This PDF file contains the front matter associated with SPIE Proceedings Volume 8750 including the Title Page, Copyright information, Table of Contents, Introduction, and Conference Committee listing.

Before the invention of orthogonal wavelet systems by Yves Meyer¹ in 1986 Gabor expansions (viewed as discretized inversion of the Short-Time Fourier Transform² using the overlap and add OLA) and (what is now perceived as) wavelet expansions have been treated more or less at an equal footing. The famous paper on painless expansions by Daubechies, Grossman and Meyer³ is a good example for this situation. The description of atomic decompositions for functions in modulation spaces⁴ (including the classical Sobolev spaces) given by the author⁵ was directly modeled according to the corresponding atomic characterizations by Frazier and Jawerth,^{6, 7} more or less with the idea of replacing the dyadic partitions of unity of the Fourier transform side by uniform partitions of unity (so-called BUPU's, first named as such in the early work on Wiener-type spaces by the author in 1980⁸). Watching the literature in the subsequent two decades one can observe that the interest in wavelets "took over", because it became possible to construct orthonormal wavelet systems with compact support and of any given degree of smoothness,⁹ while in contrast the Balian-Low theorem is prohibiting the existence of corresponding Gabor orthonormal bases, even in the multi-dimensional case and for general symplectic lattices.¹⁰ It is an interesting historical fact that^* his construction of band-limited orthonormal wavelets (the Meyer wavelet, see¹¹) grew out of an attempt to prove the impossibility of the existence of such systems, and the final insight was that it was not impossible to have such systems, and in fact quite a variety of orthonormal wavelet system can be constructed as we know by now. Meanwhile it is established wisdom that wavelet theory and time-frequency analysis are two different ways of decomposing signals in orthogonal resp. non-orthogonal ways. The unifying theory, covering both cases, distilling from these two situations the common group theoretical background lead to the theory of coorbit spaces,^{12, 13} established by the author jointly with K. Gröchenig. Starting from an integrable and irreducible representation of some locally compact group (such as the "ax+b"-group or the Heisenberg group) one can derive families of Banach spaces having natural atomic characterizations, or alternatively a continuous transform associated to it. So at the end function spaces of locally compact groups come into play, and their generic properties help to explain why and how it is possible to obtain (nonorthogonal) decompositions. While unification of these two groups was one important aspect of the approach given in the late 80th, it was also clear that this approach allows to formulate and exploit the analogy to Banach spaces of analytic functions invariant under the Moebius group have been at the heart in this context. Recent years have seen further new instances and generalizations. Among them shearlets or the Blaschke product should be mentioned here, and the increased interest in the connections between wavelet theory and complex analysis. The talk will try to summarize a few of the general principles which can be derived from the general theory, but also highlight the difference between the different groups and signal expansions arising from corresponding group representations. There is still a lot more to be done, also from the point of view of applications and the numerical realization of such non-orthogonal expansions.

We shall introduce a novel methodology for data reconstruction and recovery, based on composite wavelet representations. These representations include shearlets and crystallographic wavelets, among others, and they allow for an increased directional sensitivity in comparison with the standard multiscale techniques. Our new approach allows us to recover missing data, due to sparsity of composite wavelet representations, especially when compared to inpainting algorithms induced by traditional wavelet representations, and also due to the flexibility of our variational approach.

We propose a framework for analyzing and visualizing data at multiple scales and directions by constructing a novel class of tight frames. We describe an elegant way of creating 2D tight frames from 1D sets of orthonormal vectors and show how to exploit the representation redundancy in a computationally efficient manner. Finally, we employ this framework to perform image superresolution via edge detection and characterization.

The paper deals with the increasing accuracy of voice authentication methods. The developed algorithm first extracts segmental parameters, such as Zero Crossing Rate, the Fundamental Frequency and Mel-frequency cepstral coefficients from voice. Based on these parameters, the neural network classifier detects the speaker's emotional state. These parameters shape the distribution of neurons in Kohonen maps, forming clusters of neurons on the map characterizing a particular emotional state. Using regression analysis, we can calculate the function of the parameters of individual emotional states. This relationship increases voice authentication accuracy and prevents unjust rejection.

