Anomaly Detection in Textual Speech Using Deep Learning – We propose two deep-learning techniques that improve the state-of-the-art accuracy when supervised text processing tasks are performed on a robot. Our approach combines feature extraction and supervised learning in order to automatically extract relevant features from the input text. Furthermore, a deep convolutional network trained on text is then used to infer features from the input text. Experiments on various datasets show that the proposed models can achieve state-of-the-art classification accuracies when trained on a variety of text corpora.

We propose a novel multi-task reinforcement learning algorithm for reinforcement learning, which involves the learner solving a task and an agent performing a novel task by learning a novel representation of the problem with a low reinforcement cost. The algorithm is a reinforcement learning algorithm which, given a reward or a target environment, learns a distribution over the rewards that is similar to the distribution of the target environment. The objective of the algorithm is to maximize the rewards of each task and environment, while the task and environment are non-differentiable. In this paper, we formalize this objective in terms of the distribution objective, a generalization of the distribution objective which we apply to the reinforcement learning problem that the agent plays with. Given a reinforcement learning algorithm that is a reinforcement learning algorithm in this formal sense, we propose to optimize reinforcement learning with a distribution objective. Extensive experiments on real-world data show that our algorithm achieves state of the art reward performances on various tasks, on four popular reinforcement learning tasks. We also show that our algorithm can also be easily adapted to a variety of real-world reinforcement learning tasks.

Stochastic Lifted Bayesian Networks

SNearest Neighbor Adversarial Search with Binary Codes

# Anomaly Detection in Textual Speech Using Deep Learning

Bayesian nonparametric regression with conditional probability priors

Identical Mixtures of Random Projections (ReLU) for Multiple TargetsWe propose a novel multi-task reinforcement learning algorithm for reinforcement learning, which involves the learner solving a task and an agent performing a novel task by learning a novel representation of the problem with a low reinforcement cost. The algorithm is a reinforcement learning algorithm which, given a reward or a target environment, learns a distribution over the rewards that is similar to the distribution of the target environment. The objective of the algorithm is to maximize the rewards of each task and environment, while the task and environment are non-differentiable. In this paper, we formalize this objective in terms of the distribution objective, a generalization of the distribution objective which we apply to the reinforcement learning problem that the agent plays with. Given a reinforcement learning algorithm that is a reinforcement learning algorithm in this formal sense, we propose to optimize reinforcement learning with a distribution objective. Extensive experiments on real-world data show that our algorithm achieves state of the art reward performances on various tasks, on four popular reinforcement learning tasks. We also show that our algorithm can also be easily adapted to a variety of real-world reinforcement learning tasks.