Composition of elementary skills into complex behaviors to solve challenging problems is one of the key elements toward building intelligent machines. To date, there has been plenty of work on learning new policies or skills but almost no focus on composing them to perform complex decision-making. In this paper, we propose a policy ensemble composition framework that takes the robot's primitive policies and learns to compose them concurrently or sequentially through reinforcement learning. We evaluate our method in problems where traditional approaches either fail or exhibit high sample complexity to find a solution. We show that our method not only solves the problems that require both task and motion planning but also exhibits high data efficiency, which is currently one of the main limitations of reinforcement learning.

Deep reinforcement learning (RL) policies are known to be vulnerable to adversarial perturbations to their observations, similar to adversarial examples for classifiers. However, an attacker is not usually able to directly modify another agent's observations. This might lead one to wonder: is it possible to attack an RL agent simply by choosing an adversarial policy acting in a multi-agent environment so as to create natural observations that are adversarial? We demonstrate the existence of adversarial policies in zero-sum games between simulated humanoid robots with proprioceptive observations, against state-of-the-art victims trained via self-play to be robust to opponents. The adversarial policies reliably win against the victims but generate seemingly random and uncoordinated behavior. We find that these policies are more successful in high-dimensional environments, and induce substantially different activations in the victim policy network than when the victim plays against a normal opponent.

Malware detection is an ever-present challenge for all organizational gatekeepers. Organizations often deploy numerous different malware detection tools, and then combine their output to produce a final classification for an inspected file. This approach has two significant drawbacks. First, it requires large amounts of computing resources and time since every incoming file needs to be analyzed by all detectors. Secondly, it is difficult to accurately and dynamically enforce a predefined security policy that comports with the needs of each organization (e.g., how tolerant is the organization to false negatives and false positives). In this study we propose ASPIRE, a reinforcement learning (RL)-based method for malware detection. Our approach receives the organizational policy -- defined solely by the perceived costs of correct/incorrect classifications and of computing resources -- and then dynamically assigns detection tools and sets the detection threshold for each inspected file. We demonstrate the effectiveness and robustness of our approach by conducting an extensive evaluation on multiple organizational policies. ASPIRE performed well in all scenarios, even achieving near-optimal accuracy of 96.21% (compared to an optimum of 96.86%) at approximately 20% of the running time of this baseline.

Recent advances in reinforcement learning have proved that given an environment we can learn to perform a task in that environment if we have access to some form of a reward function (dense, sparse or derived from IRL). But most of the algorithms focus on learning a single best policy to perform a given set of tasks. In this paper, we focus on an algorithm that learns to not just perform a task but different ways to perform the same task. As we know when the environment is complex enough there always exists multiple ways to perform a task. We show that using the concept of information maximization it is possible to learn latent codes for discovering multiple ways to perform any given task in an environment.

Recent reinforcement learning algorithms, though achieving impressive results in various fields, suffer from brittle training effects such as regression in results and high sensitivity to initialization and parameters. We claim that some of the brittleness stems from variance differences, i.e. when different environment areas - states and/or actions - have different rewards variance. This causes two problems: First, the "Boring Areas Trap" in algorithms such as Q-learning, where moving between areas depends on the current area variance, and getting out of a boring area is hard due to its low variance. Second, the "Manipulative Consultant" problem, when value-estimation functions used in DQN and Actor-Critic algorithms influence the agent to prefer boring areas, regardless of the mean rewards return, as they maximize estimation precision rather than rewards. This sheds a new light on how exploration contribute to training, as it helps with both challenges. Cognitive experiments in humans showed that noised reward signals may paradoxically improve performance. We explain this using the two mentioned problems, claiming that both humans and algorithms may share similar challenges. Inspired by this result, we propose the Adaptive Symmetric Reward Noising (ASRN), by which we mean adding Gaussian noise to rewards according to their states' estimated variance, thus avoiding the two problems while not affecting the environment's mean rewards behavior. We conduct our experiments in a Multi Armed Bandit problem with variance differences. We demonstrate that a Q-learning algorithm shows the brittleness effect in this problem, and that the ASRN scheme can dramatically improve the results. We show that ASRN helps a DQN algorithm training process reach better results in an end to end autonomous driving task using the AirSim driving simulator.