This paper presents various source separation methods that utilize multiple microphones. We classify them into two classes. Methods that fall into the first class apply independent component analysis (ICA) or Gaussian mixture model (GMM) to frequency bin-wise observations, and then solve the permutation problem to reconstruct separated signals. The second type of method extends non-negative matrix factorization (NMF) to a multimicrophone situation, in which NMF bases are clustered according to their spatial properties. We have a unified understanding that all methods analyze a time-frequency representation with an additional microphone axis.

Automatic video segmentation for human activity recognition has played an important role in several computer vision applications. Active contour model (ACM) has been used extensively for unsupervised adaptive segmentation and automatic object region and boundary extraction in video sequences. This paper presents optimizing Active Contour Model using recurrent architecture for automatic object region and boundary extraction in human activity video sequences. Taking advantage of the collective computational ability and energy convergence capability of the recurrent architecture, energy function of Active Contour Model is optimized with lower computational time. The system starts with initializing recurrent architecture state based on the initial boundary points and ends up with final contour which represent actual boundary points of human body region. The initial contour of the Active Contour Model is computed using background subtraction based on Gaussian Mixture Model (GMM) such that background model is built dynamically and regularly updated to overcome different challenges including illumination changes, camera oscillations, and changes in background geometry. The recurrent nature is useful for dealing with optimization problems due to its dynamic nature, thus, ensuring convergence of the system. The proposed boundary detection and region extraction can be used for real time processing. This method results in an effective segmentation that is less sensitive to noise and complex environments. Experiments on different databases of human activity show that our method is effective and can be used for real-time video segmentation.

Non-mass enhancing lesions represent one of the most challenging types of lesions when it comes to both manual and computer-assisted diagnosis. Compared to the well-characterized mass-enhancing lesions, non-masses have not well-defined and blurred tumor borders and a kinetic behavior that is not easily generalizable and thus non-discriminative for malignant and benign non-masses. A valuable feature descriptor should capture the heterogeneity of enhancement as well as the speed of enhancement in the tissue. We apply and evaluate both textural and spatio-temporal descriptors to the pertinent feature extraction of these lesions. An automated computer-aided diagnosis system evaluates the atypical behavior of these lesions, and additionally considers the impact of non-rigid motion compensation on a correct diagnosis.

This paper explores the application of a wavelet neural network (WNN), whose hidden layer is comprised of neurons with adjustable wavelets as activation functions, to stock prediction. We discuss some basic rationales behind technical analysis, and based on which, inputs of the prediction system are carefully selected. This system is tested on Istanbul Stock Exchange National 100 Index and compared with traditional neural networks. The results show that the WNN can achieve very good prediction accuracy.

Spectral interference arising from direct, wing or background-induced spectral overlaps are a key concern in optical emission spectrometry even if an optical spectrometer with a 1 m focal length is used (thus resulting in peaks with halfwidth of ~80 pm). The problem of spectral interferences becomes even more acute when a portable spectrometer with a relatively short focal length (e.g., 10-15 cm) is used. In our lab, we are addressing spectral interference correction methods using artificial neural networks (ANNs) and partial least squares (PLS). In this paper, the application of ANNS and of PLS for spectral interference correction is compared using spectral simulations (to avoid the effects of 1/f noise).

The paper discusses the use of Discrete Wavelet Transform (DWT) and Stationary Wavelet Transform (SWT) wavelet in removing noise from voice samples and evaluation of its impact on speech quality. One significant part of Quality of Service (QoS) in communication technology is the speech quality assessment. However, this part is seriously overlooked as telecommunication providers often focus on increasing network capacity, expansion of services offered and their enforcement in the market. Among the fundamental factors affecting the transmission properties of the communication chain is noise, either at the transmitter or the receiver side. A wavelet transform (WT) is a modern tool for signal processing. One of the most significant areas in which wavelet transforms are used is applications designed to suppress noise in signals. To remove noise from the voice sample in our experiment, we used the reference segment of the voice which was distorted by Gaussian white noise. An evaluation of the impact on speech quality was carried out by an intrusive objective algorithm Perceptual Evaluation of Speech Quality (PESQ). DWT and SWT transformation was applied to voice samples that were devalued by Gaussian white noise. Afterwards, we determined the effectiveness of DWT and SWT by means of objective algorithm PESQ. The decisive criterion for determining the quality of a voice sample once the noise had been removed was Mean Opinion Score (MOS) which we obtained in PESQ. The contribution of this work lies in the evaluation of efficiency of wavelet transformation to suppress noise in voice samples.

