AdaPool

Reinforcement learning is a mathematical framework for agents to interact intelligently with their environment. Unlike supervised learning, where a system learns with the help of labeled data, reinforcement learning agents learn how to act by trial and error only receiving a reward signal from their environments. A field where reinforcement learning has been prominently successful is robotics [3]. However, real-world control problems are also particularly challenging because of the noise and high- dimensionality of input data (e.g., visual input). In recent years, in the field of supervised learning, deep neural networks have been successfully used to extract meaning from this kind of data. Building on these advances, deep reinforcement learning was used to solve complex problems like Atari games and Go. Mnih et al. [1] built a system with fixed hyper parameters able to learn to play 49 different Atari games only from raw pixel inputs. However, in order to apply the same methods to real-world control problems, deep reinforcement learning has to be able to deal with continuous action spaces. Discretizing continuous action spaces would scale poorly, since the number of discrete actions grows exponentially with the dimensionality of the action. Furthermore, having a parametrized policy can be advantageous because it can generalize in the action space. Therefore with this thesis we study state-of-the-art deep reinforcement learning algorithm, Deep Deterministic Policy Gradients. We provide a theoretical comparison to other popular methods, an evaluation of its performance, identify its limitations and investigate future directions of research. The remainder of the thesis is organized as follows. We start by introducing the field of interest, machine learning, focusing our attention of deep learning and reinforcement learning. We continue by describing in details the two main algorithms, core of this study, namely Deep Q-Network (DQN) and Deep Deterministic Policy Gradients (DDPG). We then provide implementatory details of DDPG and our test environment, followed by a description of benchmark test cases. Finally, we discuss the results of our evaluation, identifying limitations of the current approach and proposing future avenues of research.

Learning ability is the basic characteristic of human intelligence. The July 1, 2005 issue of Science published a list of 125 important questions in science. Among them, the question 94 “What are the limits of learning by machines?”. The annotation “Computers can already beat the world’s best chess players, and they have a wealth of information on the Web to draw on. But abstract reasoning is still beyond any machine”. In recent artificial intelligence has made great progresses. In 1997, the rise of the man-machine war, IBM Supercomputer Deep Blue defeated the chess master Garry Kasparov. On February 14, 2011, IBM’s Watson supercomputer won a practice round against Jeopardy champions Ken Jennings and Brad Rutter. In March 2016, Google DeepMind’s AlphaGo sealed a 4-1 victory over a South Korean Go grandmaster Lee Se-dol. This paper focuses on the machine learning methods of AlphaGo, including reinforcement learning, deep learning, deep reinforcement learning, analysis of the existing problems and the latest research progress. Deep reinforcement learning is the combination of deep learning and reinforcement learning, which can realize the learning algorithm from the perception to action. Simply said, this is the same as human behavior, input sensing information such as vision, and then, direct output action through the deep neural network. Deep reinforcement learning has the potential to learn a variety of skills for the robot to achieve full autonomy. Even though reinforcement learning is practiced successfully, but feature states need to manually set, for complex scene is a difficult thing, especially easy to cause the dimension disaster, and expression is not good. In 2010, Sascha Lange and Martin Riedmiller proposed deep auto-encoder neural networks in reinforcement learning to extract feature, which is used to control the visual correlation. In 2013, DeepMind proposed deep Q-network (DQN) in NIPS 2013, using convolution neural network to extract features, and then applied in reinforcement learning. They continue to improve and published an improved version of DQN on Nature in 2015, which has aroused widespread concern. In order to break through the limits of learning by machines, cognitive machine learning is proposed, which is the combination of machine learning and brain cognition, so that the machine intelligence is constantly evolving, and gradually reaches the human level of artificial intelligence. A cognitive model entitled Consciousness And Memory (CAM) is proposed by author, which consists of memory, consciousness, high-level cognitive functions, perception and motor. High-level cognitive functions of the brain include learning, language, thinking, decision making, emotion, and so on. Learning is a course to accept the stimulus through the nervous system and obtain new behavior, habits and accumulation experience. According to the current research progress of brain science and cognitive science, cognitive machine learning may be interested in learning emergence, procedural memory knowledge learning, learning evolution and so on. For intelligence, so-called evolution is refers to the learning of learning and the structure also follows the change. It is important to record the learning result by structure changing and improve the learning method.

