulate the PwD to talk as much as possible while keeping the PwD's states as positive as possible. In certain conditions, the achieved strategy gives the PwD choices to continue or change the topic, or stop the conversation, so that the PwD has a sense of control to mitigate the conversation stress. To achieve this, the standard QL algorithm is revised to deliberately integrate the impact of PwD's choices into the Q-value updates. Finally, the simulation results demonstrate the learning convergence and validate the efficacy of the achieved strategy. Tests show that the strategy is capable to duly adjust the difficulty level of prompt according to the PwD's states, take actions (e.g., repeat or explain the prompt, or comfort) to help the PwD out of bad states, and allow the PwD to control the conversation tendency when bad states continue.

arXiv:2109.01417

Event-Based Communication in Distributed Q-Learning

We present an approach to reduce the communication of information needed on a Distributed Q-Learning system inspired by Event Triggered Control (ETC) techniques. We consider a baseline scenario of a distributed Q-learning problem on a Markov Decision Process (MDP). Following an event-based approach, N agents explore the MDP and communicate experiences to a central learner only when necessary, which performs updates of the actor Q functions. We design an Event Based distributed Q learning system (EBd-Q), and derive convergence guarantees with respect to a vanilla Q-learning algorithm. We present experimental results showing that event-based communication results in a substantial reduction of data transmission rates in such distributed systems. Additionally, we discuss what effects (desired and undesired) these event-based approaches have on the learning processes studied, and how they can be applied to more complex multi-agent systems.

arXiv:2109.00525공식 코드 · Papers with Code

Catastrophic Interference in Reinforcement Learning: A Solution Based on Context Division and Knowledge Distillation

The powerful learning ability of deep neural networks enables reinforcement learning agents to learn competent control policies directly from continuous environments. In theory, to achieve stable performance, neural networks assume i.i.d. inputs, which unfortunately does no hold in the general reinforcement learning paradigm where the training data is temporally correlated and non-stationary. This issue may lead to the phenomenon of "catastrophic interference" and the collapse in performance. In this paper, we present IQ, i.e., interference-aware deep Q-learning, to mitigate catastrophic interference in single-task deep reinforcement learning. Specifically, we resort to online clustering to achieve on-the-fly context division, together with a multi-head network and a knowledge distillation regularization term for preserving the policy of learned contexts. Built upon deep Q networks, IQ consistently boosts the stability and performance when compared to existing methods, verified with extensive experiments on classic control and Atari tasks. The code is publicly available at: https://github.com/Sweety-dm/Interference-aware-Deep-Q-learning.

arXiv:2108.13157

DQLEL: Deep Q-Learning for Energy-Optimized LoS/NLoS UWB Node Selection

Recent advancements in Internet of Things (IoTs) have brought about a surge of interest in indoor positioning for the purpose of providing reliable, accurate, and energy-efficient indoor navigation/localization systems. Ultra Wide Band (UWB) technology has been emerged as a potential candidate to satisfy the aforementioned requirements. Although UWB technology can enhance the accuracy of indoor positioning due to the use of a wide-frequency spectrum, there are key challenges ahead for its efficient implementation. On the one hand, achieving high precision in positioning relies on the identification/mitigation Non Line of Sight (NLoS) links, leading to a significant increase in the complexity of the localization framework. On the other hand, UWB beacons have a limited battery life, which is especially problematic in practical circumstances with certain beacons located in strategic positions. To address these challenges, we introduce an efficient node selection framework to enhance the location accuracy without using complex NLoS mitigation methods, while maintaining a balance between the remaining battery life of UWB beacons. Referred to as the Deep Q-Learning Energy-optimized LoS/NLoS (DQLEL) UWB node selection framework, the mobile user is autonomously trained to determine the optimal set of UWB beacons to be localized based on the 2-D Time Difference of Arrival (TDoA) framework. The effectiveness of the proposed DQLEL framework is evaluated in terms of the link condition, the deviation of the remaining battery life of UWB beacons, location error, and cumulative rewards. Based on the simulation results, the proposed DQLEL framework significantly outperformed its counterparts across the aforementioned aspects.

