Efficient computation based on stochastic spikes

The speed and reliability of mammalian perception indicate that cortical computations can rely on very few action potentials per involved neuron. Together with the stochasticity of single-spike events in cortex, this appears to imply that large populations of redundant neurons are needed for rapid computations with action potentials. Here we demonstrate that very fast and precise computations can be realized also in small networks of stochastically spiking neurons. We present a generative network model for which we derive biologically plausible algorithms that perform spike-by-spike updates of the neuron's internal states and adaptation of its synaptic weights from maximizing the likelihood of the observed spike patterns. Paradigmatic computational tasks demonstrate the online performance and learning efficiency of our framework. The potential relevance of our approach as a model for cortical computation is discussed.     

Human eye-head co-ordination in natural exploration

During natural behavior humans continuously adjust their gaze by moving head and eyes, yielding rich dynamics of the retinal input. Sensory coding models, however, typically assume visual input as smooth or a sequence of static images interleaved by volitional gaze shifts. Are these assumptions valid during free exploration behavior in natural environments? We used an innovative technique to simultaneously record gaze and head movements in humans, who freely explored various environments (forest, train station, apartment). Most movements occur along the cardinal axes, and the predominance of vertical or horizontal movements depends on the environment. Eye and head movements co-occur more frequently than their individual statistics predicts under an independence assumption. The majority of co-occurring movements point in opposite directions, consistent with a gaze-stabilizing role of eye movements. Nevertheless, a substantial fraction of eye movements point in the same direction as co-occurring head movements. Even under the very most conservative assumptions, saccadic eye movements alone cannot account for these synergistic movements. Hence nonsaccadic eye movements that interact synergistically with head movements to adjust gaze cannot be neglected in natural visual input. Natural retinal input is continuously dynamic, and cannot be faithfully modeled as a mere sequence of static frames with interleaved large saccades.     

Conjoint analysis to measure the perceived quality in volume rendering

Visualization algorithms can have a large number of parameters, making the space of possible rendering results rather high-dimensional. Only a systematic analysis of the perceived quality can truly reveal the optimal setting for each such parameter. However, an exhaustive search in which all possible parameter permutations are presented to each user within a study group would be infeasible to conduct. Additional complications may result from possible parameter co-dependencies. Here, we will introduce an efficient user study design and analysis strategy that is geared to cope with this problem. The user feedback is fast and easy to obtain and does not require exhaustive parameter testing. To enable such a framework we have modified a preference measuring methodology, conjoint analysis, that originated in psychology and is now also widely used in market research. We demonstrate our framework by a study that measures the perceived quality in volume rendering within the context of large parameter spaces.     

Some Relations between Approximation Problems and PCPs over the Real Numbers

In [10] it was recently shown that that is the existence of transparent long proofs for was established. The latter denotes the class of real number decision problems verifiable in polynomial time as introduced by Blum et al. [6]. The present paper is devoted to the question what impact a potential full real number theorem would have on approximation issues in the BSS model of computation. We study two natural optimization problems in the BSS model. The first, denoted by MAX-QPS, is related to polynomial systems; the other, MAX-q-CAP, deals with algebraic circuits. Our main results combine the PCP framework over with approximation issues for these two problems. We also give a negative approximation result for a variant of the MAX-QPS problem.     

Discovering the events expert practitioners extract from dynamic data streams: the modified unit marking technique

One of the cornerstones of expert performance in complex domains is the ability to perceive problem situations in terms of their task-relevant semantic properties. One such class of properties consists of phenomena that are defined in terms of patterns of change over time, i.e., events. A basic pre-requisite for working towards tools to support event recognition is a method for understanding the events that expert practitioners find meaningful in a given field of practice. In this article we present the modified unit marking procedure (mUMP), a technique adapted from work on social perception to facilitate identification of the meaningful phenomena which observers attend to in a dynamic data array. The mUMP and associated data analysis techniques are presented with examples from a first of a kind study where they were used to elicit and understand the events practitioners found meaningful in a scenario from an actual complex work domain.

