We theoretically and empirically explore the explainability benefits of adversarial learning in logistic regression models on structured datasets. In particular we focus on improved explainability due to significantly higher $\textit{feature-concentration}$ in adversarially-learned models: Compared to natural training, adversarial training tends to more efficiently shrink the weights of non-predictive and weakly-predictive features, while model performance on natural test data only degrades slightly (and even sometimes improves), compared to that of a naturally trained model. We provide theoretical insight into this phenomenon via an analysis of the expectation of the logistic model weight updates by an SGD-based adversarial learning algorithm, where examples are drawn from a random binary data-generation process. We empirically demonstrate the feature-pruning effect on a synthetic dataset, some datasets from the UCI repository, and real-world large-scale advertising response-prediction data-sets from MediaMath. In several of the MediaMath datasets there are 10s of millions of data points, and on the order of 100,000 sparse categorical features, and adversarial learning often results in model-size reduction by a factor of 20 or higher, and yet the model performance on natural test data (measured by AUC) is comparable to (and sometimes even better than) that of the naturally trained model. We also show that traditional $\ell_1$ regularization does not even come close to achieving this level of feature-concentration. We measure "feature concentration" using the Integrated Gradients-based feature-attribution method of Sundararajan et. al (2017), and derive a new closed-form expression for 1-layer networks, which substantially speeds up computation of aggregate feature attributions across a large dataset.

In many reinforcement learning tasks, the goal is to learn a policy to manipulate an agent, whose design is fixed, to maximize some notion of cumulative reward. The design of the agent's physical structure is rarely optimized for the task at hand. In this work, we explore the possibility of learning a version of the agent's design that is better suited for its task, jointly with the policy. We propose a minor alteration to the OpenAI Gym framework, where we parameterize parts of an environment, and allow an agent to jointly learn to modify these environment parameters along with its policy. We demonstrate that an agent can learn a better structure of its body that is not only better suited for the task, but also facilitates policy learning. Joint learning of policy and structure may even uncover design principles that are useful for assisted-design applications. Videos of results at https://designrl.github.io/

In this paper, we present a deep reinforcement learning method for quadcopter bypassing the obstacle on the flying path. In the past study, the algorithm only controls the forward direction about quadcopter. In this letter, we use two functions to control quadcopter. One is quadcopter navigating function. It is based on calculating coordination point and find the straight path to the goal. The other function is collision avoidance function. It is implemented by deep Q-network model. Both two function will output rotating degree, the agent will combine both output and turn direct. Besides, deep Q-network can also make quadcopter fly up and down to bypass the obstacle and arrive at the goal. Our experimental result shows that the collision rate is 14% after 500 flights. Based on this work, we will train more complex sense and transfer model to the real quadcopter.

The ability to learn optimal control policies in systems where action space is defined by sentences in natural language would allow many interesting real-world applications such as automatic optimisation of dialogue systems. Text-based games with multiple endings and rewards are a promising platform for this task, since their feedback allows us to employ reinforcement learning techniques to jointly learn text representations and control policies. We argue that the key property of AI agents, especially in the text-games context, is their ability to generalise to previously unseen games. We present a minimalistic text-game playing agent, testing its generalisation and transfer learning performance and showing its ability to play multiple games at once. We also present pyfiction, an open-source library for universal access to different text games that could, together with our agent that implements its interface, serve as a baseline for future research.

A longstanding goal in character animation is to combine data-driven specification of behavior with a system that can execute a similar behavior in a physical simulation, thus enabling realistic responses to perturbations and environmental variation. We show that well-known reinforcement learning (RL) methods can be adapted to learn robust control policies capable of imitating a broad range of example motion clips, while also learning complex recoveries, adapting to changes in morphology, and accomplishing user-specified goals. Our method handles keyframed motions, highly-dynamic actions such as motion-captured flips and spins, and retargeted motions. By combining a motion-imitation objective with a task objective, we can train characters that react intelligently in interactive settings, e.g., by walking in a desired direction or throwing a ball at a user-specified target. This approach thus combines the convenience and motion quality of using motion clips to define the desired style and appearance, with the flexibility and generality afforded by RL methods and physics-based animation. We further explore a number of methods for integrating multiple clips into the learning process to develop multi-skilled agents capable of performing a rich repertoire of diverse skills. We demonstrate results using multiple characters (human, Atlas robot, bipedal dinosaur, dragon) and a large variety of skills, including locomotion, acrobatics, and martial arts.

