Category Archives: Probability And Statistics

On the effects of large variances in the transition function for Q-learning

October 10, 2019 14:29 ,

D. Lee and W. B. Powell, Bias-Corrected Q-Learning With Multistate Extension. IEEE Transactions on Automatic Control, vol. 64, no. 10, pp. 4011-4023, DOI: 10.1109/TAC.2019.2912443.

Q-learning is a sample-based model-free algorithm that solves Markov decision problems asymptotically, but in finite time, it can perform poorly when random rewards and transitions result in large variance of value estimates. We pinpoint its cause to be the estimation bias due to the maximum operator in Q-learning algorithm, and present the evidence of max-operator bias in its Q value estimates. We then present an asymptotically optimal bias-correction strategy and construct an extension to bias-corrected Q-learning algorithm to multistate Markov decision processes, with asymptotic convergence properties as strong as those from Q-learning. We report the empirical performance of the bias-corrected Q-learning algorithm with multistate extension in two model problems: A multiarmed bandit version of Roulette and an electricity storage control simulation. The bias-corrected Q-learning algorithm with multistate extension is shown to control max-operator bias effectively, where the bias-resistance can be tuned predictably by adjusting a correction parameter.

Posted in: Reinforcement learning theory , Tagged:

A nice (short) survey of deep RL

May 10, 2019 11:43 ,

Matthew Botvinick, Sam Ritter, Jane X. Wang, Zeb Kurth-Nelson, Charles Blundell, Demis Hassabis, Reinforcement Learning, Fast and Slow, Trends in Cognitive Sciences, Volume 23, Issue 5, 2019, Pages 408-422 DOI: 10.1016/j.tics.2019.02.006.

Deep reinforcement learning (RL) methods have driven impressive advances in artificial intelligence in recent years, exceeding human performance in domains ranging from Atari to Go to no-limit poker. This progress has drawn the attention of cognitive scientists interested in understanding human learning. However, the concern has been raised that deep RL may be too sample-inefficient – that is, it may simply be too slow – to provide a plausible model of how humans learn. In the present review, we counter this critique by describing recently developed techniques that allow deep RL to operate more nimbly, solving problems much more quickly than previous methods. Although these techniques were developed in an AI context, we propose that they may have rich implications for psychology and neuroscience. A key insight, arising from these AI methods, concerns the fundamental connection between fast RL and slower, more incremental forms of learning.

Posted in: Psycho-physiological bases of engineering, Reinforcement learning theory , Tagged: ,

An interesting review of criticisms of deep learning in cognitive science

March 20, 2019 10:22 ,

Radoslaw M. Cichy, Daniel Kaiser, Deep Neural Networks as Scientific Models, Trends in Cognitive Sciences, Volume 23, Issue 4, 2019, Pages 305-317, DOI: 10.1016/j.tics.2019.01.009.

Artificial deep neural networks (DNNs) initially inspired by the brain enable computers to solve cognitive tasks at which humans excel. In the absence of explanations for such cognitive phenomena, in turn cognitive scientists have started using DNNs as models to investigate biological cognition and its neural basis, creating heated debate. Here, we reflect on the case from the perspective of philosophy of science. After putting DNNs as scientific models into context, we discuss how DNNs can fruitfully contribute to cognitive science. We claim that beyond their power to provide predictions and explanations of cognitive phenomena, DNNs have the potential to contribute to an often overlooked but ubiquitous and fundamental use of scientific models: exploration.

Posted in: Machine learning , Tagged: ,

A new framework for fitting jump models

September 4, 2018 15:23 ,

Alberto Bemporad, Valentina Breschi, Dario Piga, Stephen P. Boyd, Fitting jump models, Automatica, Volume 96, 2018, Pages 11-21, DOI: 10.1016/j.automatica.2018.06.022.

We describe a new framework for fitting jump models to a sequence of data. The key idea is to alternate between minimizing a loss function to fit multiple model parameters, and minimizing a discrete loss function to determine which set of model parameters is active at each data point. The framework is quite general and encompasses popular classes of models, such as hidden Markov models and piecewise affine models. The shape of the chosen loss functions to minimize determines the shape of the resulting jump model.

Posted in: Probability and statistics , Tagged: , ,

Regression to help in finding the optimal policy in MDPs based on duality theory

April 26, 2018 07:44 ,

H. Zhu, F. Ye and E. Zhou, Solving the Dual Problems of Dynamic Programs via Regression, IEEE Transactions on Automatic Control, vol. 63, no. 5, pp. 1340-1355, DOI: 10.1109/TAC.2017.2747405.