We introduce a new algorithm for multichannel blind deconvolution. Given the outputs of K linear time- invariant channels driven by a common source, we wish to recover their impulse responses without knowledge of the source signal. Abstractly, this problem amounts to finding a solution to an overdetermined system of quadratic equations. We show how we can recast the problem as solving a system of underdetermined linear equations with a rank constraint. Recent results in the area of low rank recovery have shown that there are effective convex relaxations to problems of this type that are also scalable computationally, allowing us to recover 100s of channel responses after a moderate observation time. We illustrate the effectiveness of our methodology with a numerical simulation of a passive noise imaging" experiment.

Ambiguity in binocular ranging (David Marr’s paradox) may be resolved by using two eyes moving from side to side behind an optical bench while integrating multiple views. Moving a head from left to right with one eye closed can also help resolve the foreground and background range uncertainty. That empirical experiment implies redundancy in image data, which may be reduced by adopting a 3-D camera imaging model to perform compressive sensing. Here, the compressive sensing concept is examined from the perspective of redundancy reduction in images subject to diurnal and weather variations for the purpose of resolving range uncertainty at all weather conditions such as the dawn or dusk, the daytime with different light level or the nighttime at different spectral band. As an example, a scenario at an intersection of a country road at dawn/dusk is discussed where the location of the traffic signs needs to be resolved by passive ranging to answer whether it is located on the same side of the road or the opposite side, which is under the influence of temporal light/color level variation. A spectral band extrapolation via application of Lagrange Constrained Neural Network (LCNN) learning algorithm is discussed to address lost color restoration at dawn/dusk. A numerical simulation is illustrated along with the code example.

Automatic and real-time face recognition can be applied into many attractive applications. For example, at a checkpoint it is expected that there are no burdens on a passing person and a security guard in addition to low cost. Normally a unique 3D person is projected into 2D images with information loss. It means a person is no longer unique in 2D space. Furthermore the various conditions such as pose variance, illumination variance and different expression make face recognition difficult. In order to separate a person, his or her subspace should have several faces and be redundant. That is why the database naturally becomes large. Under this situation the efficient face recognition is a key to a surveillance system. Face recognition by spars representation classification (SRC) could be one of promising candidates to realize rapid face recognition. This method can be understood in a similar way to compressive sensing (CS). In this paper, we propose the efficient approach of face recognition by SRC for multiple poses from the viewpoint of CS. The part-cropped database (PCD) is suggested to avoid position misalignments by discarding the information of topological linkages among eyes, a nose and a mouth. Although topological linkages are important for face recognition in general, they cause position misalignments among multiple poses which decrease recognition rate. Our approach solves one of trade-off problem between keeping topological linkages and avoiding position misalignments. According to the simulated experiments, PCD works well to avoid position misalignments and acquires correct recognition despite less information on topological linkages.

For the past two decades, there has been an ongoing research effort at Los Alamos National Laboratory to learn more about the Earth’s radiofrequency (RF) background utilizing satellite-based RF observations of terrestrial lightning. The Fast On-orbit Recording of Transient Events (FORTE) satellite provided a rich RF lighting database, comprising of five years of data recorded from its two RF payloads. While some classification work has been done previously on the FORTE RF database, application of modern pattern recognition techniques may advance lightning research in the scientific community and potentially improve on-orbit processing and event discrimination capabilities for future satellite payloads. We now develop and implement new event classification capability on the FORTE database using state-of-the-art adaptive signal processing combined with compressive sensing and machine learning techniques. The focus of our work is improved feature extraction using sparse representations in learned dictionaries. Conventional localized data representations for RF transients using analytical dictionaries, such as a short-time Fourier basis or wavelets, can be suitable for analyzing some types of signals, but not others. Instead, we learn RF dictionaries directly from data, without relying on analytical constraints or additional knowledge about the signal characteristics, using several established machine learning algorithms. Sparse classification features are extracted via matching pursuit search over the learned dictionaries, and used in conjunction with a statistical classifier to distinguish between lightning types. We present preliminary results of our work and discuss classification scenarios and future development.

