We designed and developed DOOM (Adversarial-DRL based Opcode level Obfuscator to generate Metamorphic malware), a novel system that uses adversarial deep reinforcement learning to obfuscate malware at the op-code level for the enhancement of IDS. The ultimate goal of DOOM is not to give a potent weapon in the hands of cyber-attackers, but to create defensive-mechanisms against advanced zero-day attacks. Experimental results indicate that the obfuscated malware created by DOOM could effectively mimic multiple-simultaneous zero-day attacks. To the best of our knowledge, DOOM is the first system that could generate obfuscated malware detailed to individual op-code level. DOOM is also the first-ever system to use efficient continuous action control based deep reinforcement learning in the area of malware generation and defense. Experimental results indicate that over 67% of the metamorphic malware generated by DOOM could easily evade detection from even the most potent IDS. This achievement gains significance, as with this, even IDS augment with advanced routing sub-system can be easily evaded by the malware generated by DOOM.

Navigating fluently around pedestrians is a necessary capability for mobile robots deployed in human environments, such as office buildings and homes. While related literature has addressed the co-navigation problem focused on the scalability with the number of pedestrians in open spaces, typical indoor environments present the additional challenge of constrained spaces such as corridors, doorways and crosswalks that limit maneuverability and influence patterns of pedestrian interaction. We present an approach based on reinforcement learning to learn policies capable of dynamic adaptation to the presence of moving pedestrians while navigating between desired locations in constrained environments. The policy network receives guidance from a motion planner that provides waypoints to follow a globally planned trajectory, whereas the reinforcement component handles the local interactions. We explore a compositional principle for multi-layout training and find that policies trained in a small set of geometrically simple layouts successfully generalize to unseen and more complex layouts that exhibit composition of the simple structural elements available during training. Going beyond wall-world like domains, we show transfer of the learned policy to unseen 3D reconstructions of two real environments (market, home). These results support the applicability of the compositional principle to real-world environments and indicate promising usage of agent simulation within reconstructed environments for tasks that involve interaction.

Learning dexterous manipulation in high-dimensional state-action spaces is an important open challenge with exploration presenting a major bottleneck. Although in many cases the learning process could be guided by demonstrations or other suboptimal experts, current RL algorithms for continuous action spaces often fail to effectively utilize combinations of highly off-policy expert data and on-policy exploration data. As a solution, we introduce Relative Entropy Q-Learning (REQ), a simple policy iteration algorithm that combines ideas from successful offline and conventional RL algorithms. It represents the optimal policy via importance sampling from a learned prior and is well-suited to take advantage of mixed data distributions. We demonstrate experimentally that REQ outperforms several strong baselines on robotic manipulation tasks for which suboptimal experts are available. We show how suboptimal experts can be constructed effectively by composing simple waypoint tracking controllers, and we also show how learned primitives can be combined with waypoint controllers to obtain reference behaviors to bootstrap a complex manipulation task on a simulated bimanual robot with human-like hands. Finally, we show that REQ is also effective for general off-policy RL, offline RL, and RL from demonstrations. Videos and further materials are available at sites.google.com/view/rlfse.

Recent advances in multi-agent reinforcement learning (MARL) have achieved superhuman performance in games like Quake 3 and Dota 2. Unfortunately, these techniques require orders-of-magnitude more training rounds than humans and may not generalize to slightly altered environments or new agent configurations (i.e., ad hoc team play). In this work, we propose Collaborative Q-learning (CollaQ) that achieves state-of-the-art performance in the StarCraft multi-agent challenge and supports ad hoc team play. We first formulate multi-agent collaboration as a joint optimization on reward assignment and show that under certain conditions, each agent has a decentralized Q-function that is approximately optimal and can be decomposed into two terms: the self-term that only relies on the agent's own state, and the interactive term that is related to states of nearby agents, often observed by the current agent. The two terms are jointly trained using regular DQN, regulated with a Multi-Agent Reward Attribution (MARA) loss that ensures both terms retain their semantics. CollaQ is evaluated on various StarCraft maps, outperforming existing state-of-the-art techniques (i.e., QMIX, QTRAN, and VDN) by improving the win rate by 40% with the same number of environment steps. In the more challenging ad hoc team play setting (i.e., reweight/add/remove units without retraining or finetuning), CollaQ outperforms previous SoTA by over 30%. In recent years, multi-agent deep reinforcement learning (MARL) has drawn increasing interest from the research community. MARL algorithms have shown superhuman level performance in various games like Dota 2 (Berner et al., 2019), Quake 3 Arena (Jaderberg et al., 2019), and StarCraft (Samvelyan et al., 2019). However, the algorithms (Schulman et al., 2017; Mnih et al., 2013) are far less sample efficient than humans.