In this article, we consider a subclass of partially observable Markov decision process (POMDP) problems which we termed confounding POMDPs. In these types of POMDPs, temporal difference (TD)-based reinforcement learning (RL) algorithms struggle, as TD error cannot be easily derived from observations. We solve these types of problems using a new bio-inspired neural architecture that combines a modulated Hebbian network (MOHN) with deep Q-network (DQN), which we call modulated Hebbian plus Q-network architecture (MOHQA). The key idea is to use a Hebbian network with rarely correlated bio-inspired neural traces to bridge temporal delays between actions and rewards when confounding observations and sparse rewards result in inaccurate TD errors. In MOHQA, DQN learns low-level features and control, while the MOHN contributes to high-level decisions by associating rewards with past states and actions. Thus, the proposed architecture combines two modules with significantly different learning algorithms, a Hebbian associative network and a classical DQN pipeline, exploiting the advantages of both. Simulations on a set of POMDPs and on the Malmo environment show that the proposed algorithm improved DQN's results and even outperformed control tests with advantage-actor critic (A2C), quantile regression DQN with long short-term memory (QRDQN + LSTM), Monte Carlo policy gradient (REINFORCE), and aggregated memory for reinforcement learning (AMRL) algorithms on most difficult POMDPs with confounding stimuli and sparse rewards.

To decrease the impact of Partially Observable MDP on deep reinforcement learning performance of intersection signal control, A deep reinforcement learning is proposed in this paper with utilizing the real-time GPS data as well as learning the control of the traffic lights in single intersection. We integrate deep reinforcement learning network (DRQN) with recurrent neural network (RNN) and apply deep network, experience pool and greedy strategy in deep reinforcement learning strategy. It solves the problem of overestimation of target Q value and insufficient long-term experience learning in the standard reinforcement learning of traffic signal control. The comparison of performance was made between the proposed method and standard Deep Q-Network (DQN) on the partial observation of traffic situations. The experimental results show that both DQN and DRQN methods can adjust their traffic signal timing control strategies according to the specific traffic conditions as well as calculating a lower average delay time of vehicle than that of fixed-time control. Besides, the simulation effect of DRQN learning method is better than that of DQN learning method in different probe vehicle proportion environment.

Reinforcement Learning (RL) has become one of the most powerful methods in Artificial Intelligence (AI) which allows a system to learn as it is engaged in an interaction with its environment and to maximize its actions by performing a trial and error process. Although classical RL algorithms can give us a fundamental idea on RL, like the Q-Learning, its usefulness wanes in high-dimensional or continuous-state spaces. To overcome these shortcomings, Deep Reinforcement Learning (DRL) algorithms including Deep Q-Networks (DQN) build on neural networks to learn action-value functions, allowing horizontal scalability and more efficient learning. This paper reports on the experimental comparison of Q-Learning and DQN based on the environment CartPole-v1. Both of the algorithms were trained in the controlled environment and tested under various metrics that include cumulative rewards, success rate, convergence speed, safety (failure rate), and path efficiency. Findings indicate that Q-Learning moves towards moderate stability with slower convergence but DQN takes shorter learning time, is more reliable and performs better. Such results indicate the ground-breaking potential of the DRL in the progress of AI toward transcending classical RL restrictions. Future research topics encompass implementation of more challenging experimental settings, a test of other DRL algorithms (e.g., PPO, SAC), and generalization, scalability and safety in real-world environments.

With the rapid development of AI technology, personalized learning path planning is becoming an important research field. Dual system teaching models have great influence on the learning effect of students. Therefore, how to optimize the learning process of students through the personalized learning path is a great challenge. A Deep Reinforcement Learning (DRL) algorithm for personalized learning path planning is proposed. Based on deep Q-networks and reinforcement learning feedback mechanisms, the algorithm achieves adaptive optimization. The experimental results show that the algorithm based on the DRL is superior to the traditional method in learning performance, learning efficiency and satisfaction. Compared with traditional path planning algorithms based on rules and collaborative filtering, the proposed algorithm improves learning performance by 20%. Moreover, it has obvious advantages in learning motivation and efficiency. This paper provides a new way of thinking to intelligent education system, especially dual-system teaching model, and it is of great practical significance.

The usage of distributed Peer-to-Peer (P2P) networks has been growing steadily for a reasonable period. Various applications rely on the infrastructure of P2P networks, where nodes communicate to accomplish a task without the need for a central authority. One of the significant challenges in P2P networks is the ability of the network nodes to reach a consensus on a shared item; the challenge increases as time passes. Thus, this work proposes a new effective method for tweaking the Deep Reinforcement Learning (DRL) algorithm to train Deep Q Network (DQN) learning agents to reach a consensus among the P2P nodes. We propose various hierarchies of deep agents to address this crucial challenge in P2P networks. DRL is utilized to build and train agents; more precisely, DQN learning agents are constructed and trained. Two scenarios are proposed and evaluated. In the first scenario, one DQN agent is trained to find the consensus between the network nodes. In the second scenario, three hierarchies with different numbers of layers of agents are proposed and evaluated. In both scenarios, the P2P network used is a blockchain network. The best result was obtained using the third hierarchy of the second scenario; the overall accuracy of the model is 87.8%.