Pulse and respiration rates provide vital information for evaluating the physiological state of an individual during triage. Traditionally, pulse and respiration have been tracked by means of contact sensors. Recent work has shown that visible cameras can passively and remotely obtain pulse signals under controlled environmental conditions [2] [5] [14] [27]. This paper introduces methods for extracting and characterizing pulse and respiration signals from skin reflectivity data captured in peak sensitivity range for silicon detector (400nm-1100nm). Based on the physiological understanding [12] [13] [15] of human skin and reflectivity at various skin depths, we optimize a group of spectral bands to determine pulse and respiration with high Peak Signal-to-Noise Ratio (PSNR) and correlation values [27] [30]. Our preliminary results indicate top six optimal waveband groups in about 100nm - 200nm resolution in each, with rank-ordered peaks at 409nm, 512nm, 584nm, 667nm, 885nm and 772nm. This work, collected under an approved IRB protocol enhances non-contact, remote, passive, and real-time measurement of pulse and respiration for security and medical applications.

In recent years, sparse coding has drawn considerable research attention in developing feature representations for visual recognition problems. In this paper, we devise sparse coding algorithms to learn a dictionary of basis functions from Scale- Invariant Feature Transform (SIFT) descriptors extracted from images. The learned dictionary is used to code SIFT-based inputs for the feature representation that is further pooled via spatial pyramid matching kernels and fed into a Support Vector Machine (SVM) for object classification on the large-scale ImageNet dataset. We investigate the advantage of SIFT-based sparse coding approach by combining different dictionary learning and sparse representation algorithms. Our results also include favorable performance on different subsets of the ImageNet database.

We have recently introduced a progressive compressing sensing method based on an appropriate sampling scheme of optical Radon projections. By choosing the sampling steps to be of the size of the golden angle new information is optimally acquired with each angular sampling step thus permitting gradual improvement of the reconstructed image. A comparison between the progressive sampling scheme with the conventional one based on uniform sampling is given in terms of their coherence parameter.

Calcification detection in mammogram is important in breast cancer diagnosis. A super-resolution reconstruction method is proposed to reconstruct mammogram image from one single low resolution mammogram based on the compressed sensing by the contourlet transform. The initial estimation of the super-resolution mammogram is obtained by the interpolation method of the low resolution mammogram reconstructed by compressed sensing, then contourlet transform is applied respectively to the initial estimation and the reconstructed low resolution mammogram. From the statistical characteristics of the mutiscale frequency bands between the initial estimation and the reconstructed low resolution mammogram, the thresholds are estimated to integrate the high frequency of the initial estimation and the low frequency of the reconstructed low resolution mammogram. The super-resolution mammogram is achieved through the reconstruction of contourlet inverse transform. The proposed method can retrieve some details of the low resolution images. The calcification in mammogram can be detected efficiently.

In this paper, a denoise approach is proposed to reduce the speckle noise in SAR images based on compress sensing. Through the skill of compressed sensing, we divide the image into some blocks, and propose an image reconstruction method based on block compressing sensing with Orthogonal Matching Pursuit. By adding some simulated speckle noise in the SAR image, the performance of the proposed approach is shown and compared with a conventional algorithm. the result has been shown that our method can get better result in terms of peak signal noise ratio (PSNR).

At the nanoscale, three dimensional manipulation and assembly becomes extremely challenging and also cost prohibitive. Self-assembly provides an attractive and possibly the only highly parallel methodology to structure truly three dimensional patterned materials and devices at this size scale for applications in electronics, optics, robotics and medicine. This is a concise review along with a perspective of an important and exciting field in nanotechnology and is related to a Nanoengineering Pioneer Award that I received at this SPIE symposium for my contributions to the 3D selfassembly of nanostructures. I detail a historical account of 3D self-assembly and outline important developments in this area which is put into context with the larger research areas of 3D nanofabrication, assembly and nanomanufacturing. A focus in this review is on our work as it relates to the self-assembly with lithographically patterned units; this approach provides a means for heterogeneous integration of periodic, curved and angled nanostructures with precisely defined three dimensional patterns.

This paper focuses on the optimization of video coding standards motion estimation algorithms using Altera Custom Instructions based-paradigm and the combination of SDRAM with On-Chip memory in NIOS II processors. On one hand a complete algorithm profiling is achieved before the optimization, in order to find the code time leaks, afterward is developing a custom instruction set which will be added to the specific embedded design enhancing the original system. On the other hand, all possible permitted memories combinations between On-Chip memory and SDRAM have been tested for achieving the best performance combination. The final performance of the final design (memory optimization and custom instruction acceleration) is shown. This contribution, thus, outlines a low cost system, mapped on a Very Large Scale Integration (VLSI) technology which accelerates software algorithms by converting them to custom hardware logic block and shows the best combination between On-Chip memory and SDRAM for the NIOS II processor.

