Attacks on vulnerable computer networks and cyber-infrastructure--often called zero-day attacks--can quickly overwhelm traditional defenses, resulting in billions of dollars of damage and requiring weeks of manual patching work to shore up the systems after the intrusion. Now, a Penn State-led team of researchers used a machine learning approach, based on a technique known as reinforcement learning, to create an adaptive cyber defense against these attacks. According to Minghui Zhu, associate professor of electrical engineering and computer science and Institute for Computational and Data Sciences co-hire, the team developed this adaptive machine learning-driven method to address current limitations in a method to detect and respond to cyber-attacks, called moving target defense, or MTD. "These adaptive manual target-defense techniques can dynamically and proactively reconfigure deployed defenses that can increase uncertainty and complexity for attackers during vulnerability windows," said Zhu. "However, existing MTD techniques suffer from two limitations. First, manual selection can be very time consuming. Secondly, manually selected configurations might not be the most cost-effective method to handle this."

A Lawrence Livermore National Laboratory team has developed a new deep reinforcement learning framework for a type of discrete optimization called symbolic regression, showing it could outperform several common methods, including commercial software gold standards, on benchmark problems. The work is being featured at the upcoming International Conference on Learning Representations. From left: LLNL team members Brenden Petersen, Mikel Landajuela, Nathan Mudhenk, Soo Kim, Ruben Glatt and Joanne Kim. Lawrence Livermore National Laboratory (LLNL) computer scientists have developed a new framework and an accompanying visualization tool that leverages deep reinforcement learning for symbolic regression problems, outperforming baseline methods on benchmark problems. The paper was recently accepted as an oral presentation at the International Conference on Learning Representations (ICLR 2021), one of the top machine learning conferences in the world.

This work introduces Bilinear Classes, a new structural framework, which permit generalization in reinforcement learning in a wide variety of settings through the use of function approximation. The framework incorporates nearly all existing models in which a polynomial sample complexity is achievable, and, notably, also includes new models, such as the Linear $Q^*/V^*$ model in which both the optimal $Q$-function and the optimal $V$-function are linear in some known feature space. Our main result provides an RL algorithm which has polynomial sample complexity for Bilinear Classes; notably, this sample complexity is stated in terms of a reduction to the generalization error of an underlying supervised learning sub-problem. These bounds nearly match the best known sample complexity bounds for existing models. Furthermore, this framework also extends to the infinite dimensional (RKHS) setting: for the the Linear $Q^*/V^*$ model, linear MDPs, and linear mixture MDPs, we provide sample complexities that have no explicit dependence on the explicit feature dimension (which could be infinite), but instead depends only on information theoretic quantities.

In real-world tasks, reinforcement learning (RL) agents frequently encounter situations that are not present during training time. To ensure reliable performance, the RL agents need to exhibit robustness against worst-case situations. The robust RL framework addresses this challenge via a worst-case optimization between an agent and an adversary. Previous robust RL algorithms are either sample inefficient, lack robustness guarantees, or do not scale to large problems. We propose the Robust Hallucinated Upper-Confidence RL (RH-UCRL) algorithm to provably solve this problem while attaining near-optimal sample complexity guarantees. RH-UCRL is a model-based reinforcement learning (MBRL) algorithm that effectively distinguishes between epistemic and aleatoric uncertainty and efficiently explores both the agent and adversary decision spaces during policy learning. We scale RH-UCRL to complex tasks via neural networks ensemble models as well as neural network policies. Experimentally, we demonstrate that RH-UCRL outperforms other robust deep RL algorithms in a variety of adversarial environments.

The combination of cloud computing capabilities at the network edge and artificial intelligence promise to turn future mobile networks into service- and radio-aware entities, able to address the requirements of upcoming latency-sensitive applications. In this context, a challenging research goal is to exploit edge intelligence to dynamically and optimally manage the Radio Access Network Slicing (that is a less mature and more complex technology than fifth-generation Network Slicing) and Radio Resource Management, which is a very complex task due to the mostly unpredictably nature of the wireless channel. This paper presents a novel architecture that leverages Deep Reinforcement Learning at the edge of the network in order to address Radio Access Network Slicing and Radio Resource Management optimization supporting latency-sensitive applications. The effectiveness of our proposal against baseline methodologies is investigated through computer simulation, by considering an autonomous-driving use-case.

