The pressure field around a two-dimensional gas bubble in a fluidized bed

Measurements of the entire pressure field around a two-dimensional bubble rising in an incipiently fluidized bed are reported. Above the center of pressure line, the field is Laplacian and reasonably well described by Davidson's theory. The pressure wake is closed. It extends for 12 bubble radii below the stagnation point and four radii on either side of the vertical axis of symmetry.     

Learning and planning in environments with delayed feedback

This work considers the problems of learning and planning in Markovian environments with constant observation and reward delays. We provide a hardness result for the general planning problem and positive results for several special cases with deterministic or otherwise constrained dynamics. We present an algorithm, Model Based Simulation, for planning in such environments and use model-based reinforcement learning to extend this approach to the learning setting in both finite and continuous environments. Empirical comparisons show this algorithm holds significant advantages over others for decision making in delayed-observation environments.     

Optimal one-dimensional apodizations and shaped pupils for planet finding coronagraphy

The realization that direct imaging of extrasolar planets could be technologically feasible within the next decade or so has inspired a great deal of recent research into high-contrast imaging. We have contributed several design ideas, all of which can be described as shaped pupil coronagraphs. We offer a complete and unified survey of one-dimensional shaped pupil designs, some of which have been published in our previous papers. We also introduce a promising new design, which we call bar-code masks. With these masks we can achieve the required contrast with a fairly large discovery zone and throughput, but most importantly they are perhaps the easiest to manufacture and might therefore stand up best to refined analyses.     

Gas-solid contacting in a two-dimensional fluidized bed containing mobile internals: Non-reactive tracer studies

The effect of 50.8 mm diameter disc shaped mobile internals (MIs) on gas-solid contacting was studied in a large two-dimensional fluidized bed by means of non-reactive tracers. MIs were found to improve substantially the mass transfer coefficient between the bubble and dense phases. However, they also increased the gas dispersion in the dense phase.     

An axisymmetric model of flow in the annulus of a spouted bed of coarse particles. Model,experimental verification and residence time distribution

An axisymmetric model of flow in the annulus of a spouted bed of coarse particles which takes into account non-Darcy flow and utilizes a new and general axial pressure boundary condition at the spout-annulus interface is presented. The pressure boundary condition and the annulus pressure distribution are experimentally verified and the model used to calculate the residence time distribution and the flowrate at the top of the annulus. In addition, the average axial velocity profile in the annulus is calculated and compared with predictions of several one-dimensional flow models.     

A unified analysis of value-function-based reinforcement- learning algorithms

Reinforcement learning is the problem of generating optimal behavior in a sequential decision-making environment given the opportunity of interacting with it. Many algorithms for solving reinforcement-learning problems work by computing improved estimates of the optimal value function. We extend prior analyses of reinforcement-learning algorithms and present a powerful new theorem that can provide a unified analysis of such value-function-based reinforcement-learning algorithms. The usefulness of the theorem lies in how it allows the convergence of a complex asynchronous reinforcement-learning algorithm to be proved by verifying that a simpler synchronous algorithm converges. We illustrate the application of the theorem by analyzing the convergence of Q-learning, model-based reinforcement learning, Q-learning with multistate updates, Q-learning for Markov games, and risk-sensitive reinforcement learning.