Learning to Perceive and Act by Trial and Error

This article considers adaptive control architectures that integrate active sensory-motor systems with decision systems based on reinforcement learning. One unavoidable consequence of active perception is that the agent's internal representation often confounds external world states. We call this phoenomenon perceptual aliasingand show that it destabilizes existing reinforcement learning algorithms with respect to the optimal decision policy. We then describe a new decision system that overcomes these difficulties for a restricted class of decision problems. The system incorporates a perceptual subcycle within the overall decision cycle and uses a modified learning algorithm to suppress the effects of perceptual aliasing. The result is a control architecture that learns not only how to solve a task but also where to focus its visual attention in order to collect necessary sensory information.     

Electrophoretic deposition of monochrome and color phosphor screens for information displays

This paper reviews our research on the electrophoretic deposition (EPD) of phosphors for the processing of monochromatic and color screens for information displays. Our investigation began with the study of the fundamentals of the EPD process for phosphors. The processing variables which enhance the adhesion strength of phosphor deposits were determined. The optical performance of phosphors deposited by EPD was shown to be not affected by the process itself nor by the conditions which enhance phosphor adhesion. Processes developed to produce high-resolution color screens by combining EPD and photolithography techniques are described. Also, a method to electrophoretically deposit phosphor in a thermo-reversible gel from mixtures of poly(butyl methacrylate) and isopropanol was examined.     

Progress in the development of polymer cholesteric liquid crystal flakes for display applications

Polymer cholesteric liquid crystal (PCLC) flake technology is being developed as an alternative display technology for flexible, reflective particle displays. The motion of PCLC flakes suspended in a host fluid can be controlled with an electric field, creating a way to electrically control the flakes' ability to brightly reflect light that is circularly polarized. The PCLC flake/host fluid dispersion has been successfully microencapsulated both in a polymer matrix and in gelatin micro-capsules. Microencapsulation will not only expand the applications scope of the technology, but also may aid in addressing some potential problem areas that are inherent to many forms of particle display technology. A second important development in PCLC flake technology involves the manufacture of shaped flakes based on soft lithography techniques. The size and shape of a flake impact its reorientation, and uniformly shaped flakes respond in a similar manner. The unique reflective properties of PCLC flakes also provide possible applications in areas such as optics and photonics, switchable ‘smart windows’ or conformal coatings, and information displays such as ‘electronic paper.’     

A classification of tournaments having an acyclic tournament as a minimum feedback arc set

Given a tournament with an acyclic tournament as a feedback arc set we give necessary and sufficient conditions for this feedback arc set to have minimum size.     

Image denoising based on wavelets and multifractals for singularity detection.

This paper presents a very efficient algorithm for image denoising based on wavelets and multifractals for singularity detection. A challenge of image denoising is how to preserve the edges of an image when reducing noise. By modeling the intensity surface of a noisy image as statistically self-similar multifractal processes and taking advantage of the multiresolution analysis with wavelet transform to exploit the local statistical self-similarity at different scales, the pointwise singularity strength value characterizing the local singularity at each scale was calculated. By thresholding the singularity strength, wavelet coefficients at each scale were classified into two categories: the edge-related and regular wavelet coefficients and the irregular coefficients. The irregular coefficients were denoised using an approximate minimum mean-squared error (MMSE) estimation method, while the edge-related and regular wavelet coefficients were smoothed using the fuzzy weighted mean (FWM) filter aiming at preserving the edges and details when reducing noise. Furthermore, to make the FWM-based filtering more efficient for noise reduction at the lowest decomposition level, the MMSE-based filtering was performed as the first pass of denoising followed by performing the FWM-based filtering. Experimental results demonstrated that this algorithm could achieve both good visual quality and high PSNR for the denoised images.     

A Pervasive Computing Curriculum for Engineering and Science Students

Today, there's an urgent need for well-trained pervasive computing engineers and scientists. Pervasive computing has evolved into a critical field, influencing a broad range of industries from healthcare to manufacturing to defense. However, most computer engineering and computer science programs pay marginal attention, at best, to the field. In this issue, the authors of the Rochester Institute of Technology describes a set of courses on pervasive computing topics and a supporting laboratory designed for both computer engineering and computer science students     

Intermediate order in tetrahedrally coordinated silicon: evidence for chainlike objects

In this report, we describe the nature of intermediate order in silicon as determined by recent measurements on thin films using transmission electron microscopy (TEM) and Raman scattering. The TEM images show in addition to the expected continuous random network (CRN), the presence of highly ordered quasi-one-dimensional “chain-like objects” (CLO's) that are 1–2 nm wide and tens of nm long that meander and show some evidence of cross-linking with each other. The presence of these objects correlate to a Raman feature centered at 490 cm⁻¹ whose width is 35–40 cm⁻¹, and is used to quantify the heterogeneity in terms of the CLO and CRN (=475 cm⁻¹ scattering) concentrations. The 490 and 35 cm⁻¹ values are consistent with bond angle deviations approaching 0°, and thus reinforces an association with the CLOs. We find that in reference quality a-Si:H (made using pure SiH₄), the CLO concentration is about 5 vol%, while in state-of-the-art material using high H₂ levels of dilution during processing, it increases to about 15%. Increased stability of such material to light-soaking is thus not mediated by a direct volumetric replacement of poor with high-quality components. Rather, an important characteristic of intermediate order in silicon is the low-dimensional aspect of its order, which allows it to influence more total volume than which it is itself composed. Consistent with these and other recent findings, we propose a tensegrity model of amorphous silicon.     

Learning Automata from Ordered Examples

Connectionist learning models have had considerable empirical success, but it is hard to characterize exactly what they learn. The learning of finite-state languages (FSL) from example strings is a domain which has been extensively studied and might provide an opportunity to help understand connectionist learning. A major problem is that traditional FSL learning assumes the storage of all examples and thus violates connectionist principles. This paper presents a provably correct algorithm for inferring any minimum-state deterministic finite-state automata (FSA) from a complete ordered sample using limited total storage and without storing example strings. The algorithm is an iterative strategy that uses at each stage a current encoding of the data considered so far, and one single sample string. One of the crucial advantages of our algorithm is that the total amount of space used in the course of learning for encoding any finite prefix of the sample is polynomial in the size of the inferred minimum state deterministic FSA. The algorithm is also relatively efficient in time and has been implemented. More importantly, there is a connectionist version of the algorithm that preserves these properties. The connectionist version requires much more structure than the usual models and has been implemented using the Rochester Connectionist Simulator. We also show that no machine with finite working storage can iteratively identify the FSL from arbitrary presentations.