Two branches of AI - Deep Learning, and Reinforcement Learning are now responsible for many real-world applications. Machine Translation, Speech Recognition, Object Detection, Robot Control, and Drug Discovery - are some of the numerous examples. Both approaches are data hungry - DL requires many examples of each class, and RL needs to play through many episodes to learn a policy. A small child can typically see an image just once, and instantly recognize it in other contexts and environments. We seem to possess an innate model/representation of how the world works, which helps us grasp new concepts and adapt to new situations fast. Humans are excellent one/few shot learners.

In this paper, we derive a new model of synaptic plasticity, b ased on recent algorithms for reinforcement learning (in which an age nt attempts to learn appropriate actions to maximize its long-term averag e reward). We show that these direct reinforcement learning algorithms a lso give locally optimal performance for the problem of reinforcement learn ing with multiple agents, without any explicit communication between a gents. By considering a network of spiking neurons as a collection of agen ts attempting to maximize the long-term average of a reward signal, we deri ve a synaptic update rule that is qualitatively similar to Hebb's post ulate. This rule requires only simple computations, such as addition and lea ky integration, and involves only quantities that are available in the vicin ity of the synapse. Furthermore, it leads to synaptic connection strengths tha t give locally optimal values of the long term average reward. The reinforcem ent learning paradigm is sufficiently broad to encompass many learning pr oblems that are solved by the brain. We illustrate, with simulations, th at the approach is effective for simple pattern classification and motor learn ing tasks. It is widely accepted that the functions performed by neural circuits are modified by adjustments to the strength of the synaptic connectio ns between neurons. 1 In the 1940s, Donald Hebb speculated that such adjustments a re associated with simultaneous (or nearly simultaneous) firing of the presyna ptic and postsynaptic neurons [14]: When an axon of cell A ... persistently takes part in firing [cell B ], some growth process or metabolic change takes place [to incr ease] A's efficacy as one of the cells firing B .

The IKEA Furniture Assembly Environment is one of the first benchmarks for testing and accelerating the automation of complex manipulation tasks. The environment is designed to advance reinforcement learning from simple toy tasks to complex tasks requiring both long-term planning and sophisticated low-level control. Our environment supports over 80 different furniture models, Sawyer and Baxter robot simulation, and domain randomization. The IKEA Furniture Assembly Environment is a testbed for methods aiming to solve complex manipulation tasks. The environment is publicly available at https://clvrai.com/furniture

Off-policy deep reinforcement learning (RL) algorithms are incapable of learning solely from batch offline data without online interactions with the environment, due to the phenomenon known as \textit{extrapolation error}. This is often due to past data available in the replay buffer that may be quite different from the data distribution under the current policy. We argue that most off-policy learning methods fundamentally suffer from a \textit{state distribution shift} due to the mismatch between the state visitation distribution of the data collected by the behavior and target policies. This data distribution shift between current and past samples can significantly impact the performance of most modern off-policy based policy optimization algorithms. In this work, we first do a systematic analysis of state distribution mismatch in off-policy learning, and then develop a novel off-policy policy optimization method to constraint the state distribution shift. To do this, we first estimate the state distribution based on features of the state, using a density estimator and then develop a novel constrained off-policy gradient objective that minimizes the state distribution shift. Our experimental results on continuous control tasks show that minimizing this distribution mismatch can significantly improve performance in most popular practical off-policy policy gradient algorithms.

A BSTRACT Transformers have increasingly outperformed gated RNNs in obtaining new state-of-the-art results on supervised tasks involving text sequences. Inspired by this trend, we study the question of how Transformer-based models can improve the performance of sequential decision-making agents. We present the Working Memory Graph (WMG), an agent that employs multi-head self-attention to reason over a dynamic set of vectors representing observed and recurrent state. We evaluate WMG in two partially observable environments, one that requires complex reasoning over past observations, and another that features factored observations. We find that WMG significantly outperforms gated RNNs on these tasks, supporting the hypothesis that WMG's inductive bias in favor of learning and leveraging factored representations can dramatically boost sample efficiency in environments featuring such structure. In the RNN-based approach of Sutskever et al. (2014), an encoder RNN maps an input sentence to a series of internal hidden state vectors. The encoder's final hidden state is copied into a decoder RNN, which then generates another sequence of hidden states that determine the selection of output tokens in the target language. This model can be trained to translate sentences, but translation quality deteriorates on long sentences where long-term dependencies become critical.