arXiv:2108.12054

Deep Reinforcement Learning for Dynamic Band Switch in Cellular-Connected UAV

The choice of the transmitting frequency to provide cellular-connected Unmanned Aerial Vehicle (UAV) reliable connectivity and mobility support introduce several challenges. Conventional sub-6 GHz networks are optimized for ground Users (UEs). Operating at the millimeter Wave (mmWave) band would provide high-capacity but highly intermittent links. To reach the destination while minimizing a weighted function of traveling time and number of radio failures, we propose in this paper a UAV joint trajectory and band switch approach. By leveraging Double Deep Q-Learning we develop two different approaches to learn a trajectory besides managing the band switch. A first blind approach switches the band along the trajectory anytime the UAV-UE throughput is below a predefined threshold. In addition, we propose a smart approach for simultaneous learning-based path planning of UAV and band switch. The two approaches are compared with an optimal band switch strategy in terms of radio failure and band switches for different thresholds. Results reveal that the smart approach is able in a high threshold regime to reduce the number of radio failures and band switches while reaching the desired destination.

arXiv:2108.05030

DQ-GAT: Towards Safe and Efficient Autonomous Driving with Deep Q-Learning and Graph Attention Networks

Autonomous driving in multi-agent dynamic traffic scenarios is challenging: the behaviors of road users are uncertain and are hard to model explicitly, and the ego-vehicle should apply complicated negotiation skills with them, such as yielding, merging and taking turns, to achieve both safe and efficient driving in various settings. Traditional planning methods are largely rule-based and scale poorly in these complex dynamic scenarios, often leading to reactive or even overly conservative behaviors. Therefore, they require tedious human efforts to maintain workability. Recently, deep learning-based methods have shown promising results with better generalization capability but less hand engineering efforts. However, they are either implemented with supervised imitation learning (IL), which suffers from dataset bias and distribution mismatch issues, or are trained with deep reinforcement learning (DRL) but focus on one specific traffic scenario. In this work, we propose DQ-GAT to achieve scalable and proactive autonomous driving, where graph attention-based networks are used to implicitly model interactions, and deep Q-learning is employed to train the network end-to-end in an unsupervised manner. Extensive experiments in a high-fidelity driving simulator show that our method achieves higher success rates than previous learning-based methods and a traditional rule-based method, and better trades off safety and efficiency in both seen and unseen scenarios. Moreover, qualitative results on a trajectory dataset indicate that our learned policy can be transferred to the real world for practical applications with real-time speeds. Demonstration videos are available at https://caipeide.github.io/dq-gat/.

arXiv:2108.04623

Maximizing Influence with Graph Neural Networks

Finding the seed set that maximizes the influence spread over a network is a well-known NP-hard problem. Though a greedy algorithm can provide near-optimal solutions, the subproblem of influence estimation renders the solutions inefficient. In this work, we propose \textsc{Glie}, a graph neural network that learns how to estimate the influence spread of the independent cascade. \textsc{Glie} relies on a theoretical upper bound that is tightened through supervised training. Experiments indicate that it provides accurate influence estimation for real graphs up to 10 times larger than the train set. Subsequently, we incorporate it into two influence maximization techniques. We first utilize Cost Effective Lazy Forward optimization substituting Monte Carlo simulations with \textsc{Glie}, surpassing the benchmarks albeit with a computational overhead. To improve computational efficiency we develop a provably submodular influence spread based on \textsc{Glie}'s representations, to rank nodes while building the seed set adaptively. The proposed algorithms are inductive, meaning they are trained on graphs with less than 300 nodes and up to 5 seeds, and tested on graphs with millions of nodes and up to 200 seeds. The final method exhibits the most promising combination of time efficiency and influence quality, outperforming several baselines.

arXiv:2108.04115

Modified Double DQN: addressing stability

Inspired by Double Q-learning algorithm, the Double-DQN (DDQN) algorithm was originally proposed in order to address the overestimation issue in the original DQN algorithm. The DDQN has successfully shown both theoretically and empirically the importance of decoupling in terms of action evaluation and selection in computation of target values; although, all the benefits were acquired with only a simple adaption to DQN algorithm, minimal possible change as it was mentioned by the authors. Nevertheless, there seems a roll-back in the proposed algorithm of DDQN since the parameters of policy network are emerged again in the target value function which were initially withdrawn by DQN with the hope of tackling the serious issue of moving targets and the instability caused by it (i.e., by moving targets) in the process of learning. Therefore, in this paper three modifications to the DDQN algorithm are proposed with the hope of maintaining the performance in the terms of both stability and overestimation. These modifications are focused on the logic of decoupling the best action selection and evaluation in the target value function and the logic of tackling the moving targets issue. Each of these modifications have their own pros and cons compared to the others. The mentioned pros and cons mainly refer to the execution time required for the corresponding algorithm and the stability provided by the corresponding algorithm. Also, in terms of overestimation, none of the modifications seem to underperform compared to the original DDQN if not outperform it. With the intention of evaluating the efficacy of the proposed modifications, multiple empirical experiments along with theoretical experiments were conducted. The results obtained are represented and discussed in this article.

