Portrait de Raihan Seraj n'est pas disponible

Raihan Seraj

Alumni

Publications

Approximate information state for approximate planning and reinforcement learning in partially observed systems
Jayakumar Subramanian
Amit Sinha
Raihan Seraj
Aditya Mahajan
We propose a theoretical framework for approximate planning and learning in partially observed systems. Our framework is based on the fundam… (voir plus)ental notion of information state. We provide two equivalent definitions of information state---i) a function of history which is sufficient to compute the expected reward and predict its next value; ii) equivalently, a function of the history which can be recursively updated and is sufficient to compute the expected reward and predict the next observation. An information state always leads to a dynamic programming decomposition. Our key result is to show that if a function of the history (called approximate information state (AIS)) approximately satisfies the properties of the information state, then there is a corresponding approximate dynamic program. We show that the policy computed using this is approximately optimal with bounded loss of optimality. We show that several approximations in state, observation and action spaces in literature can be viewed as instances of AIS. In some of these cases, we obtain tighter bounds. A salient feature of AIS is that it can be learnt from data. We present AIS based multi-time scale policy gradient algorithms. and detailed numerical experiments with low, moderate and high dimensional environments.
2021-12-31
Journal of Machine Learning Research (publié)
arxiv.org
arxiv.org
Mean-field approximation for large-population beauty-contest games
Raihan Seraj
Jerome Le Ny
Aditya Mahajan
We study a class of Keynesian beauty contest games where a large number of heterogeneous players attempt to estimate a common parameter base… (voir plus)d on their own observations. The players are rewarded for producing an estimate close to a certain multiplicative factor of the average decision, this factor being specific to each player. This model is motivated by scenarios arising in commodity or financial markets, where investment decisions are sometimes partly based on following a trend. We provide a method to compute Nash equilibria within the class of affine strategies. We then develop a mean-field approximation, in the limit of an infinite number of players, which has the advantage that computing the best-response strategies only requires the knowledge of the parameter distribution of the players, rather than their actual parameters. We show that the mean-field strategies lead to an ε-Nash equilibrium for a system with a finite number of players. We conclude by analyzing the impact on individual behavior of changes in aggregate population behavior.
2021-12-13
IEEE Conference on Decision and Control (publié)
doi.org
Entropy Regularization with Discounted Future State Distribution in Policy Gradient Methods
Riashat Islam
Raihan Seraj
Pierre-Luc Bacon
Doina Precup
The policy gradient theorem is defined based on an objective with respect to the initial distribution over states. In the discounted case, t… (voir plus)his results in policies that are optimal for one distribution over initial states, but may not be uniformly optimal for others, no matter where the agent starts from. Furthermore, to obtain unbiased gradient estimates, the starting point of the policy gradient estimator requires sampling states from a normalized discounted weighting of states. However, the difficulty of estimating the normalized discounted weighting of states, or the stationary state distribution, is quite well-known. Additionally, the large sample complexity of policy gradient methods is often attributed to insufficient exploration, and to remedy this, it is often assumed that the restart distribution provides sufficient exploration in these algorithms. In this work, we propose exploration in policy gradient methods based on maximizing entropy of the discounted future state distribution. The key contribution of our work includes providing a practically feasible algorithm to estimate the normalized discounted weighting of states, i.e, the \textit{discounted future state distribution}. We propose that exploration can be achieved by entropy regularization with the discounted state distribution in policy gradients, where a metric for maximal coverage of the state space can be based on the entropy of the induced state distribution. The proposed approach can be considered as a three time-scale algorithm and under some mild technical conditions, we prove its convergence to a locally optimal policy. Experimentally, we demonstrate usefulness of regularization with the discounted future state distribution in terms of increased state space coverage and faster learning on a range of complex tasks.
2019-12-10
ArXiv (prépublication)
arxiv.org
arxiv.org
Doubly Robust Off-Policy Actor-Critic Algorithms for Reinforcement Learning
Riashat Islam
Raihan Seraj
Samin Yeasar Arnob
Doina Precup
We study the problem of off-policy critic evaluation in several variants of value-based off-policy actor-critic algorithms. Off-policy actor… (voir plus)-critic algorithms require an off-policy critic evaluation step, to estimate the value of the new policy after every policy gradient update. Despite enormous success of off-policy policy gradients on control tasks, existing general methods suffer from high variance and instability, partly because the policy improvement depends on gradient of the estimated value function. In this work, we present a new way of off-policy policy evaluation in actor-critic, based on the doubly robust estimators. We extend the doubly robust estimator from off-policy policy evaluation (OPE) to actor-critic algorithms that consist of a reward estimator performance model. We find that doubly robust estimation of the critic can significantly improve performance in continuous control tasks. Furthermore, in cases where the reward function is stochastic that can lead to high variance, doubly robust critic estimation can improve performance under corrupted, stochastic reward signals, indicating its usefulness for robust and safe reinforcement learning.
2019-12-09
arXiv (prépublication)
doi.org
arxiv.org