In recent years, deep reinforcement learning (DRL) has made impressive achievements in many fields. However, existing DRL algorithms usually require a large amount of exploration to obtain a good action policy. In addition, in many complex situations, the reward function cannot be well designed to meet task requirements. These two problems will make it difficult for DRL to learn a good action policy within a relatively short period. The use of expert data can provide effective guidance and avoid unnecessary exploration. This study proposes a deep imitation reinforcement learning (DIRL) algorithm that uses a certain amount of expert demonstration data to speed up the training of DRL. In the proposed method, the learning agent imitates the expert's action policy by learning from demonstration data. After imitation learning, DRL is used to optimise the action policy in a self‐learning way. By experimental comparison on a video game called the Mario racing game, it is shown that the proposed DIRL algorithm with expert demonstration data can obtain much better performance than previous DRL algorithms without expert guidance.

Reinforcement learning (RL), a promising framework for data-driven decision making in an uncertain environment, has successfully been applied in many real-world operation and control problems. However, the application of RL in a large-scale decentralized multi-agent environment remains a challenging problem due to the partial observability and limited communications between agents. In this paper, we develop a model-based kernel RL approach and a model-free deep RL approach for learning a decentralized, shared policy among homogeneous agents. By leveraging the strengths of both these methods, we further propose a novel deep ensemble multi-agent reinforcement learning (MARL) method that efficiently learns to arbitrate between the decisions of the local kernel-based RL model and the wider-reaching deep RL model. We validate the proposed deep ensemble method on a highly challenging real-world air traffic control problem, where the goal is to provide effective guidance to aircraft to avoid air traffic congestion, conflicting situations, and to improve arrival timeliness, by dynamically recommending adjustments of aircraft speeds in real-time. Extensive empirical results from an open-source air traffic management simulation model, developed by Eurocontrol and built on a real-world data set including thousands of aircrafts, demonstrate that our proposed deep ensemble MARL method significantly outperforms three state-of-the-art benchmark approaches.

Due to limited infrastructure capacity and the increasing number of vehicles, traffic congestion remains a major problem in urban management. The scalability and adaptability issues with traditional traffic signal control (TSC) techniques call for innovative solutions. In order to optimize intelligent transportation systems (ITS), this paper investigates the use of deep reinforcement learning (DRL) models, specifically deep Q-networks (DQN) and their variations. In this study, we employed a Markov Decision Process (MDP) framework to assess how various RL models improve traffic flow efficiency and alleviate congestion. This allows adaptive real-time decision-making to improve traffic flow. We assess how well two RL agents—"Choose Next Phase" and "Next or Not"—manage traffic at a four-way intersection. The simulation results show that RL-based methods perform better than conventional methods, reducing emergency vehicle delays, average waiting time, and queue length. By obtaining higher and more consistent rewards over training iterations, the "Choose Next Phase" agent showed exceptional adaptability. Additionally, under abnormal traffic conditions, we compare RL-based TSC with traditional heuristic-based techniques, emphasizing DRL’s superior handling of complex scenarios. To further increase scalability and decision-making reliability in practical applications, future research is highlighted with an emphasis on improving RL models using multi-agent learning, transformer-based architectures, and decentralized computing.

The flying sidekick traveling salesman problem with stochastic travel time: A reinforcement learning approach

To solve problems involving sequential decision-making, Deep Reinforcement Learning (DRL) combines the advantages of Reinforcement Learning (RL) and Deep Learning (DL).These fusions are skilled at training agents for video games because they use neural networks to approximate nonlinear functions.The Deep Q Network (DQN), a key algorithm in this field, tends to overestimate Q-values despite its effectiveness.The Double Deep Q Network (Double DQN) was presented to address this.The Kera's framework in PyCharm is used in this study to examine the practical application and comparative analysis of DQN and Double DQN in the Flappy Bird game.To speed up its training, improvements were also made to the DQN model.The enhanced DQN performed better than the traditional DQN but less well than the Double DQN, according to the results.This study delves into the practical application and comparative analysis of DQN and Double DQN in the Flappy Bird game, utilizing the Kera's framework in PyCharm.Additionally, enhancements were made to the DQN model to expedite its training.Results affirmed that the enhanced DQN outperformed the conventional DQN but lagged the Double DQN.The training loss trajectory further substantiated Double DQN's superiority in mitigating overestimation issues.

