Recurrent Self-organizing Map Research Articles

Application of temporal self-organizing maps to patterning short-time series of fish behavior responding to environmental stress

Response behaviors of animals to stressors are complicated. Especially, it is difficult to analyze temporal data in addressing the time course development of behavioral states to stressors. Not much study has been conducted in identifying the short-term behavior conditions of indicator organisms responding to stressors. Hence, in the present study, we selected three temporal networks, Temporal Kohonen Map (TKM), Recurrent Self-Organizing Map (RSOM), and Recursive Self-Organizing Map (RecSOM) to characterize the behavior strength of individual fish exposed to Atrazine (ATZ, 0.12 mg/L) for 15 days. The trained networks were able to detect the characteristic behavior strength segments presenting the Stepwise Behavior Response Model (SBRM). TKM was found suitable for detecting strong curves in a short training time, but clustering was limited in grouped segments. In contrast, RecSOM was found to be most suitable for detecting behavior segments with both high and low contrast response under toxic exposure. RecSOM was effective in clustering but could not detect segments with strong changes. Finally, our results presented that the temporal SOMs, especially RecSOMs, were feasible in identifying short-term time series data and could be utilized for on-line monitoring of test organisms under stressful conditions.

Analyzing the Response Behavior of Lumbriculus variegatus (Oligochaeta: Lumbriculidae) to Different Concentrations of Copper Sulfate Based on Line Body Shape Detection and a Recurrent Self-Organizing Map.

Point detection (e.g., the centroid of the body) of species has been conducted in numerous studies. However, line detection (i.e., the line body shape) of elongated species has rarely been investigated under stressful conditions. We analyzed the line movements of an Oligochaeta Lumbriculus variegatus in response to treatments with a toxic chemical, copper sulfate, at low concentrations (0.01 mg/L and 0.1 mg/L). The automatic line-tracking system was devised to identify the movement of body segments (body length) and the movements of segments (i.e., the speed and angles between segments) were recorded before and after treatment. Total body length was shortened from 31.22 (±5.18) mm to 20.91 (±4.65) mm after the 0.1 mg/L treatment. The Shannon entropy index decreased from 0.44 (±0.1) to 0.28 (±0.08) after treatment. On the other hand, the body and movement segments did not significantly change after the 0.01 mg/L treatment. Sequential movements of test organisms were further analyzed with a recurrent self-organizing map (RSOM) to determine the pattern of time-series line movements. The RSOM made it feasible to classify sequential behaviors of indicator organisms and identify various continuous body movements under stressful conditions.

Software Aging Forecast Using Recurrent SOM with Local Model

Studies of software aging problems are important since they are related to QoS. Previous studies have used many methods to guarantee QoS. In this article, a recurrent self-organizing map with multi-layerperceptron is proposed to forecast resource consumption in a web server which suffered from a software aging problem. First, a resource consumption series in a web server is split into p dimensional space vectors. Second, the split series is clustered into local sets by using a recurrent self-organizing map. Last, a local prediction method called multi-layerperceptron is used to predict on each local set. The results indicated that the recurrent self-organizing map with multi-layerperceptron generates a slightly better estimation than multi-layerperceptron and autoregressive integrated moving average in the resource consumption predictions of system and application level of web server.

Human activity prediction by mapping grouplets to recurrent Self-Organizing Map

Human activity prediction is defined as inferring the high-level activity category with the observation of only a few action units. It is very meaningful for time-critical applications such as emergency surveillance. For efficient prediction, we represent the ongoing human activity by using body part movements and taking full advantage of inherent sequentiality, then find the best matching activity template by a proper aligning measurement.In streaming videos, dense spatio-temporal interest points (STIPs) are first extracted as low-level descriptors for their high detection efficiency. Then, sparse grouplets, i.e., clustered point groups, are located to represent body part movements, for which we propose a scale-adaptive mean shift method that can determine grouplet number and scale for each frame adaptively. To learn the sequentiality, located grouplets are successively mapped to Recurrent Self-Organizing Map (RSOM), which has been pre-trained to preserve the temporal topology of grouplet sequences. During this mapping, a growing RSOM trajectory, which represents the ongoing activity, is obtained. For the special structure of RSOM trajectory, a combination of dynamic time warping (DTW) distance and edit distance, called DTW-E distance, is designed for similarity measurement. Four activity datasets with different characteristics such as complex scenes and inter-class ambiguities serve for performance evaluation. Experimental results confirm that our method is very efficient for predicting human activity and yields better performance than state-of-the-art works.

Cooperative Growing Hierarchical Recurrent Self Organizing Model for Phoneme Recognition

In this paper, we propose a system of a tree evolutionary recurrent self-organizing models. Inherited from the Growing Hierarchical Self-Organizing Map GHSOM. The proposed GHSOM variants are characterized by a hierarchical model, composed of independent RSOMs (recurrent Self-Organizing Map). The case study of the proposed system is phoneme recognition in continuous speech and speaker independent context. GHSOM variants serve az tools for developing intel- ligent systems and proposed artificial intelligence application.

