Sanming Song

Local Autoencoding for Parameter Estimation in a Hidden Potts-Markov Random Field

A local-autoencoding (LAE) method is proposed for the parameter estimation in a Hidden Potts-Markov random field model. Due to sampling cost, Markov chain Monte Carlo methods are rarely used in real-time applications. Like other heuristic methods, LAE is based on a conditional independence assumption. It adapts, however, the parameters in a block-by-block style with a simple Hebbian learning rule. Experiments with given label fields show that the LAE is able to converge in far less time than required for a scan. It is also possible to derive an estimate for LAE based on a Cramer-Rao bound that is similar to the classical maximum pseudolikelihood method. As a general algorithm, LAE can be used to estimate the parameters in anisotropic label fields. Furthermore, LAE is not limited to the classical Potts model and can be applied to other types of Potts models by simple label field transformations and straightforward learning rule extensions. Experimental results on image segmentations demonstrate the efficiency and generality of the LAE algorithm.

A modular latching chain

Many cognitive tasks involve transitions between distinct mental processes, which may range from discrete states to complex strategies. The ability of cortical networks to combine discrete jumps with continuous glides along ever changing trajectories, dubbed latching dynamics, may be essential for the emergence of the unique cognitive capacities of modern humans. Novel trajectories have to be followed in the multidimensional space of cortical activity for novel behaviours to be produced; yet, not everything changes: several lines of evidence point at recurring patterns in the sequence of activation of cortical areas in a variety of behaviours. To extend a mathematical model of latching dynamics beyond the simple unstructured auto-associative Potts network previously analysed, we introduce delayed structured connectivity and hetero-associative connection weights, and we explore their effects on the dynamics. A modular model in the small-world regime is considered, with modules arranged on a ring. The synaptic weights include a standard auto-associative component, stabilizing distinct patterns of activity, and a hetero-associative component, favoring transitions from one pattern, expressed in one module, to the next, in the next module. We then study, through simulations, how structural parameters, like those regulating rewiring probability, noise and feedback connections, determine sequential association dynamics.

Label field initialization for MRF-based sonar image segmentation by selective autoencoding

The optimal solution of a Markov random field (MRF) can be solved by constructing a Markov chain that eventually goes to a balance state. However, in most situations, only an suboptimal solution can be obtained, because it is hard to choose the ideal initial state and the updating strategy. While the updating strategy has been extensively investigated, the initialization issue has been fully neglected. Though k-means-clustering has been used exclusively in initializing the label field, it suffers from the lack of account of the local constraints, which is the most essential part of the MRF model. A structural method based on selective autoencoding (SAE) is proposed for the label field initialization of MRF model in the task of sonar image segmentation. SAE is similar to the AutoEncoder, with the largest difference on the activation function, where a piece-wise sigmoid activation function with two different slop parameters is used to selectively encode image patches that resemble shadow ares or other areas. The synapse matrixes of SAE network act as information filters, preserve specific area adaptively and selectively, generating a label field that is much closer to the balance state. Experiments on sonar image segmentation demonstrate the efficiency of the SAE algorithm.

Latching chains in K-nearest-neighbor and modular small-world networks

Latching dynamics retrieve pattern sequences successively by neural adaption and pattern correlation. We have previously proposed a modular latching chain model in Song et al. (2014) to better accommodate the structured transitions in the brain. Different cortical areas have different network structures. To explore how structural parameters like rewiring probability, threshold, noise and feedback connections affect the latching dynamics, two different connection schemes, K-nearest-neighbor network and modular network both having modular structure are considered. Latching chains are measured using two proposed measures characterizing length of intra-modular latching chains and sequential inter-modular association transitions. Our main findings include: (1) With decreasing threshold coefficient and rewiring probability, both the K-nearest-neighbor network and the modular network experience quantitatively similar phase change processes. (2) The modular network exhibits selectively enhanced latching in the small-world range of connectivity. (3) The K-nearest-neighbor network is more robust to changes in rewiring probability, while the modular network is more robust to the presence of noise pattern pairs and to changes in the strength of feedback connections. According to our findings, the relationships between latching chains in K-nearest-neighbor and modular networks and different forms of cognition and information processing emerging in the brain are discussed.