Against the backdrop of the rapid development of big data technology, the phenomenon of price discrimination, colloquially known as "big data-enabled price gouging," has sparked widespread social concern and a crisis of consumer trust. To investigate this issue and seek solutions, this study combines traditional empirical analysis with cutting-edge machine learning simulation. First, through questionnaire surveys and regression analysis, and based on equity theory and price sensitivity theory, we verify the direct negative impact of big data-driven price discrimination on customer loyalty, revealing the mediating role of perceived unfairness and the moderating role of price sensitivity. Second, based on these empirical results, we construct a high-fidelity market simulation environment and employ the Deep Q-Network (DQN) algorithm from Deep Reinforcement Learning (DRL) to train a dynamic pricing agent. The agent's objective is to maximize long-term profit while maintaining customer loyalty. The simulation results show that the trained reinforcement learning strategy significantly outperforms both myopic greedy pricing and fixed-price strategies in terms of total revenue and customer retention. The final policy learned by the agent confirms that to maintain long-term customer relationships, it is necessary to offer discounts to price-sensitive, low-loyalty users while adopting moderate pricing for high-loyalty users, thereby effectively validating our initial theoretical hypotheses. This study not only reveals the internal mechanism through which price discrimination affects customer loyalty but also provides a scientific basis and practical path for enterprises to optimize pricing and achieve sustainable development using advanced algorithms.

Deep reinforcement learning (DRL) algorithms have been widely applied in user cold-start recommender systems because they can gradually capture users' dynamic interest preferences. Deep Q-Networks (DQN) have become the most popular reinforcement learning (RL) method due to their simple update strategy and excellent performance. In many user cold-start scenarios, the action space is gradually reduced to avoid recommending duplicate items to users. However, current DQN-based RL recommender systems output the entire action space fixedly, inevitably leading to discrepancies with the gradually shrinking action space. This paper demonstrates that such discrepancies cause a decrement error in the action space corresponding to the temporal difference (TD) in the original RL, rendering standard DQN reinforcement learning methods inaccurate in Q-value estimation. Moreover, in long-term recommendation scenarios, the differences in the lengths of interactions recommended to different users are significant, making it difficult to ignore such errors, thereby challenging the applicability of these methods in scenarios where the action space gradually reduces. To address this issue, this paper introduces a new algorithm called Q-AD (Q-learning Action Decrease), which is based on DQN and aims to mitigate the reduction error in the action space by buffering the Q-value estimation error at each update. Q-AD augments the standard DQN with an error reduction term for TD updates. Through experiments, it was observed that the Q-AD algorithm significantly reduces value estimation errors and achieves better accuracy and efficiency compared to previous methods across different datasets.

By addressing the drawbacks of static optimization techniques, this research seeks to improve the dynamic energy management of thermoelectric generators (TEGs). Finding the best deep reinforcement learning (DRL) algorithm to maximize energy distribution, prolong battery life, and boost system efficiency in the face of the variable conditions present in waste heat recovery systems is the goal. The TEG system was modeled using the Markov decision process and implemented in a computer-simulated environment. Three advanced DRL algorithms were used: soft actor-critic (SAC), proximal policy optimization (PPO), and deep Q-networks (DQN); which were trained to act as intelligent controllers. The performance of each algorithm was systematically evaluated and compared using key metrics, including average cumulative reward, battery health, system efficiency, and a novel metric termed the energy fulfillment rate, which measures the ability to meet demand while storing surplus energy. The comparative analysis revealed a critical trade-off between maximizing performance and ensuring hardware longevity. The SAC algorithm demonstrated the best overall performance, achieving the highest average reward (− 7.03) and energy fulfillment rate (22.84%). However, the A2C, DDPG, and PPO algorithms all achieved a perfect average battery health of 100.00%, highlighting their superior capability for preserving system longevity, albeit with slightly lower rewards. The DQN algorithm consistently showed the least effective performance across all metrics, particularly in maintaining battery health (60.73%). The SAC algorithm is the most suitable of the methods tested for dynamically managing TEG systems. Its underlying principle of maximization of entropy enables a better exploration of control strategies, leading to a better balance between immediate energy dispatch and long-term storage goals. The findings confirm the significant potential of DRL to create efficient and adaptive controllers for renewable energy applications, although further validation of physical hardware is required to confirm real-world viability.Supplementary InformationThe online version contains supplementary material available at 10.1038/s41598-025-27210-7.