In recent years, information relaxation and duality in dynamic programs have been studied extensively, and the resulted primal-dual approach has become a powerful procedure in solving dynamic programs by providing lower-upper bounds on the optimal value function. Theoretically, with the so-called value-based optimal dual penalty, the optimal value function could be recovered exactly via strong duality. However, in practice, obtaining tight dual bounds usually requires good approximations of the optimal dual penalty, which could be time consuming if analytical computation is not possible and nested simulation has to be used to estimate the conditional expectations inside the dual penalty. In this paper, we will develop a framework of a regression approach to approximating the optimal dual penalty in a nonnested manner, by exploring the structure of the function space consisting of all feasible dual penalties. The resulted approximations maintain to be feasible dual penalties, and thus yielding valid dual bounds on the optimal value function. We show that the proposed framework is computationally efficient, and the resulted dual penalties lead to numerically tractable dual problems. Finally, we apply the framework to a high-dimensional dynamic trading problem to demonstrate its effectiveness in solving the dual problems of complex dynamic programs.

Posted in: Probability and statistics , Tagged: ,

POMDPs aware of the data association problem

March 29, 2018 09:32 ,

Shashank Pathak, Antony Thomas, and Vadim Indelman, A unified framework for data association aware robust belief space planning and perception, The International Journal of Robotics Research Vol 37, Issue 2-3, pp. 287 – 315, DOI: 10.1177/0278364918759606.

We develop a belief space planning approach that advances the state of the art by incorporating reasoning about data association within planning, while considering additional sources of uncertainty. Existing belief space planning approaches typically assume that data association is given and perfect, an assumption that can be harder to justify during operation in the presence of localization uncertainty, or in ambiguous and perceptually aliased environments. By contrast, our data association aware belief space planning (DA-BSP) approach explicitly reasons about data association within belief evolution owing to candidate actions, and as such can better accommodate these challenging real-world scenarios. In particular, we show that, owing to perceptual aliasing, a posterior belief can become a mixture of probability distribution functions and design cost functions, which measure the expected level of ambiguity and posterior uncertainty given candidate action. Furthermore, we also investigate more challenging situations, such as when prior belief is multimodal and when data association aware planning is performed over several look-ahead steps. Our framework models the belief as a Gaussian mixture model. Another unique aspect of this approach is that the number of components of this Gaussian mixture model can increase as well as decrease, thereby reflecting reality more accurately. Using these and standard costs (e.g. control penalty, distance to goal) within the objective function yields a general framework that reliably represents action impact and, in particular, is capable of active disambiguation. Our approach is thus applicable to both robust perception in a passive setting with data given a priori and in an active setting, such as in autonomous navigation in perceptually aliased environments. We demonstrate key aspects of DA-BSP in a theoretical example, in a Gazebo-based realistic simulation, and also on the real robotic platform using a Pioneer robot in an office environment.

Posted in: Probability and statistics, Robot task planning , Tagged: ,

Using EKF estimation in a PI controller for improving its performance under noise

March 29, 2018 09:29 ,

Y. Zhou, Q. Zhang, H. Wang, P. Zhou and T. Chai, EKF-Based Enhanced Performance Controller Design for Nonlinear Stochastic Systems, IEEE Transactions on Automatic Control, vol. 63, no. 4, pp. 1155-1162, DOI: 10.1109/TAC.2017.2742661.

In this paper, a novel control algorithm is presented to enhance the performance of the tracking property for a class of nonlinear and dynamic stochastic systems subjected to non-Gaussian noises. Although the existing standard PI controller can be used to obtain the basic tracking of the systems, the desired tracking performance of the stochastic systems is difficult to achieve due to the random noises. To improve the tracking performance, an enhanced performance loop is constructed using the EKF-based state estimates without changing the existing closed loop with a PI controller. Meanwhile, the gain of the enhanced performance loop can be obtained based upon the entropy optimization of the tracking error. In addition, the stability of the closed loop system is analyzed in the mean-square sense. The simulation results are given to illustrate the effectiveness of the proposed control algorithm.

Posted in: Bayesian filtering, Control Engineering , Tagged: ,

A novel method of mathematical compression of the value function for polynomial (in the state) time complexity of value iteration / policy iteration

March 27, 2018 09:01 ,

Alex Gorodetsky, Sertac Karaman, and Youssef Marzouk, High-dimensional stochastic optimal control using continuous tensor decompositions, The International Journal of Robotics Research Vol 37, Issue 2-3, pp. 340 – 377, DOI: 10.1177/0278364917753994.