We describe an approach to learning optimal control policies for a large, linear particle accelerator using deep reinforcement learning coupled with a high-fidelity physics engine. The framework consists of an AI controller that uses deep neural nets for state and action-space representation and learns optimal policies using reward signals that are provided by the physics simulator. For this work, we only focus on controlling a small section of the entire accelerator. Nevertheless, initial results indicate that we can achieve better-than-human level performance in terms of particle beam current and distribution. The ultimate goal of this line of work is to substantially reduce the tuning time for such facilities by orders of magnitude, and achieve near-autonomous control.

While Deep Reinforcement Learning (DRL) has emerged as a promising approach to many complex tasks, it remains challenging to train a single DRL agent that is capable of undertaking multiple different continuous control tasks. In this paper, we present a Knowledge Transfer based Multi-task Deep Reinforcement Learning framework (KTM-DRL) for continuous control, which enables a single DRL agent to achieve expert-level performance in multiple different tasks by learning from task-specific teachers. In KTM-DRL, the multi-task agent first leverages an offline knowledge transfer algorithm designed particularly for the actor-critic architecture to quickly learn a control policy from the experience of task-specific teachers, and then it employs an online learning algorithm to further improve itself by learning from new online transition samples under the guidance of those teachers. We perform a comprehensive empirical study with two commonly-used benchmarks in the MuJoCo continuous control task suite. The experimental results well justify the effectiveness of KTM-DRL and its knowledge transfer and online learning algorithms, as well as its superiority over the state-of-the-art by a large margin.

This paper studies the safe reinforcement learning (RL) problem without assumptions about prior knowledge of the system dynamics and the constraint function. We employ an uncertainty-aware neural network ensemble model to learn the dynamics, and we infer the unknown constraint function through indicator constraint violation signals. We use model predictive control (MPC) as the basic control framework and propose the robust cross-entropy method (RCE) to optimize the control sequence considering the model uncertainty and constraints. We evaluate our methods in the Safety Gym environment. The results show that our approach achieves better constraint satisfaction than baseline safe RL methods while maintaining good task performance. Additionally, we are able to achieve several orders of magnitude better sample efficiency when compared to constrained model-free RL approaches. The code is available at https://github.com/liuzuxin/safe-mbrl.

Today, human operators primarily perform voltage control of the electric transmission system. As the complexity of the grid increases, so does its operation, suggesting additional automation could be beneficial. A subset of machine learning known as deep reinforcement learning (DRL) has recently shown promise in performing tasks typically performed by humans. This paper applies DRL to the transmission voltage control problem, presents open-source DRL environments for voltage control, proposes a novel modification to the "deep Q network" (DQN) algorithm, and performs experiments at scale with systems up to 500 buses. The promise of applying DRL to voltage control is demonstrated, though more research is needed to enable DRL-based techniques to consistently outperform conventional methods.

We consider reinforcement learning (RL) in episodic Markov decision processes (MDPs) with linear function approximation under drifting environment. Specifically, both the reward and state transition functions can evolve over time, as long as their respective total variations, quantified by suitable metrics, do not exceed certain \textit{variation budgets}. We first develop the $\texttt{LSVI-UCB-Restart}$ algorithm, an optimistic modification of least-squares value iteration combined with periodic restart, and establish its dynamic regret bound when variation budgets are known. We then propose a parameter-free algorithm, $\texttt{Ada-LSVI-UCB-Restart}$, that works without knowing the variation budgets, but with a slightly worse dynamic regret bound. We also derive the first minimax dynamic regret lower bound for nonstationary MDPs to show that our proposed algorithms are near-optimal. As a byproduct, we establish a minimax regret lower bound for linear MDPs, which is unsolved by \cite{jin2020provably}. In addition, we provide numerical experiments to demonstrate the effectiveness of our proposed algorithms. As far as we know, this is the first dynamic regret analysis in nonstationary reinforcement learning with function approximation.

In Cooperative Multi-Agent Reinforcement Learning (MARL) and under the setting of Centralized Training with Decentralized Execution (CTDE), agents observe and interact with their environment locally and independently. With local observation and random sampling, the randomness in rewards and observations leads to randomness in long-term returns. Existing methods such as Value Decomposition Network (VDN) and QMIX estimate the value of long-term returns as a scalar that does not contain the information of randomness. Our proposed model QR-MIX introduces quantile regression, modeling joint state-action values as a distribution, combining QMIX with Implicit Quantile Network (IQN). However, the monotonicity in QMIX limits the expression of joint state-action value distribution and may lead to incorrect estimation results in non-monotonic cases. Therefore, we proposed a flexible loss function to approximate the monotonicity found in QMIX. Our model is not only more tolerant of the randomness of returns, but also more tolerant of the randomness of monotonic constraints. The experimental results demonstrate that QR-MIX outperforms the previous state-of-the-art method QMIX in the StarCraft Multi-Agent Challenge (SMAC) environment.