There is an emerging trend in the reinforcement learning for healthcare literature. In order to prepare longitudinal, irregularly sampled, cli nical datasets for reinforcement learning algorithms, many researchers will resa mple the time series data to short, regular intervals and use last-observation- carried-forward (LOCF) imputation to fill in these gaps. Typically, they will not mai ntain any explicit information about which values were imputed. In this work, w e (1) call attention to this practice and discuss its potential implication s; (2) propose an alternative representation of the patient state that addresses som e of these issues; and (3) demonstrate in a novel but representative clinical data set that our alternative representation yields consistently better results for ach ieving optimal control, as measured by off-policy policy evaluation, compared to repr esentations that do not incorporate missingness information.

Imitation learning trains a policy from expert demonstrations. Imitation learning approaches have been designed from various principles, such as behavioral cloning via supervised learning, apprenticeship learning via inverse reinforcement learning, and GAIL via generative adversarial learning. In this paper, we propose a framework to analyze the theoretical property of imitation learning approaches based on discrepancy propagation analysis. Under the infinite-horizon setting, the framework leads to the value discrepancy of behavioral cloning in an order of O((1-\gamma)^{-2}). We also show that the framework leads to the value discrepancy of GAIL in an order of O((1-\gamma)^{-1}). It implies that GAIL has less compounding errors than behavioral cloning, which is also verified empirically in this paper. To the best of our knowledge, we are the first one to analyze GAIL's performance theoretically. The above results indicate that the proposed framework is a general tool to analyze imitation learning approaches. We hope our theoretical results can provide insights for future improvements in imitation learning algorithms.

Thank you to Jeff, Gerald, Milan, Ian, Becky, Jino, Daniel, Narskogr, Jason, and Mariano for being $5 /month Patrons! Follow me on Twitter! http://twitter.com/jordanbharrod

Improved Exploration through Latent Trajectory Optimization in Deep Deterministic Policy Gradient Kevin Sebastian Luck 1, Mel V ecerik 2, Simon Stepputtis 1, Heni Ben Amor 1 and Jonathan Scholz 2 Abstract -- Model-free reinforcement learning algorithms such as Deep Deterministic Policy Gradient (DDPG) often require additional exploration strategies, especially if the actor is of deterministic nature. This work evaluates the use of model-based trajectory optimization methods used for exploration in Deep Deterministic Policy Gradient when trained on a latent image embedding. In addition, an extension of DDPG is derived using a value function as critic, making use of a learned deep dynamics model to compute the policy gradient. This approach leads to a symbiotic relationship between the deep reinforcement learning algorithm and the latent trajectory optimizer . The trajectory optimizer benefits from the critic learned by the RL algorithm and the latter from the enhanced exploration generated by the planner . The developed methods are evaluated on two continuous control tasks, one in simulation and one in the real world. In particular, a Baxter robot is trained to perform an insertion task, while only receiving sparse rewards and images as observations from the environment. I NTRODUCTION Reinforcement learning (RL) methods enabled the development of autonomous systems that can autonomously learn and master a task when provided with an objective function. RL has been successfully applied to a wide range of tasks including flying [24], [17], manipulation [26], [9], [12], [3], [1], locomotion [10], [13], and even autonomous driving [6], [7].

We consider the problem of learning from demonstrated trajectories with inverse reinforcement learning (IRL). Motivated by a limitation of the classical maximum entropy model (Ziebart, Bagnell, and Dey 2010) in capturing the structure of the network of states, we propose an IRL model based on a generalized version of the causal entropy maximization problem, which allows us to generate a class of maximum entropy IRL models. Our generalized model has an advantage of being able to recover, in addition to a reward function, another expert's function that would (partially) capture the impact of the connecting structure of the states on experts' decisions. Empirical evaluation on a real-world dataset and a grid-world dataset shows that our generalized model outperforms the classical ones, in terms of recovering reward functions and demonstrated trajectories.