Motion planning and control problems are embedded and essential in almost all robotics applications. These problems are often formulated as stochastic optimal control problems and solved using dynamic programming algorithms. Unfortunately, most existing algorithms that guarantee convergence to optimal solutions suffer from the curse of dimensionality: the run time of the algorithm grows exponentially with the dimension of the state space of the system. We propose novel dynamic programming algorithms that alleviate the curse of dimensionality in problems that exhibit certain low-rank structure. The proposed algorithms are based on continuous tensor decompositions recently developed by the authors. Essentially, the algorithms represent high-dimensional functions (e.g. the value function) in a compressed format, and directly perform dynamic programming computations (e.g. value iteration, policy iteration) in this format. Under certain technical assumptions, the new algorithms guarantee convergence towards optimal solutions with arbitrary precision. Furthermore, the run times of the new algorithms scale polynomially with the state dimension and polynomially with the ranks of the value function. This approach realizes substantial computational savings in “compressible” problem instances, where value functions admit low-rank approximations. We demonstrate the new algorithms in a wide range of problems, including a simulated six-dimensional agile quadcopter maneuvering example and a seven-dimensional aircraft perching example. In some of these examples, we estimate computational savings of up to 10 orders of magnitude over standard value iteration algorithms. We further demonstrate the algorithms running in real time on board a quadcopter during a flight experiment under motion capture.

Posted in: Probability and statistics, Robot motion planning , Tagged: , , , ,

Improving the estimation of the offset parameter of heavy-tailed distributions through the injection of noise

March 8, 2018 09:17 ,

Y. Pan, F. Duan, F. Chapeau-Blondeau and D. Abbott, Noise Enhancement in Robust Estimation of Location, IEEE Transactions on Signal Processing, vol. 66, no. 8, pp. 1953-1966, DOI: 10.1109/TSP.2018.2802463.

In this paper, we investigate the noise benefits to maximum likelihood type estimators (M-estimator) for the robust estimation of a location parameter. Two distinct noise benefits are shown to be accessible under these conditions. With symmetric heavy-tailed noise distributions, the asymptotic efficiency of the estimation can be enhanced by injecting extra noise into the M-estimators. With an asymmetric contaminated noise model having a convex cumulative distribution function, we demonstrate that addition of noise can reduce the maximum bias of the median estimator. These findings extend the analysis of stochastic resonance effects for noise-enhanced signal and information processing.

Posted in: Probability and statistics , Tagged: ,

A novel approach to use POMDP in practical active perception, where rewards are needed to penalize uncertainty and therefore reomve the piecewise-linear and convex property of the value function

February 11, 2018 09:08 ,

Satsangi, Y., Whiteson, S., Oliehoek, F.A. et al., Exploiting submodular value functions for scaling up active perception, Auton Robot (2018) 42: 209, DOI: 10.1007/s10514-017-9666-5.

In active perception tasks, an agent aims to select sensory actions that reduce its uncertainty about one or more hidden variables. For example, a mobile robot takes sensory actions to efficiently navigate in a new environment. While partially observable Markov decision processes (POMDPs) provide a natural model for such problems, reward functions that directly penalize uncertainty in the agent’s belief can remove the piecewise-linear and convex (PWLC) property of the value function required by most POMDP planners. Furthermore, as the number of sensors available to the agent grows, the computational cost of POMDP planning grows exponentially with it, making POMDP planning infeasible with traditional methods. In this article, we address a twofold challenge of modeling and planning for active perception tasks. We analyze ρ POMDP and POMDP-IR, two frameworks for modeling active perception tasks, that restore the PWLC property of the value function. We show the mathematical equivalence of these two frameworks by showing that given a ρ POMDP along with a policy, they can be reduced to a POMDP-IR and an equivalent policy (and vice-versa). We prove that the value function for the given ρ POMDP (and the given policy) and the reduced POMDP-IR (and the reduced policy) is the same. To efficiently plan for active perception tasks, we identify and exploit the independence properties of POMDP-IR to reduce the computational cost of solving POMDP-IR (and ρ POMDP). We propose greedy point-based value iteration (PBVI), a new POMDP planning method that uses greedy maximization to greatly improve scalability in the action space of an active perception POMDP. Furthermore, we show that, under certain conditions, including submodularity, the value function computed using greedy PBVI is guaranteed to have bounded error with respect to the optimal value function. We establish the conditions under which the value function of an active perception POMDP is guaranteed to be submodular. Finally, we present a detailed empirical analysis on a dataset collected from a multi-camera tracking system employed in a shopping mall. Our method achieves similar performance to existing methods but at a fraction of the computational cost leading to better scalability for solving active perception tasks.

Posted in: Probability and statistics, Robotics , Tagged: ,