The tokamak offers a promising path to fusion energy, but plasma disruptions pose a major economic risk, motivating considerable advances in disruption avoidance. This work develops a reinforcement learning approach to this problem by training a policy to safely ramp-down the plasma current while avoiding limits on a number of quantities correlated with disruptions. The policy training environment is a hybrid physics and machine learning model trained on simulations of the SPARC primary reference discharge (PRD) ramp-down, an upcoming burning plasma scenario which we use as a testbed. To address physics uncertainty and model inaccuracies, the simulation environment is massively parallelized on GPU with randomized physics parameters during policy training. The trained policy is then successfully transferred to a higher fidelity simulator where it successfully ramps down the plasma while avoiding user-specified disruptive limits. We also address the crucial issue of safety criticality by demonstrating that a constraint-conditioned policy can be used as a trajectory design assistant to design a library of feed-forward trajectories to handle different physics conditions and user settings. As a library of trajectories is more interpretable and verifiable offline, we argue such an approach is a promising path for leveraging the capabilities of reinforcement learning in the safety-critical context of burning plasma tokamaks. Finally, we demonstrate how the training environment can be a useful platform for other feed-forward optimization approaches by using an evolutionary algorithm to perform optimization of feed-forward trajectories that are robust to physics uncertainty.

Deep reinforcement learning has demonstrated remarkable achievements across diverse domains such as video games, robotic control, autonomous driving, and drug discovery. Common methodologies in partially-observable domains largely lean on end-to-end learning from high-dimensional observations, such as images, without explicitly reasoning about true state. We suggest an alternative direction, introducing the Partially Supervised Reinforcement Learning (PSRL) framework. At the heart of PSRL is the fusion of both supervised and unsupervised learning. The approach leverages a state estimator to distill supervised semantic state information from high-dimensional observations which are often fully observable at training time. This yields more interpretable policies that compose state predictions with control. In parallel, it captures an unsupervised latent representation. These two-the semantic state and the latent state-are then fused and utilized as inputs to a policy network. This juxtaposition offers practitioners a flexible and dynamic spectrum: from emphasizing supervised state information to integrating richer, latent insights. Extensive experimental results indicate that by merging these dual representations, PSRL offers a potent balance, enhancing model interpretability while preserving, and often significantly outperforming, the performance benchmarks set by traditional methods in terms of reward and convergence speed.

We study the problem of reward poisoning attacks against general offline reinforcement learning with deep neural networks for function approximation. We consider a black-box threat model where the attacker is completely oblivious to the learning algorithm and its budget is limited by constraining both the amount of corruption at each data point, and the total perturbation. We propose an attack strategy called `policy contrast attack'. The high-level idea is to make some low-performing policies appear as high-performing while making high-performing policies appear as low-performing. To the best of our knowledge, we propose the first black-box reward poisoning attack in the general offline RL setting. We provide theoretical insights on the attack design and empirically show that our attack is efficient against current state-of-the-art offline RL algorithms in different kinds of learning datasets.

One of the main challenges of decentralized machine learning paradigms such as Federated Learning (FL) is the presence of local non-i.i.d. datasets. Device-to-device transfers (D2D) between distributed devices has been shown to be an effective tool for dealing with this problem and robust to stragglers. In an unsupervised case, however, it is not obvious how data exchanges should take place due to the absence of labels. In this paper, we propose an approach to create an optimal graph for data transfer using Reinforcement Learning. The goal is to form links that will provide the most benefit considering the environment's constraints and improve convergence speed in an unsupervised FL environment. Numerical analysis shows the advantages in terms of convergence speed and straggler resilience of the proposed method to different available FL schemes and benchmark datasets.

Offline policy learning aims to discover decision-making policies from previously-collected datasets without additional online interactions with the environment. As the training dataset is fixed, its quality becomes a crucial determining factor in the performance of the learned policy. This paper studies a dataset characteristic that we refer to as multi-behavior, indicating that the dataset is collected using multiple policies that exhibit distinct behaviors. In contrast, a uni-behavior dataset would be collected solely using one policy. We observed that policies learned from a uni-behavior dataset typically outperform those learned from multi-behavior datasets, despite the uni-behavior dataset having fewer examples and less diversity. Therefore, we propose a behavior-aware deep clustering approach that partitions multi-behavior datasets into several uni-behavior subsets, thereby benefiting downstream policy learning. Our approach is flexible and effective; it can adaptively estimate the number of clusters while demonstrating high clustering accuracy, achieving an average Adjusted Rand Index of 0.987 across various continuous control task datasets. Finally, we present improved policy learning examples using dataset clustering and discuss several potential scenarios where our approach might benefit the offline policy learning community.