Two-dimensional forward-looking sonar image registration by maximization of peripheral mutual information

Monitoring the field of operation of an underwater vehicle is crucial during missions near the sea floor. The forward-looking sonar is often the only available sensor for the observation of the ambient turbid water environment. Sonar image registration is not only a first step towards a panoramic mosaic but it also provides an initial motion parameter estimation for the vehicle self-localization and navigation. In this article, a peripheral mutual information (PMI) maximization method is proposed for the sonar image registration. Peripheral mutual information is inspired by regional mutual information (RMI) which makes use of the closed-form solution for the Shannon entropy by the assumption that the data vectors made of neighbouring pixels are normally distributed, an assumption that ignores correlations between the pixels in sonar images. To accommodate the fact that the neighbouring pixels show dependencies due to acoustic reverberation and dispersion, only the peripheral information in the neighbourhood of a pixel is used in peripheral mutual information for the calculation of the mutual information. Experiments show that the peripheral mutual information registration function is much smoother than that of regional mutual information. Further experiments on the two-dimensional forward-looking sonar image registration demonstrate the efficiency of peripheral mutual information.

Forward-looking sonar image mosaicking by feature tracking

High-frequency forward-looking sonars are appropriate for the operation or search close to the seabed. Constructing a panoramic mosaic not only facilitates an interpretation of the underwater environment, but also supports the vehicles self-localization. In this paper, a method to register sonar sequences is proposed that is based on the feature tracking using the particle filtering. Our methods starts with the extraction of the intensity, the texture and the shape features from the unstructured seabed environment. Next a region of interest (ROI) is tracked until it disappears from the view field of an autonomous unterwater vehicle (AUV). Then, another ROI is selected and the tracking procedures are repeated. Experimental results show that (1) Feature tracking is feasible for the forward-looking sonar image mosaicking. (2) Fusion of the texture and the shape feature lead to a robust feature extraction method for more precise motion estimation. (3) The prior information on the AUVs movement is necessary for the tracking of highlighted regions.

Probe the potts states in the minicolumn dynamics

Minicolumn has been widely accepted not only as the basic structural element of the cortex anatomically, but also as the fundamental functional unit physiologically. And, it is believed by many theorists that the minicolumn may function as a Potts spine, only takes on finite discrete states. But its feasibility is unclear. In order to provide a biophysical evidence for the Potts assumption, a model is proposed to analyze the dynamics of minicolumn. With simulation, we found that Potts states may originate from the temporary high synchronization of neuronal subsets. Furthermore, after analyzing the average number of synchronous spiking neurons, we propose a novel and important assumption that, intrinsically-busting neurons may play a critical role in stabilizing the Potts states.

Modular scale-free function subnetworks in auditory areas

Function connectivity analysis is set to probe the whole-brain network architecture. Only several specific areas have to be focused when a specific modal is considered. To explore the microscopic subnetworks in auditory modality, the mean shift algorithm is proposed to cluster the fMRI time courses in the corresponding activation areas and several heuristic conclusions are obtained. 1) The voxel degree distribution supports scale-free hypothesis, but the exponential is relatively small. 2) More global subnetworks appear in the more abstract cognition process. 3) At least half of the subnetworks are local networks and they seldom cross with each other, acting as independent modules.

Stable fast rewiring depends on the activation of skeleton voxels

Compared with the relatively stable structural networks, the functional networks, defined by the temporal correlation between remote neurophysiological events, are highly complex and variable. However, the transitions should never be random. So it was proposed that some stable fast rewiring mechanisms probably exist in the brain. In order to probe the underlying mechanisms, we analyze the fMRI signal in temporal dimension and obtain several heuristic conclusions. 1) There is a stable time delay, 7~14 seconds, between the stimulus onset and the activation of corresponding functional regions. 2) In analyzing the biophysical factors that support stable fast rewiring, it is, to our best knowledge, the first to observe that skeleton voxels may be essential for the fast rewiring process. 3) Our analysis on the structure of functional network supports the scale-free hypothesis.

Who dominates the retinotectal mapping

Retina ganglia cells (RGCs) connect to the tectum neurons in the midbrain through 1–1 strict topic mapping, which benefits the precise information transmission and is called retinotopic mapping. Its significant to understand the underlying mechanisms that contributing to the topic mapping. Inspired by the experimental observations given by [Nicol et al 2007 Nat. Neurosci. 10(3):340], we divide the retinotectal mapping into three almost independent stages, axon overshooting, axon retraction, and the final refinement of axons and arborization. In order to analyze the roles that forward signaling, reverse signaling and neural activity plays in different stages, an axon retraction model and a refinement procedure have been proposed to simulate the retinotopic mapping process. We testify the asymmetry of gradients between forward signaling and reverse signaling, and find that it is the forward signaling that dominates the axon retraction. Also, the refinement procedure demonstrates that the reverse-signaling-induced gradient branching and the synapse-dependent neural activity lead the RGC axons projecting to the topic target tectum neurons.