Non-mass enhancing lesions represent one of the most challenging types of lesions for both the clinician as well as current computer-aided diagnosis (CAD) systems. Differently from the well-studied mass-enhancing tumors these lesions do not exhibit a typical kinetic behavior that can be further easily categorized into benign or malignant based on feature descriptors. Furthermore, the poorly defined tumor borders pose a difficulty to even the most sophisticated segmentation algorithms. To address these challenges in terms of segmentation and atypical contrast enhancement dynamics, we apply an ICA-based segmentation on these lesions and extract from the average signal intensity curve of the most representative independent component (IC). Subsequently the dynamics of this IC is modeled based on mathematical models such as the empirical mathematical model and the phenomenological universalities. An automated computer-aided diagnosis system evaluates the atypical behavior of these lesions, and additionally compares the benefit of ICA-segmentation versus active contour segmentation.

In this paper, a modified particle swarm optimization (PSO) approach, particle swarm optimization with ε- greedy exploration εPSO), is used to tackle the object tracking. In the modified εPSO algorithm, the cooperative learning mechanism among individuals has been introduced, namely, particles not only adjust its own flying speed according to itself and the best individual of the swarm but also learn from other best individuals according to certain probability. This kind of biologically-inspired mutual-learning behavior can help to find the global optimum solution with better convergence speed and accuracy. The εPSO algorithm has been tested on benchmark function and demonstrated its effectiveness in high-dimension multi-modal optimization. In addition to the standard benchmark study, we also combined our new εPSO approach with the traditional particle filter (PF) algorithm on the object tracking task, such as car tracking in complex environment. Comparative studies between our εPSO combined PF algorithm with those of existing techniques, such as the particle filter (PF) and classic PSO combined PF will be used to verify and validate the performance of our approach.

Visual saliency is a bottom-up process that identifies those regions in an image that stand out from their surroundings. We oversegment an image as a collection of “super pixels” (SPs). Each SP is salient if it is different in color from all other SPs and if its most similar SPs are nearby. We test our method on image sequences collected by a vehicle. We consider an SP in a frame as salient if it stands out from all frames in a collection that consists of an ensemble of images from different road segments and a sequence of immediate past frames.

Proposed is a new fast algorithm for entropy estimation of a given input word. The algorithm utilizes k-nearest neighbor search of a given dictionary. The time complexity of the search is independent of the dictionary size.

Computer-aided medical image analysis has been widely used in clinics to facilitate objective disease diagnosis. This facilitation, however, is often qualitative instead of quantitative due to the analysis challenges associated with medical images such as low signal-to-noise ratio, signal dropout, and large variations. Consequently, physicians have to rely on their personal experiences to make diagnostic decisions, which in turn is expertise-dependent and prone to individual bias. Recently, low-rank modeling based approaches have achieved great success in natural image analysis. There is a trend that low-rank modeling will find its applications in medical image analysis. In this review paper, we like to review the recent progresses along this direction. Concretely, we will first explain the mathematical background of low-rank modeling, categorize existing low-rank modeling approaches and their applications in natural image analysis. After that, we will illustrate some application examples of using low-rank modeling in medical image analysis. Finally, we will discuss some possibilities of developing more robust analysis methods to better analyze cardiac images.

We have embedded Adaptive Compressive Sensing (ACS) algorithm on Charge-Coupled-Device (CCD) camera based on the simplest concept that each pixel is a charge bucket, and the charges comes from Einstein photoelectric conversion effect. Applying the manufactory design principle, we only allow altering each working component at a minimum one step. We then simulated what would be such a camera can do for real world persistent surveillance taking into account of diurnal, all weather, and seasonal variations. The data storage has saved immensely, and the order of magnitude of saving is inversely proportional to target angular speed. We did design two new components of CCD camera. Due to the matured CMOS (Complementary metal–oxide–semiconductor) technology, the on-chip Sample and Hold (SAH) circuitry can be designed for a dual Photon Detector (PD) analog circuitry for changedetection that predicts skipping or going forward at a sufficient sampling frame rate. For an admitted frame, there is a purely random sparse matrix [Φ] which is implemented at each bucket pixel level the charge transport bias voltage toward its neighborhood buckets or not, and if not, it goes to the ground drainage. Since the snapshot image is not a video, we could not apply the usual MPEG video compression and Hoffman entropy codec as well as powerful WaveNet Wrapper on sensor level. We shall compare (i) Pre-Processing FFT and a threshold of significant Fourier mode components and inverse FFT to check PSNR; (ii) Post-Processing image recovery will be selectively done by CDT&D adaptive version of linear programming at L1 minimization and L2 similarity. For (ii) we need to determine in new frames selection by SAH circuitry (i) the degree of information (d.o.i) K(t) dictates the purely random linear sparse combination of measurement data a la [Φ]_M,N M(t) = K(t) Log N(t).

