Imitation learning has traditionally been applied to learn a single task from demonstrations thereof. The requirement of structured and isolated demonstrations limits the scalability of imitation learning approaches as they are difficult to apply to real-world scenarios, where robots have to be able to execute a multitude of tasks. In this paper, we propose a multi-modal imitation learning framework that is able to segment and imitate skills from unlabelled and unstructured demonstrations by learning skill segmentation and imitation learning jointly. The extensive simulation results indicate that our method can efficiently separate the demonstrations into individual skills and learn to imitate them using a single multi-modal policy.

We present an algorithm based on posterior sampling (aka Thompson sampling) that achieves near-optimal worst-case regret bounds when the underlying Markov Decision Process (MDP) is communicating with a finite, though unknown, diameter. Our main result is a high probability regret upper bound of $\tilde{O}(D\sqrt{SAT})$ for any communicating MDP with $S$ states, $A$ actions and diameter $D$, when $T\ge S^5A$. Here, regret compares the total reward achieved by the algorithm to the total expected reward of an optimal infinite-horizon undiscounted average reward policy, in time horizon $T$. This result improves over the best previously known upper bound of $\tilde{O}(DS\sqrt{AT})$ achieved by any algorithm in this setting, and matches the dependence on $S$ in the established lower bound of $\Omega(\sqrt{DSAT})$ for this problem. Our techniques involve proving some novel results about the anti-concentration of Dirichlet distribution, which may be of independent interest.

Imitation learning has been commonly applied to solve different tasks in isolation. This usually requires either careful feature engineering, or a significant number of samples. This is far from what we desire: ideally, robots should be able to learn from very few demonstrations of any given task, and instantly generalize to new situations of the same task, without requiring task-specific engineering. In this paper, we propose a meta-learning framework for achieving such capability, which we call one-shot imitation learning. Specifically, we consider the setting where there is a very large (maybe infinite) set of tasks, and each task has many instantiations. For example, a task could be to stack all blocks on a table into a single tower, another task could be to place all blocks on a table into two-block towers, etc. In each case, different instances of the task would consist of different sets of blocks with different initial states. At training time, our algorithm is presented with pairs of demonstrations for a subset of all tasks. A neural net is trained that takes as input one demonstration and the current state (which initially is the initial state of the other demonstration of the pair), and outputs an action with the goal that the resulting sequence of states and actions matches as closely as possible with the second demonstration. At test time, a demonstration of a single instance of a new task is presented, and the neural net is expected to perform well on new instances of this new task. Our experiments show that the use of soft attention allows the model to generalize to conditions and tasks unseen in the training data. We anticipate that by training this model on a much greater variety of tasks and settings, we will obtain a general system that can turn any demonstrations into robust policies that can accomplish an overwhelming variety of tasks.

Reinforcement learning is a powerful paradigm for learning optimal policies from experimental data. However, to find optimal policies, most reinforcement learning algorithms explore all possible actions, which may be harmful for real-world systems. As a consequence, learning algorithms are rarely applied on safety-critical systems in the real world. In this paper, we present a learning algorithm that explicitly considers safety, defined in terms of stability guarantees. Specifically, we extend control-theoretic results on Lyapunov stability verification and show how to use statistical models of the dynamics to obtain high-performance control policies with provable stability certificates. Moreover, under additional regularity assumptions in terms of a Gaussian process prior, we prove that one can effectively and safely collect data in order to learn about the dynamics and thus both improve control performance and expand the safe region of the state space. In our experiments, we show how the resulting algorithm can safely optimize a neural network policy on a simulated inverted pendulum, without the pendulum ever falling down.

In reinforcement learning, agents learn by performing actions and observing their outcomes. Sometimes, it is desirable for a human operator to interrupt an agent in order to prevent dangerous situations from happening. Yet, as part of their learning process, agents may link these interruptions, that impact their reward, to specific states and deliberately avoid them. The situation is particularly challenging in a multi-agent context because agents might not only learn from their own past interruptions, but also from those of other agents. Orseau and Armstrong defined safe interruptibility for one learner, but their work does not naturally extend to multi-agent systems. This paper introduces dynamic safe interruptibility, an alternative definition more suited to decentralized learning problems, and studies this notion in two learning frameworks: joint action learners and independent learners. We give realistic sufficient conditions on the learning algorithm to enable dynamic safe interruptibility in the case of joint action learners, yet show that these conditions are not sufficient for independent learners. We show however that if agents can detect interruptions, it is possible to prune the observations to ensure dynamic safe interruptibility even for independent learners.

Reinforcement Learning is a type of machine learning that allows machines and software agents to act smart and automatically detect the ideal behavior within a specific environment, in order to maximize its performance and productivity. Reinforcement Learning is becoming popular because it not only serves as an way to study how machine and software agents learn to act, it is also been used as a tool for constructing autonomous systems that improve themselves with experience. This video will give you a brief introduction to Reinforcement Learning; it will help you navigate the "Grid world" to calculate likely successful outcomes using the popular MDPToolbox package. This video will show you how the Stimulus - Action - Reward algorithm works in Reinforcement Learning. By the end of this video you will have a basic understanding of the concept of reinforcement learning, you will have compiled your first Reinforcement Learning program, and will have mastered programming the environment for Reinforcement Learning.

About this course: The goal of «Intro to Reinforcement learning» is in its name: introduce students to reinforcement learning – the prominent area of modern research in artificial intelligence. The reinforcement learning differs much from both supervised and unsupervised learning and is more about how humans learn in reality. Students will learn from this course both theoretical core and recent practical RL methods. Most importantly, they will learn how to apply such methods to practical problems. In six weeks students will be guided through the basics of Reinforcement Learning (RL): we will talk about essential theory of RL, value-based methods (such as SARSA and Q-learning), policy based algorithms and methods, designed to solve the optimal exploration problem.

R is a high-level statistical language and is widely used among statisticians and data miners to develop statistical applications. If you want to learn reward-based learning with R, then you should surely go for this Learning Path. Packt's Video Learning Paths are a series of individual video products put together in a logical and stepwise manner such that each video builds on the skills learned in the video before it. Beginning with the basics of R programming, this Learning Path provides step-by-step resources and time-saving methods to help you solve programming problems efficiently. You will be able to boost your productivity with the most popular R packages and data structures such as matrices, lists, and factors.

In this last module, we explore supervised learning and reinforcement learning. The first lecture introduces you to supervised learning with the help of famous faces from politics and Bollywood, casts neurons as classifiers, and gives you a taste of that bedrock of supervised learning, backpropagation, with whose help you will learn to back a truck into a loading dock.The second and third lectures focus on reinforcement learning. The second lecture will teach you how to predict rewards à la Pavlov's dog and will explore the connection to that important reward-related chemical in our brains: dopamine. In the third lecture, we will learn how to select the best actions for maximizing rewards, and examine a possible neural implementation of our computational model in the brain region known as the basal ganglia.

As a request from my friend Richaldo, in this post I'm going to explain the types of machine learning algorithms and when you should use each of them. I particularly think that getting to know the types of Machine learning algorithms is like getting to see the Big Picture of AI and what is the goal of all the things that are being done in the field and put you in a better position to break down a real problem and design a machine learning system. In the previous two types, either there are no labels for all the observation in the dataset or labels are present for all the observations. Semi-supervised learning falls in between these two. In many practical situations, the cost to label is quite high, since it requires skilled human experts to do that.