Mixture models are an expressive hypothesis class that can approximate a rich set of policies. However, using mixture policies in the Maximum Entropy (MaxEnt) framework is not straightforward. The entropy of a mixture model is not equal to the sum of its components, nor does it have a closed-form expression in most cases. Using such policies in MaxEnt algorithms, therefore, requires constructing a tractable approximation of the mixture entropy. In this paper, we derive a simple, low-variance mixture-entropy estimator. We show that it is closely related to the sum of marginal entropies. Equipped with our entropy estimator, we derive an algorithmic variant of Soft Actor-Critic (SAC) to the mixture policy case and evaluate it on a series of continuous control tasks.

Deep reinforcement learning is having a superstar moment. Trouncing physicians at medical diagnoses and crushing humanity's best gamers at Go and Atari. While far from achieving the flexible, quick thinking that comes naturally to humans, this powerful machine learning idea seems unstoppable as a harbinger of better thinking machines. Except there's a massive roadblock: they take forever to run. Because the concept behind these algorithms is based on trial and error, a reinforcement learning AI "agent" only learns after being rewarded for its correct decisions.

Real-world decision-making tasks are generally complex, requiring trade-offs between multiple, often conflicting, objectives. Despite this, the majority of research in reinforcement learning and decision-theoretic planning either assumes only a single objective, or that multiple objectives can be adequately handled via a simple linear combination. Such approaches may oversimplify the underlying problem and hence produce suboptimal results. This paper serves as a guide to the application of multi-objective methods to difficult problems, and is aimed at researchers who are already familiar with single-objective reinforcement learning and planning methods who wish to adopt a multi-objective perspective on their research, as well as practitioners who encounter multi-objective decision problems in practice. It identifies the factors that may influence the nature of the desired solution, and illustrates by example how these influence the design of multi-objective decision-making systems for complex problems.

This paper provides a roadmap that explores the question of how to imbue learning agents with the ability to understand and generate contextually relevant natural language in service of achieving a goal. We hypothesize that two key components in creating such agents are interactivity and environment grounding, shown to be vital parts of language learning in humans, and posit that interactive narratives should be the environments of choice for such training these agents. These games are simulations in which an agent interacts with the world through natural language -- "perceiving", "acting upon", and "talking to" the world using textual descriptions, commands, and dialogue -- and as such exist at the intersection of natural language processing, storytelling, and sequential decision making. We discuss the unique challenges a text games' puzzle-like structure combined with natural language state-and-action spaces provides: knowledge representation, commonsense reasoning, and exploration. Beyond the challenges described so far, progress in the realm of interactive narratives can be applied in adjacent problem domains. These applications provide interesting challenges of their own as well as extensions to those discussed so far. We describe three of them in detail: (1) evaluating AI system's commonsense understanding by automatically creating interactive narratives; (2) adapting abstract text-based policies to include other modalities such as vision; and (3) enabling multi-agent and human-AI collaboration in shared, situated worlds.

In this paper, we investigate the sample complexity of policy evaluation in infinite-horizon offline reinforcement learning (also known as the off-policy evaluation problem) with linear function approximation. We identify a hard regime $d\gamma^{2}>1$, where $d$ is the dimension of the feature vector and $\gamma$ is the discount rate. In this regime, for any $q\in[\gamma^{2},1]$, we can construct a hard instance such that the smallest eigenvalue of its feature covariance matrix is $q/d$ and it requires $\Omega\left(\frac{d}{\gamma^{2}\left(q-\gamma^{2}\right)\varepsilon^{2}}\exp\left(\Theta\left(d\gamma^{2}\right)\right)\right)$ samples to approximate the value function up to an additive error $\varepsilon$. Note that the lower bound of the sample complexity is exponential in $d$. If $q=\gamma^{2}$, even infinite data cannot suffice. Under the low distribution shift assumption, we show that there is an algorithm that needs at most $O\left(\max\left\{ \frac{\left\Vert \theta^{\pi}\right\Vert _{2}^{4}}{\varepsilon^{4}}\log\frac{d}{\delta},\frac{1}{\varepsilon^{2}}\left(d+\log\frac{1}{\delta}\right)\right\} \right)$ samples ($\theta^{\pi}$ is the parameter of the policy in linear function approximation) and guarantees approximation to the value function up to an additive error of $\varepsilon$ with probability at least $1-\delta$.