This paper presents novel graph algorithms and modern control solutions applied to the graph networks resulting from specific experiments to discover disease-related pathways and drug targets in glioma cancer stem cells (GSCs). The theoretical framework applies to many other high-throughput data from experiments relevant to a variety of diseases. In addition to developing novel graph and control networks to predict therapeutic targets, these algorithms will provide biochemists with techniques to identify more metabolic regions and biological pathways for complex diseases, and design and test novel therapeutic solutions.

The majority of elderly with Alzheimer’s Disease (AD) receive care at home from caregivers. In contrast to standard tethered clinical settings, a wireless, real-time, body-area smartphone-based remote monitoring of electroencephalogram (EEG) can be extremely advantageous for home care of those patients. Such wearable tools pave the way to personalized medicine, for example giving the opportunity to control the progression of the disease and the effect of drugs. By applying Compressive Sensing (CS) techniques it is in principle possible to overcome the difficulty raised by smartphones spatial-temporal throughput rate bottleneck. Unfortunately, EEG and other physiological signals are often non-sparse. In this paper, it is instead shown that the EEG of AD patients becomes actually more compressible with the progression of the disease. EEG of Mild Cognitive Impaired (MCI) subjects is also showing clear tendency to enhanced compressibility. This feature favor the use of CS techniques and ultimately the use of telemonitoring with wearable sensors.

There is a general chain of events that applies to infections. Human body infection could causes by many different types of bacteria and virus in different areas or organ systems. In general, doctor can’t find out the right solution/treatment for infections unless some certain types of bacteria or virus are detected. These detecting processes, usually, take few days to one week to accomplish. However, some infections of the body may not be able to detect at first round and the patient may lose the timing to receive the proper treatment. In this works, we base on Chi’s theory which is an invisible circulation system existed inside the body and propose a novel health sensor which summarizes human’s infrared, acoustic energy and magnetic signature and find out, in minutes, the most possible area or organ system that cause the infection just like what Chi-Kung master can accomplish. Therefore, the detection process by doctor will be shortened and it raises the possibility to give the proper treatment to the patient in the earliest timing.

Modeling and Simulation (M&S) has been evolving along two general directions: (i) data-rich approach suffering the curse of dimensionality and (ii) equation-rich approach suffering computing power and turnaround time. We suggest a third approach. We call it (iii) compressive M&S (CM&S); because the basic Minimum Free-Helmholtz Energy (MFE) facilitating CM&S can reproduce and generalize Candes, Romberg, Tao & Donoho (CRT&D) Compressive Sensing (CS) paradigm as a linear Lagrange Constraint Neural network (LCNN) algorithm. CM&S based MFE can generalize LCNN to 2^nd order as Nonlinear augmented LCNN. For example, during the sunset, we can avoid a reddish bias of sunlight illumination due to a long-range Rayleigh scattering over the horizon. With CM&S we can take instead of day camera, a night vision camera. We decomposed long wave infrared (LWIR) band with filter into 2 vector components (8~10μm and 10~12μm) and used LCNN to find pixel by pixel the map of Emissive-Equivalent Planck Radiation Sources (EPRS). Then, we up-shifted consistently, according to de-mixed sources map, to the sub-micron RGB color image. Moreover, the night vision imaging can also be down-shifted at Passive Millimeter Wave (PMMW) imaging, suffering less blur owing to dusty smokes scattering and enjoying apparent smoothness of surface reflectivity of man-made objects under the Rayleigh resolution. One loses three orders of magnitudes in the spatial Rayleigh resolution; but gains two orders of magnitude in the reflectivity, and gains another two orders in the propagation without obscuring smog . Since CM&S can generate missing data and hard to get dynamic transients, CM&S can reduce unnecessary measurements and their associated cost and computing in the sense of super-saving CS: measuring one & getting one’s neighborhood free .