Real time bidding (RTB) enables demand side platforms (bidders) to scale ad campaigns across multiple publishers affiliated to an RTB ad exchange. While driving multiple campaigns for mobile app install ads via RTB, the bidder typically has to: (i) maintain each campaign's efficiency (i.e., meet advertiser's target cost-per-install), (ii) be sensitive to advertiser's budget, and (iii) make profit after payouts to the ad exchange. In this process, there is a sense of delayed rewards for the bidder's actions; the exchange charges the bidder right after the ad is shown, but the bidder gets to know about resultant installs after considerable delay. This makes it challenging for the bidder to decide beforehand the bid (and corresponding cost charged to advertiser) for each ad display opportunity. To jointly handle the objectives mentioned above, we propose a state space based policy which decides the exchange bid and advertiser cost for each opportunity. The state space captures the current efficiency, budget utilization and profit. The policy based on this state space is trained on past decisions and outcomes via a novel Q-learning algorithm which accounts for the delay in install notifications. In our experiments based on data from app install campaigns managed by Yahoo's Gemini advertising platform, the Q-learning based policy led to a significant increase in the profit and number of efficient campaigns.

Pursuit-evasion is a multi-agent sequential decision problem wherein a group of agents known as pursuers coordinate their traversal of a spatial domain to locate an agent trying to evade them. Pursuit evasion problems arise in a number of import application domains including defense and route planning. Learning to optimally coordinate pursuer behaviors so as to minimize time to capture of the evader is challenging because of a large action space and sparse noisy state information; consequently, previous approaches have relied primarily on heuristics. We propose a variant of Thompson Sampling for pursuit-evasion that allows for the application of existing model-based planning algorithms. This approach is general in that it allows for an arbitrary number of pursuers, a general spatial domain, and the integration of auxiliary information provided by informants. In a suite of simulation experiments, Thompson Sampling for pursuit evasion significantly reduces time-to-capture relative to competing algorithms.

I describe an optimal control view of adversarial machine learning, where the dynamical system is the machine learner, the input are adversarial actions, and the control costs are defined by the adversary's goals to do harm and be hard to detect. This view encompasses many types of adversarial machine learning, including test-item attacks, training-data poisoning, and adversarial reward shaping. The view encourages adversarial machine learning researcher to utilize advances in control theory and reinforcement learning.

Visual attention serves as a means of feature selection mechanism in the perceptual system. Motivated by Broadbent's leaky filter model of selective attention, we evaluate how such mechanism could be implemented and affect the learning process of deep reinforcement learning. We visualize and analyze the feature maps of DQN on a toy problem Catch, and propose an approach to combine visual selective attention with deep reinforcement learning. We experiment with optical flow-based attention and A2C on Atari games. Experiment results show that visual selective attention could lead to improvements in terms of sample efficiency on tested games. An intriguing relation between attention and batch normalization is also discovered.

OpenAI today announced the launch of Spinning Up, a program designed to teach anyone deep reinforcement learning. OpenAI is well known for making funky-looking agents in virtual environments that learn how to walk on their own such as Humanoid v2 or POLO, a collaboration with University of Washington. Reinforcement learning involves providing reward signals to an agent in an environment incentivized to maximize its reward to meet a goal. RL has played a role in major AI breakthroughs such as Google DeepMind's AlphaGo and agents trained in environments like Dota 2. Spinning Up includes a collection of important reinforcement learning research papers, a glossary of terminology necessary to understand RL, and a collection of algorithms for running exercises. The program is being launched not just to help people learn how reinforcement learning works, but to make progress towards OpenAI's general goal of safely creating artificial general intelligence (AGI) by involving more people from fields beyond computer science. "Solving AI safety will require people with a wide range of expertise and perspectives, and many relevant professions have no connection to engineering or computer science at all.