Proteomic expression profiling of thyroid neoplasms

Thyroid cancer is the most common endocrine neoplasm with multiple histologic subtypes, each associated with different treatments and outcomes. Differentiating benign neoplasms such as follicular adenomas from malignant entities such as follicular carcinomas and papillary carcinoma can be challenging. To define the proteomic profile of different thyroid tumors, we screened an antibody array of 330 features against five thyroid neoplasms: follicular adenoma, follicular carcinoma, papillary carcinoma, anaplastic carcinoma, and medullary carcinoma as well as normal thyroid epithelium. Eight candidate biomarkers; c-erbB-2, Stat5a, Annexin IV, IL-11, RARα, FGF7, Caspase 9, and phospho-c-myc were identified as differentially expressed on the antibody array, and validated with immunohistochemistry on tissue microarrays, with a total of 144 samples of the same variety of thyroid neoplasms. Analysis revealed c-erbB-2, Annexin IV, and Stat5a have potential clinical utility to differentiate follicular adenoma, follicular carcinoma, and papillary carcinoma from each other. By using an antibody array as a discovery platform and a tissue microarray as a first step in validation on a large number of specimens, we have identified new markers that have potential utility in the diagnosis of thyroid neoplasms.     

Fluid dynamics of a quantized vortex filament in a hole

The behavior of nucleated vortex loops and of remanent vortex filaments in idealized circular and slit-like orifices has been investigated by direct computation. It is found that such vortices can be stretched by the diverging flow on the exit side of the orifice. The energy needed to stretch the vortex is abstracted from the flow field and observed as dissipation. This occurs in the form of discrete phase-slip events in the case of nucleated vortex loops, and in the form of multiple-phase-slip cascades when a remanent vortex is involved.     

Value and limits of μ-CT for nondemineralized bone tissue processing

An experimental approach was performed on 20 giant rabbits to establish the possibilities and limitations of μ-CT for routine processing of nondemineralized bone tissue. Hydroxyapatite (HA) or β-tricalciumphosphate (β-TCP) bead implants or a melange of both, microchambered and solid, were implanted into a standardized and precise defect in the patellar groove. The bone-healing phase was chosen for the histology considering 1 or 2 days, and 2, 3, and 6 weeks. Normal X-ray and μ-CT were applied on all specimens; five specimens in the 6-week stage were additionally processed according to the full range of conventional nondemineralized bone processing methods. μ-CT increased the possibilities of nondemineralized histology with respect to bone morphometry and a complete sequence of sections, thus providing a complete analysis of the bone response. μ-CT was limited in differentiating bone quality, cell analyses, and mineralization stages. The investigation based on normal X-rays is limited to defining integration and excluding the fibrous and bony encapsulation of loose implants. μ-CT allows a 3D evaluation of newly formed bone which is clearly marked against the ceramic implant. It does not allow, however, for the differentiation between woven and lamellar bone, the presentation of the canalicular lacunar system, or on the cell level, revealing canaliculi or details of the mineralization process which can be documented by high-resolution microradiography. Titer dynamics of bone formation remains the domain of polychromatic sequential labeling. The complete sequence of μ-CT slices enhances the possibilities for routine histology, tremendously allowing to the focus on detail histology to topographically well-defined cuts, thus providing more precise conclusions which take into consideration the whole implant.     

Generating feature spaces for linear algorithms with regularized sparse kernel slow feature analysis

Without non-linear basis functions many problems can not be solved by linear algorithms. This article proposes a method to automatically construct such basis functions with slow feature analysis (SFA). Non-linear optimization of this unsupervised learning method generates an orthogonal basis on the unknown latent space for a given time series. In contrast to methods like PCA, SFA is thus well suited for techniques that make direct use of the latent space. Real-world time series can be complex, and current SFA algorithms are either not powerful enough or tend to over-fit. We make use of the kernel trick in combination with sparsification to develop a kernelized SFA algorithm which provides a powerful function class for large data sets. Sparsity is achieved by a novel matching pursuit approach that can be applied to other tasks as well. For small data sets, however, the kernel SFA approach leads to over-fitting and numerical instabilities. To enforce a stable solution, we introduce regularization to the SFA objective. We hypothesize that our algorithm generates a feature space that resembles a Fourier basis in the unknown space of latent variables underlying a given real-world time series. We evaluate this hypothesis at the example of a vowel classification task in comparison to sparse kernel PCA. Our results show excellent classification accuracy and demonstrate the superiority of kernel SFA over kernel PCA in encoding latent variables.