Over the last several years, much progress has been made for in vitro culture of mouse and human ES cells. Our laboratory focuses on the molecular and cellular mechanisms of neural differentiation from pluripotent cells. Pluripotent cells first become committed to the ectodermal fate and subsequently differentiate into uncommitted neuroectodermal cells. Both previous mammalian and amphibian studies on pluripotent cells have indicated that the neural fate is a sort of the basal direction of the differentiation of these cells while mesoendodermal differentiation requires extrinsic inductive signals. ES cells differentiate into neuroectodermal cells with a rostral-most character (telencephalon and hypothalamus) when they are cultured in the absence of strong patterning signals. In this talk, I first discuss this issue by referring to our recent data on the mechanism of spontaneous neural differentiation in serum-free culture of mouse ES cells. Then, I will talk about self-organization phenomena observed in 3D culture of ES cells, which lead to tissue-autonomous formation of regional structures such as layered cortical tissues. I also discuss our new attempt to monitor these in vitro morphogenetic processes by live imaging, in particular, self-organizing morphogenesis of the optic cup in three-dimensional cultures.

Discovering the variations in human torso shape plays a key role in many design-oriented applications, such as suit designing. With recent advances in 3D surface imaging technologies, people can obtain 3D human torso data that provide more information than traditional measurements. However, how to find different human shapes from 3D torso data is still an open problem. In this paper, we propose to use spectral clustering approach on torso manifold to address this problem. We first represent high-dimensional torso data in a low-dimensional space using manifold learning algorithm. Then the spectral clustering method is performed to get several disjoint clusters. Experimental results show that the clusters discovered by our approach can describe the discrepancies in both genders and human shapes, and our approach achieves better performance than the compared clustering method.

In this paper, we propose a general approach for statistical inference and machine learning based on accumulated observational data. We demonstrate that a large class of machine learning problems can be formulated as the general problem of constructing random intervals with pre-specified coverage probabilities for the parameters of the model for the observational data. We show that the construction of such random intervals can be accomplished by comparing the endpoints of random intervals with confidence sequences for the parameters obtained from the observational data. Asymptotic results are obtained for such sequential methods.

In this paper, we develop an exact computational approach for simultaneous inference of population proportions. The main idea of this computational approach is to use branch and bound technique for rigorous checking of coverage probabilities and the probabilities of making wrong decisions. Applications of the proposed method can be found in machine learning and other areas.

We introduce a new metric on a space of right-sided infinite sequences drawn from a finite alphabet. Emerging from a problem of entropy estimation of a discrete stationary ergodic process, the metric is important on its own part and exhibits some interesting properties. Notably, the number of distinct metric values for a set of sequences of length m is equal to _Fm+3 − 1, where F_m is a Fibonacci number.

Due to the commercial move and gaming industries, Augmented Reality (AR) technology has matured. By definition of AR, both artificial and real humans can be simultaneously present and realistically interact among one another. With the help of physics and physiology, we can build in the AR tool together with real human day-night webcam inputs through a simple interaction of heat transfer –getting hot, action and reaction –walking or falling, as well as the physiology –sweating due to activity. Knowing the person age, weight and 3D coordinates of joints in the body, we deduce the force, the torque, and the energy expenditure during real human movements and apply to an AR human model. We wish to support the physics-physiology AR version, PPAR, as a BMW surveillance tool for senior home alone (SHA). The functionality is to record senior walking and hand movements inside a home environment. Besides the fringe benefit of enabling more visits from grand children through AR video games, the PP-AR surveillance tool may serve as a means to screen patients in the home for potential falls at points around in house. Moreover, we anticipate PP-AR may help analyze the behavior history of SHA, e.g. enhancing the Smartphone SHA Ubiquitous Care Program, by discovering early symptoms of candidate Alzheimer-like midnight excursions, or Parkinson-like trembling motion for when performing challenging muscular joint movements. Using a set of coordinates corresponding to a set of 3D positions representing human joint locations, we compute the Kinetic Energy (KE) generated by each body segment over time. The Work is then calculated, and converted into calories. Using common graphics rendering pipelines, one could invoke AR technology to provide more information about patients to caretakers. Alerts to caretakers can be prompted by a patient’s departure from their personal baseline, and the patient’s time ordered joint information can be loaded to a graphics viewer allowing for high-definition digital reconstruction. Then an entire scene can be viewed from any position in virtual space, and AR can display certain measurements values which either constituted an alert, or otherwise indicate signs of the transition from wellness to illness.