Automatic machine learning is an important problem in the forefront of machine learning. The strongest AutoML systems are based on neural networks, evolutionary algorithms, and Bayesian optimization. Recently AlphaD3M reached state-of-the-art results with an order of magnitude speedup using reinforcement learning with self-play. In this work we extend AlphaD3M by using a pipeline grammar and a pre-trained model which generalizes from many different datasets and similar tasks. Our results demonstrate improved performance compared with our earlier work and existing methods on AutoML benchmark datasets for classification and regression tasks. In the spirit of reproducible research we make our data, models, and code publicly available.

High-dimensional always-changing environments constitute a hard challenge for current reinforcement learning techniques. Artificial agents, nowadays, are often trained off-line in very static and controlled conditions in simulation such that training observations can be thought as sampled i.i.d. from the entire observations space. However, in real world settings, the environment is often non-stationary and subject to unpredictable, frequent changes. In this paper we propose and openly release CRLMaze, a new benchmark for learning continually through reinforcement in a complex 3D non-stationary task based on ViZDoom and subject to several environmental changes. Then, we introduce an end-to-end model-free continual reinforcement learning strategy showing competitive results with respect to four different baselines and not requiring any access to additional supervised signals, previously encountered environmental conditions or observations.

Joint replacement is the most common inpatient surgical treatment in the US. We investigate the clinical pathway optimization for knee replacement, which is a sequential decision process from onset to recovery. Based on episodic claims from previous cases, we view the pathway optimization as an intelligence crowdsourcing problem and learn the optimal decision policy from data by imitating the best expert at every intermediate state. We develop a reinforcement learning-based pipeline that uses value iteration, state compression and aggregation learning, kernel representation and cross validation to predict the best treatment policy. It also provides forecast of the clinical pathway under the optimized policy. Empirical validation shows that the optimized policy reduces the overall cost by 7 percent and reduces the excessive cost premium by 33 percent.

In a reinforcement learning approach, an optimal value function is learned across a set of actions, or decisions, that leads to a set of states giving different rewards, with the objective to maximize the overall reward. A policy assigns to each state-action pairs an expected return. We call an optimal policy a policy for which the value function is optimal. QLBS, Q-Learner in the Black-Scholes(-Merton) Worlds, applies the reinforcement learning concepts, and noticeably, the popular Q-learning algorithm, to the financial stochastic model described by Black, Scholes and Merton. However, QLBS is specifically optimized for the geometric Brownian motion and the pricing of vanilla options. Consequently, it suffers from the traditional over-estimation of the Q-values reflected by an over-estimation of the vanilla option prices. Furthermore, its range of application is limited to vanilla option pricing within the financial markets. We propose MQLV, Modified Q-Learner for the Vasicek model, a new reinforcement learning approach that limits the Q-values over-estimation observed in QLBS and extends the simulation to mean reverting stochastic diffusion processes. Additionally, MQLV uses a digital function to estimate the future probability of an event, thus widening the scope of the financial application to any other domain involving time series. Our experiments underline the potential of MQLV on generated Monte Carlo simulations, particularly representative of the retail banking time series. In particular, MQLV is able to determine the optimal policy of money management based on the aggregated financial transactions of the clients, unlocking new frontiers to establish personalized credit card limits or loans. Finally, MQLV is the first methodology compatible with the Vasicek model capable of an event probability estimation targeting simulation of event probabilities in retail banking.

Value function learning plays a central role in many state-of-the-art reinforcement-learning algorithms. Many popular algorithms like Q-learning do not optimize any objective function, but are fixed-point iterations of some variant of Bellman operator that is not necessarily a contraction. As a result, they may easily lose convergence guarantees, as can be observed in practice. In this paper, we propose a novel loss function, which can be optimized using standard gradient-based methods without risking divergence. The key advantage is that its gradient can be easily approximated using sampled transitions, avoiding the need for double samples required by prior algorithms like residual gradient. Our approach may be combined with general function classes such as neural networks, on either on- or off-policy data, and is shown to work reliably and effectively in several benchmarks.