Integrating recurrent SOM with wavelet-based kernel partial least square regressions for financial forecasting

This study implements a novel expert system for financial forecasting. In the first stage, wavelet analysis transforms the input space of raw data to a time-scale feature space suitable for financial forecasting, and then a Recurrent Self-Organizing Map (RSOM) algorithm is used for partitioning and storing temporal context of the feature space. In the second stage, multiple kernel partial least square regressors (as local models) that best fit partitioned regions are constructed for final forecasting. Compared with neural networks, pure SVMs or traditional GARCH models, the proposed model performs best. The root-mean-squared forecasting errors are significantly reduced.

Extracting finite structure from infinite language

This paper presents a novel connectionist memory-rule based model capable of learning the finite-state properties of an input language from a set of positive examples. The model is based upon an unsupervised recurrent self-organizing map with laterally interconnected neurons. A derivation of functional-equivalence theory is used that allows the model to exploit similarities between the future context of previously memorized sequences and the future context of the current input sequence. This bottom-up learning algorithm binds functionally related neurons together to form states. Results show that the model is able to learn the Reber grammar perfectly from a randomly generated training set and to generalize to sequences beyond the length of those found in the training set. ed sequences and the future context of the current input sequence. This bottom-up learning algorithm binds functionally related neurons together to form states. Results show that the model is able to learn the Reber grammar [A. Cleeremans, D. Schreiber, J. McClelland, Finite state automata and simple recurrent networks, Neural Computation, 1 (1989) 372-381] perfectly from a randomly generated training set and to generalize to sequences beyond the length of those found in the training set.

Self-organizing Symbol Acquisition and Motion Generation based on Dynamics-based Information Processing System

The symbol acquisition and manipulation abilities are one of the inherent characteristics of human beings comparing with other creatures. In this paper, based on recurrent self-organizing map and dynamics-based information processing system, we propose a dynamics based self-organizing map (DBSOM). This method enables designing a topological map using time sequence data, which causes recognition and generation of the robot motion. Using this method, we design the self-organizing symbol acquisition system and robot motion generation system for a humanoid robot. By implementing DBSOM to the robot in the real world, we realize the symbol acquisition from the experimental data and investigate the spatial property of the obtained DBSOM.

Nonlinear principal component analysis by learning nerve fields united by inhibitory connections

AbstractThe authors propose a model with feedforward and a recurrent self‐organizing maps model which can extract nonlinear principal components. This model consists of several learning nerve fields that are mutually inhibiting each other, and is an extension of the self‐organizing overlapping maps model proposed previously by the authors. The feedforward model is sufficient for nonlinear principal component analysis, but the recurrent model with symmetrical inhibitory connections between nerve fields has additional interesting characteristics. © 2004 Wiley Periodicals, Inc. Syst Comp Jpn, 35(3): 68–78, 2004; Published online in Wiley InterScience (www.interscience.wiley.com). DOI 10.1002/scj.1230

Period detection and representation by recurrent oscillatory self-organizing map

Whether or not periodicity is a property of the environment, for a cognitive system a period is always a dynamical mental construct. This study suggests a dynamical implementation of this hypothesis using recurrent oscillatory self-organizing map of the feature space of such streams. The mapping allows the system to assign identity and class-membership to each stream point, represented by a locus on a two-dimensional map. The model is shown to detect periodicities of various regular and behavior-originated, single- and multi-channel wave patterns, and to reproduce such signals, relying solely on the oscillating activation of the units.

Temporal Kohonen Map and the Recurrent Self-Organizing Map: Analytical and Experimental Comparison

This paper compares two Self-Organizing Map (SOM) based models for temporal sequence processing (TSP) both analytically and experimentally. These models, Temporal Kohonen Map (TKM) and Recurrent Self-Organizing Map (RSOM), incorporate leaky integrator memory to preserve the temporal context of the input signals. The learning and the convergence properties of the TKM and RSOM are studied and we show analytically that the RSOM is a significant improvement over the TKM, because the RSOM allows simple derivation of a consistent learning rule. The results of the analysis are demonstrated with experiments.

Bringing Knowing-When and Knowing-What Together: Periodically Tuned Categorization and Category-Based Timing Modeled with the Recurrent Oscillatory Self-Organizing Map (ROSOM)

The study addresses the cyclically temporal aspect of sequence recognition, storage and recall using the Recurrent Oscillatory Self-Organizing Map (ROSOM), first introduced by Kaipainen, Papadopoulos and Karhu (1997). The unique solution of the network is that oscillatory States are assigned to network units, corresponding to their `readiness-to-fire'. The ROSOM is a categorizer, a temporal sequence storage system and a periodicity detector designed for use in an ambiguous cyclically repetitive environment. As its external input, the model accepts a multidimensional stream of environment-describing feature configurations with implicit periodicities. The output of the model is one or a few closed cycles abstracted from such a stream, mapped as trajectories on a two-dimensional sheet with an organization reminiscent of multi-dimensional scaling. The model's capabilities are explored with a variety of workbench data.