Neuroclusterization method of stabilometric data

Stabilometry is a modern method for assessing the functional state of a person by the ability to maintain a stable balance of an upright posture. Technically, the implementation of the stabilometry method consists in measuring, with the help of specialized devices, the values that make up the support reaction, with the subsequent determination, according to these measurements, of the coordinates of the center of body pressure on the support. The nature of the migrations of the center of pressure during the stabilometric study is a source of information about the features of the processes of postural regulation. At the same time, up to the present time, there is a problem of the correct interpretation of the results of stabilometry. The adequacy of the conclusions is largely determined by the human factor, i.e. qualification of a specialist analyzing stabilometry data. Thus, in our opinion, the task of objectifying the assessment of stabilometry results is urgent. The aim of this work is to study the possibility of applying the neurocluster method using self-organizing neural networks to objectify the analysis of stabilometry data. The authors proposed a technique for analyzing the structure of individual and group stabilometric data by clustering them using selforganizing Kohonen neural maps with Euclidean metrics. Neuroclusterization of stabilometric data allows in automatic mode (without human intervention) to identify the type of group of subjects corresponding to the norm or pathology, various types of pathologies, as well as individual biometric characteristics of the subjects. The subsequent analysis of the individual characteristics of the data of the subjects, grouped in this way, makes it possible to detect deviations indicating the presence of abnormalities or the formation of various pathological conditions, which can be useful for the early diagnosis of diseases.

SELF ORGANIZING NEURAL MAPS IN THE PROBLEMS OF ECOLOGICAL MONITORING

There is a certain problem in ecological monitoring of the environment state according to the measured values of a certain abiotic factor. Namely, how to build a continuous map of environmental pollution throughout the controlled area, based on the results of measurements carried out at a finite number of points inside the controlled territory. The aim of the work is to study the possibility of using the method of self organizing neural maps (SOM) for the problems of the ecological monitoring of the environment, and specifically for building an accurate continuous map of environmental pollution on the ground. The materials and methods of researches are the results of measurements the ambient equivalent of the continuous X-ray and gamma radiation dose rate on a territory of the historical center of Kharkiv has been used as research materials; processing of the obtained data by SOM's methods using MatLab 8.1 and STATISTICA 10 computer programs has been done. Results: in the process of 1000 self-learning cycles of a neural network of 100 initial active neurons randomly located on the controlled area map, 25 neural clusters have been obtained, the coordinates of the centers of which practically coincided with the 25 control points coordinates. A continuous map of the background radiation on the controlled area has been built. The accuracy of this map was no worse than 0.25 μR/hour. Conclusions: the possibility of using the SOM methods to build a continuous map of the level of environmental pollution on the ground based on the results of measuring the values of a certain abiotic factor in a finite number of points has been proven. It has been proven that this method is more accurate compared to the methods of regression mapping and cluster analysis, from which it is essentially different. The possibilities for a significant improvement in the accuracy of the method lie in increasing the number of initial neurons on the terrain map and the number of iterations during their training.

Sketch-Driven Regular Expression Generation from Natural Language and Examples

Recent systems for converting natural language descriptions into regular expressions (regexes) have achieved some success, but typically deal with short, formulaic text and can only produce simple regexes. Real-world regexes are complex, hard to describe with brief sentences, and sometimes require examples to fully convey the user’s intent. We present a framework for regex synthesis in this setting where both natural language (NL) and examples are available. First, a semantic parser (either grammar-based or neural) maps the natural language description into an intermediate sketch, which is an incomplete regex containing holes to denote missing components. Then a program synthesizer searches over the regex space defined by the sketch and finds a regex that is consistent with the given string examples. Our semantic parser can be trained purely from weak supervision based on correctness of the synthesized regex, or it can leverage heuristically derived sketches. We evaluate on two prior datasets (Kushman and Barzilay 2013 ; Locascio et al. 2016 ) and a real-world dataset from Stack Overflow. Our system achieves state-of-the-art performance on the prior datasets and solves 57% of the real-world dataset, which existing neural systems completely fail on. 1

New insights on the modeling of the molecular mechanisms underlying neural maps alignment in the midbrain

We previously identified and modeled a principle of visual map alignment in the midbrain involving the mapping of the retinal projections and concurrent transposition of retinal guidance cues into the superior colliculus providing positional information for the organization of cortical V1 projections onto the retinal map (Savier et al., 2017). This principle relies on mechanisms involving Epha/Efna signaling, correlated neuronal activity and axon competition. Here, using the 3-step map alignment computational model, we predict and validate in vivo the visual mapping defects in a well-characterized mouse model. Our results challenge previous hypotheses and provide an alternative, although complementary, explanation for the phenotype observed. In addition, we propose a new quantification method to assess the degree of alignment and organization between maps, allowing inter-model comparisons. This work generalizes the validity and robustness of the 3-step map alignment algorithm as a predictive tool and confirms the basic mechanisms of visual map organization.

New insights on the modelling of the molecular mechanisms underlying neural maps alignment in the midbrain

We previously identified and modelled a principle of visual maps alignment in the midbrain involving the mapping of the retinal projections and concurrent transposition of retinal guidance cues into the superior colliculus providing positional information for the organization of cortical V1 projections onto the retinal map (Savier et al., 2017). This principle relies on mechanisms involving Epha/Efna signaling, correlated neuronal activity and axon competition. Here, using the 3-step map alignment computational model, we predict and validate in vivo the visual mapping defects in a well-characterized mouse model. Our results challenge previous hypotheses and provide an alternative, although complementary, explanation for the phenotype observed. In addition, we propose a new quantification method to assess the degree of alignment and organization between maps, allowing inter-model comparisons. This work generalizes the validity and robustness of the 3-step map alignment algorithm as a predictive tool and confirms the basic mechanisms of visual maps organization.