Over the recent years, research and development in adaptive bitrate (ABR) algorithms for live video streaming have been successful in improving users' quality of experience (QoE) by reducing latency to near real-time levels while delivering higher bitrate videos with minimal rebuffering time. However, the QoE models used by these ABR algorithms do not take into account that a large portion of live video streaming clients use mobile devices where a higher bitrate does not necessarily translate into higher perceived quality. Ignoring perceived quality results in playing videos at higher bitrates without a significant increase in perceptual video quality and becomes a burden for battery-constrained mobile devices due to higher energy consumption. In this paper, we propose LL-GABR, a deep reinforcement learning approach that models the QoE using perceived video quality instead of bitrate and uses energy consumption along with other metrics like latency, rebuffering events, and smoothness. LL-GABR makes no assumptions about the underlying video, environment, or network settings and can operate flexibly on different video titles, each having a different bitrate encoding ladder without additional re-training, unlike existing learning-based ABRs. Trace-driven experimental results show that LL-GABR outperforms the state-of-the-art approaches by up to 44% in terms of perceptual QoE and a 73% increase in energy efficiency as a result of reducing net energy consumption by 11%.

Demonstrations are commonly used to speed up the learning process of Deep Reinforcement Learning algorithms. To cope with the difficulty of accessing multiple demonstrations, some algorithms have been developed to learn from a single demonstration. In particular, the Divide & Conquer Imitation Learning algorithms leverage a sequential bias to learn a control policy for complex robotic tasks using a single state-based demonstration. The latest version, DCIL-II demonstrates remarkable sample efficiency. This novel method operates within an extended Goal-Conditioned Reinforcement Learning framework, ensuring compatibility between intermediate and subsequent goals extracted from the demonstration. However, a fundamental limitation arises from the assumption that the system can be reset to specific states along the demonstrated trajectory, confining the application to simulated systems. In response, we introduce an extension called Single-Reset DCIL (SR-DCIL), designed to overcome this constraint by relying on a single initial state reset rather than sequential resets. To address this more challenging setting, we integrate two mechanisms inspired by the Learning from Demonstrations literature, including a Demo-Buffer and Value Cloning, to guide the agent toward compatible success states. In addition, we introduce Approximate Goal Switching to facilitate training to reach goals distant from the reset state. Our paper makes several contributions, highlighting the importance of the reset assumption in DCIL-II, presenting the mechanisms of SR-DCIL variants and evaluating their performance in challenging robotic tasks compared to DCIL-II. In summary, this work offers insights into the significance of reset assumptions in the framework of DCIL and proposes SR-DCIL, a first step toward a versatile algorithm capable of learning control policies under a weaker reset assumption.

Command and control (C2) channels are an essential component of many types of cyber attacks, as they enable attackers to remotely control their malware-infected machines and execute harmful actions, such as propagating malicious code across networks, exfiltrating confidential data, or initiating distributed denial of service (DDoS) attacks. Identifying these C2 channels is therefore crucial in helping to mitigate and prevent cyber attacks. However, identifying C2 channels typically involves a manual process, requiring deep knowledge and expertise in cyber operations. In this paper, we propose a reinforcement learning (RL) based approach to automatically emulate C2 attack campaigns using both the normal (public) and the Tor networks. In addition, payload size and network firewalls are configured to simulate real-world attack scenarios. Results on a typical network configuration show that the RL agent can automatically discover resilient C2 attack paths utilizing both Tor-based and conventional communication channels, while also bypassing network firewalls.

Exploration is the key ingredient of reinforcement learning (RL) that determines the speed and success of learning. Here, we quantify and compare the amount of exploration and learning accomplished by a Reinforcement Learning (RL) algorithm. Specifically, we propose a novel measure, named Exploration Index, that quantifies the relative effort of knowledge transfer (transferability) by an RL algorithm in comparison to supervised learning (SL) that transforms the initial data distribution of RL to the corresponding final data distribution. The comparison is established by formulating learning in RL as a sequence of SL tasks, and using optimal transport based metrics to compare the total path traversed by the RL and SL algorithms in the data distribution space. We perform extensive empirical analysis on various environments and with multiple algorithms to demonstrate that the exploration index yields insights about the exploration behaviour of any RL algorithm, and also allows us to compare the exploratory behaviours of different RL algorithms.

This paper introduces innovative methods in Reinforcement Learning (RL), focusing on addressing and exploiting estimation biases in Actor-Critic methods for continuous control tasks, using Deep Double Q-Learning. We propose two novel algorithms: Expectile Delayed Deep Deterministic Policy Gradient (ExpD3) and Bias Exploiting - Twin Delayed Deep Deterministic Policy Gradient (BE-TD3). ExpD3 aims to reduce overestimation bias with a single $Q$ estimate, offering a balance between computational efficiency and performance, while BE-TD3 is designed to dynamically select the most advantageous estimation bias during training. Our extensive experiments across various continuous control tasks demonstrate the effectiveness of our approaches. We show that these algorithms can either match or surpass existing methods like TD3, particularly in environments where estimation biases significantly impact learning. The results underline the importance of bias exploitation in improving policy learning in RL.