Each algorithm optimises its parameters with respect to an objective, such as Q-learning or policy gradient, that defines its semantics.

Neural Information Processing Systems http://nips.cc/

Each algorithm optimises its parameters with respect to an objective, such as Q-learning or policy gradient, that defines its semantics.

Deep reinforcement learning includes a broad family of algorithms that parameterise an internal representation, such as a value function or policy, by a deep neural network. Each algorithm optimises its parameters with respect to an objective, such as Q-learning or policy gradient, that defines its semantics. In this work, we propose an algorithm based on meta-gradient descent that discovers its own objective, flexibly parameterised by a deep neural network, solely from interactive experience with its environment. Over time, this allows the agent to learn how to learn increasingly effectively. Furthermore, because the objective is discovered online, it can adapt to changes over time. We demonstrate that the algorithm discovers how to address several important issues in RL, such as bootstrapping, non-stationarity, and off-policy learning. On the Atari Learning Environment, the meta-gradient algorithm adapts over time to learn with greater efficiency, eventually outperforming the median score of a strong actor-critic baseline.

Temporal-Difference (TD) learning is a standard and very successful reinforcement learning approach, at the core of both algorithms that learn the value of a given policy, as well as algorithms which learn how to improve policies. TD-learning with eligibility traces provides a way to do temporal credit assignment, i.e. decide which portion of a reward should be assigned to predecessor states that occurred at different previous times, controlled by a parameter $\lambda$. However, tuning this parameter can be time-consuming, and not tuning it can lead to inefficient learning. To improve the sample efficiency of TD-learning, we propose a meta-learning method for adjusting the eligibility trace parameter, in a state-dependent manner. The adaptation is achieved with the help of auxiliary learners that learn distributional information about the update targets online, incurring roughly the same computational complexity per step as the usual value learner. Our approach can be used both in on-policy and off-policy learning. We prove that, under some assumptions, the proposed method improves the overall quality of the update targets, by minimizing the overall target error. This method can be viewed as a plugin which can also be used to assist prediction with function approximation by meta-learning feature (observation)-based $\lambda$ online, or even in the control case to assist policy improvement. Our empirical evaluation demonstrates significant performance improvements, as well as improved robustness of the proposed algorithm to learning rate variation.

In an effort to better understand the different ways in which the discount factor affects the optimization process in reinforcement learning, we designed a set of experiments to study each effect in isolation. Our analysis reveals that the common perception that poor performance of low discount factors is caused by (too) small action-gaps requires revision. We propose an alternative hypothesis, which identifies the size-difference of the action-gap across the state-space as the primary cause. We then introduce a new method that enables more homogeneous action-gaps by mapping value estimates to a logarithmic space. We prove convergence for this method under standard assumptions and demonstrate empirically that it indeed enables lower discount factors for approximate reinforcement-learning methods. This in turn allows tackling a class of reinforcement-learning problems that are challenging to solve with traditional methods.

Reinforcement learning methods carry a well known bias-variance trade-off in n-step algorithms for optimal control. Unfortunately, this has rarely been addressed in current research. This trade-off principle holds independent of the choice of the algorithm, such as n-step SARSA, n-step Expected SARSA or n-step Tree backup. A small n results in a large bias, while a large n leads to large variance. The literature offers no straightforward recipe for the best choice of this value. While currently all n-step algorithms use a fixed value of n over the state space we extend the framework of n-step updates by allowing each state to have its specific n. We propose a solution to this problem within the context of human aided reinforcement learning. Our approach is based on the observation that a human can learn more efficiently if she receives input regarding the criticality of a given state and thus the amount of attention she needs to invest into the learning in that state. This observation is related to the idea that each state of the MDP has a certain measure of criticality which indicates how much the choice of the action in that state influences the return. In our algorithm the RL agent utilizes the criticality measure, a function provided by a human trainer, in order to locally choose the best stepnumber n for the update of the Q function.