arXiv:2108.02827

An Elementary Proof that Q-learning Converges Almost Surely

Watkins' and Dayan's Q-learning is a model-free reinforcement learning algorithm that iteratively refines an estimate for the optimal action-value function of an MDP by stochastically "visiting" many state-ation pairs [Watkins and Dayan, 1992]. Variants of the algorithm lie at the heart of numerous recent state-of-the-art achievements in reinforcement learning, including the superhuman Atari-playing deep Q-network [Mnih et al., 2015]. The goal of this paper is to reproduce a precise and (nearly) self-contained proof that Q-learning converges. Much of the available literature leverages powerful theory to obtain highly generalizable results in this vein. However, this approach requires the reader to be familiar with and make many deep connections to different research areas. A student seeking to deepen their understand of Q-learning risks becoming caught in a vicious cycle of "RL-learning Hell". For this reason, we give a complete proof from start to finish using only one external result from the field of stochastic approximation, despite the fact that this minimal dependence on other results comes at the expense of some "shininess".

arXiv:2108.01832

Offline Decentralized Multi-Agent Reinforcement Learning

In many real-world multi-agent cooperative tasks, due to high cost and risk, agents cannot continuously interact with the environment and collect experiences during learning, but have to learn from offline datasets. However, the transition dynamics in the dataset of each agent can be much different from the ones induced by the learned policies of other agents in execution, creating large errors in value estimates. Consequently, agents learn uncoordinated low-performing policies. In this paper, we propose a framework for offline decentralized multi-agent reinforcement learning, which exploits value deviation and transition normalization to deliberately modify the transition probabilities. Value deviation optimistically increases the transition probabilities of high-value next states, and transition normalization normalizes the transition probabilities of next states. They together enable agents to learn high-performing and coordinated policies. Theoretically, we prove the convergence of Q-learning under the altered non-stationary transition dynamics. Empirically, we show that the framework can be easily built on many existing offline reinforcement learning algorithms and achieve substantial improvement in a variety of multi-agent tasks.

arXiv:2108.01262공식 코드 · Papers with Code

SABER: Data-Driven Motion Planner for Autonomously Navigating Heterogeneous Robots

We present an end-to-end online motion planning framework that uses a data-driven approach to navigate a heterogeneous robot team towards a global goal while avoiding obstacles in uncertain environments. First, we use stochastic model predictive control (SMPC) to calculate control inputs that satisfy robot dynamics, and consider uncertainty during obstacle avoidance with chance constraints. Second, recurrent neural networks are used to provide a quick estimate of future state uncertainty considered in the SMPC finite-time horizon solution, which are trained on uncertainty outputs of various simultaneous localization and mapping algorithms. When two or more robots are in communication range, these uncertainties are then updated using a distributed Kalman filtering approach. Lastly, a Deep Q-learning agent is employed to serve as a high-level path planner, providing the SMPC with target positions that move the robots towards a desired global goal. Our complete methods are demonstrated on a ground and aerial robot simultaneously (code available at: https://github.com/AlexS28/SABER).

arXiv:2107.13268

A Distributed Intelligence Architecture for B5G Network Automation

The management of networks is automated by closed loops. Concurrent closed loops aiming for individual optimization cause conflicts which, left unresolved, leads to significant degradation in performance indicators, resulting in sub-optimal network performance. Centralized optimization avoids conflicts, but impractical in large-scale networks for time-critical applications. Distributed, pervasive intelligence is therefore envisaged in the evolution to B5G networks. In this letter, we propose a Q-Learning-based distributed architecture (QLC), addressing the conflict issue by encouraging cooperation among intelligent agents. We design a realistic B5G network slice auto-scaling model and validate the performance of QLC via simulations, justifying further research in this direction.