Pseudo-rehearsal allows neural networks to learn a sequence of tasks without forgetting how to perform in earlier tasks. Preventing forgetting is achieved by introducing a generative network which can produce data from previously seen tasks so that it can be rehearsed along side learning the new task. This has been found to be effective in both supervised and reinforcement learning. Our current work aims to further prevent forgetting by encouraging the generator to accurately generate features important for task retention. More specifically, the generator is improved by introducing a second discriminator into the Generative Adversarial Network which learns to classify between real and fake items from the intermediate activation patterns that they produce when fed through a continual learning agent. Using Atari 2600 games, we experimentally find that improving the generator can considerably reduce catastrophic forgetting compared to the standard pseudo-rehearsal methods used in deep reinforcement learning. Furthermore, we propose normalising the Q-values taught to the long-term system as we observe this substantially reduces catastrophic forgetting by minimising the interference between tasks' reward functions.

This paper investigates the anti-jamming performance of a cognitive radar under a partially observable Markov decision process (POMDP) model. First, we obtain an explicit expression for uncertainty of jammer dynamics, which paves the way for illuminating the performance metric of probability of being jammed for the radar beyond a conventional signal-to-noise ratio ($\mathsf{SNR}$) based analysis. Considering two frequency hopping strategies developed in the framework of reinforcement learning (RL), this performance metric is analyzed with deep Q-network (DQN) and long short term memory (LSTM) networks under various uncertainty values. Finally, the requirement of the target network in the RL algorithm for both network architectures is replaced with a softmax operator. Simulation results show that this operator improves upon the performance of the traditional target network.

Similarity graphs are an active research direction for the nearest neighbor search (NNS) problem. New algorithms for similarity graph construction are continuously being proposed and analyzed by both theoreticians and practitioners. However, existing construction algorithms are mostly based on heuristics and do not explicitly maximize the target performance measure, i.e., search recall. Therefore, at the moment it is not clear whether the performance of similarity graphs has plateaued or more effective graphs can be constructed with more theoretically grounded methods. In this paper, we introduce a new principled algorithm, based on adjacency matrix optimization, which explicitly maximizes search efficiency. Namely, we propose a probabilistic model of a similarity graph defined in terms of its edge probabilities and show how to learn these probabilities from data as a reinforcement learning task. As confirmed by experiments, the proposed construction method can be used to refine the state-of-the-art similarity graphs, achieving higher recall rates for the same number of distance computations. Furthermore, we analyze the learned graphs and reveal the structural properties that are responsible for more efficient search.

Moving target defense (MTD) is a proactive defense approach that aims to thwart attacks by continuously changing the attack surface of a system (e.g., changing host or network configurations), thereby increasing the adversary's uncertainty and attack cost. To maximize the impact of MTD, a defender must strategically choose when and what changes to make, taking into account both the characteristics of its system as well as the adversary's observed activities. Finding an optimal strategy for MTD presents a significant challenge, especially when facing a resourceful and determined adversary who may respond to the defender's actions. In this paper, we propose finding optimal MTD strategies using deep reinforcement learning. Based on an established model of adaptive MTD, we formulate finding an MTD strategy as finding a policy for a partially-observable Markov decision process. To significantly improve training performance, we introduce compact memory representations. To demonstrate our approach, we provide thorough numerical results, showing significant improvement over existing strategies.

Reinforcement learning has attracted great attention recently, especially policy gradient algorithms, which have been demonstrated on challenging decision making and control tasks. In this paper, we propose an active multi-step TD algorithm with adaptive stepsizes to learn actor and critic. Specifically, our model consists of two components: active stepsize learning and adaptive multi-step TD algorithm. Firstly, we divide the time horizon into chunks and actively select state and action inside each chunk. Then given the selected samples, we propose the adaptive multi-step TD, which generalizes TD($\lambda$), but adaptively switch on/off the backups from future returns of different steps. Particularly, the adaptive multi-step TD introduces a context-aware mechanism, here a binary classifier, which decides whether or not to turn on its future backups based on the context changes. Thus, our model is kind of combination of active learning and multi-step TD algorithm, which has the capacity for learning off-policy without the need of importance sampling. We evaluate our approach on both discrete and continuous space tasks in an off-policy setting respectively, and demonstrate competitive results compared to other reinforcement learning baselines.

In TrojDRL: Trojan Attacks on Deep Reinforcement Learning Agents, a group of Boston University researchers demonstrate an attack on machine learning systems trained with "reinforcement learning" in which ML systems derive solutions to complex problems by iteratively trying multiple solutions. The attack is related to adversarial examples, a class of attacks that involve probing a machine-learning model to find "blind spots" -- very small changes (usually imperceptible to humans) that cause machine learning classifiers' accuracy to shelve off rapidly (for example, a small change to a model of a gun can make an otherwise reliable classifier think it's looking at a helicopter). It's not clear whether it's possible to create a machine learning model that's immune to adversarial examples (the expert I trust most on this told me off the record that they think it's not), but what the researchers behind Trojdrl propose is a method for deliberately introducing adversarial examples by slipping difficult-to-spot changes into training data, which will produce defects in the eventual model that can serve as a "backdoor" that future adversaries can exploit. Training data sets are often ad-hoc in nature; they're so large that it's hard to create version-by-version snapshots, and they're also so prone to mislabeling that researchers are always making changes to them in order to improve their accuracy. All of this suggests that poisoning training data might be easier than it sounds.

One of the most important innovations in the present era for the development of highly-advanced AI systems has been the introduction of Reinforcement Learning (RL). It has the potential to solve complex decision-making problems. It generally follows a "trial and error" method to learn optimal policies of a given problem. It has been used to achieve superhuman performance in competitive strategy games, including Go, Starcraft, Dota, among others. Despite the promise shown by reinforcement algorithms in many decision-making problems, there are few glitches and challenges, which still need to be addressed.

In environments with continuous state and action spaces, state-of-the-art actor-critic reinforcement learning algorithms can solve very complex problems, yet can also fail in environments that seem trivial, but the reason for such failures is still poorly understood. In this paper, we contribute a formal explanation of these failures in the particular case of sparse reward and deterministic environments. First, using a very elementary control problem, we illustrate that the learning process can get stuck into a fixed point corresponding to a poor solution. Then, generalizing from the studied example, we provide a detailed analysis of the underlying mechanisms which results in a new understanding of one of the convergence regimes of these algorithms. The resulting perspective casts a new light on already existing solutions to the issues we have highlighted, and suggests other potential approaches.

This paper studies the problem of spectrum shortage in an unmanned aerial vehicle (UAV) network during critical missions such as wildfire monitoring, search and rescue, and disaster monitoring. Such applications involve a high demand for high-throughput data transmissions such as real-time video-, image-, and voice- streaming where the assigned spectrum to the UAV network may not be adequate to provide the desired Quality of Service (QoS). In these scenarios, the aerial network can borrow an additional spectrum from the available terrestrial networks in the trade of a relaying service for them. We propose a spectrum sharing model in which the UAVs are grouped into two classes of relaying UAVs that service the spectrum owner and the sensing UAVs that perform the disaster relief mission using the obtained spectrum. The operation of the UAV network is managed by a hierarchical mechanism in which a central controller assigns the tasks of the UAVs based on their resources and determine their operation region based on the level of priority of impacted areas and then the UAVs autonomously fine-tune their position using a model-free reinforcement learning algorithm to maximize the individual throughput and prolong their lifetime. We analyze the performance and the convergence for the proposed method analytically and with extensive simulations in different scenarios.