• The Transactions of the Korean Institute of Electrical Engineers A
• /
• v.48 no.11
• /
• pp.1464-1471
• /
• 1999

This paper introduces complex-valued competitive learning neural network for spatio-temporal pattern recognition. There have been quite a few neural networks for spatio-temporal pattern recognition. Among them, recurrent neural network, TDNN, and avalanche model are acknowledged as standard neural network paradigms for spatio-temporal pattern recognition. Recurrent neural network has complicated learning rules and does not guarantee convergence to global minima. TDNN requires too many neurons, and can not be regarded to deal with spatio-temporal pattern basically. Grossberg's avalanche model is not able to distinguish long patterns, and has to be indicated which layer is to be used in learning. In order to remedy drawbacks of the above networks, unsupervised competitive learning using complex umber is proposed. Suggested neural network also features simultaneous recognition, time-shift invariant recognition, stable categorizing, and learning rate modulation. The network is evaluated by computer simulation with randomly generated patterns.

• Journal of the Korean Institute of Intelligent Systems
• /
• v.19 no.1
• /
• pp.54-61
• /
• 2009

This paper presents a new electromyogram (EMG) pattern recognition method based on the Hierarchical Temporal Memory (HTM) algorithm which is originally devised for image pattern recognition. In the modified HTM algorithm, a simplified two-level structure with spatial pooler, temporal pooler, and supervised mapper is proposed for efficient learning and classification of the EMG signals. To enhance the recognition performance, the category information is utilized not only in the supervised mapper but also in the temporal pooler. The experimental results show that the ten kinds of hand motion are successfully recognized.

• Proceedings of the Korea Water Resources Association Conference
• /
• 2021.06a
• /
• pp.97-97
• /
• 2021

Hydrometeorological variables contain highly complex system for temporal revolution and it is quite challenging to illustrate the system with a temporal linear and nonlinear models. In recent years, deep learning algorithms have been developed and a number of studies has focused to model the complex hydrometeorological system with deep learning models. In the current study, we investigated the temporal structure inside deep learning models for the hydrometeorological variables such as streamflow and climate indices. The results present a quite striking such that each hidden unit of the deep learning model presents different dependence structure and when the number of hidden units meet a proper boundary, it reaches the best model performance. This indicates that the deep dependence structure of deep learning models can be used to model selection or investigating whether the constructed model setup present efficient or not.

• PNF and Movement
• /
• v.18 no.2
• /
• pp.287-296
• /
• 2020

Purpose: The study aims were to develop a wearable inertial sensor-based gait analysis device that uses machine learning algorithms, and to validate this novel device using temporal gait parameters. Methods: Thirty-four healthy young participants (22 male, 12 female, aged 25.76 years) with no musculoskeletal disorders were asked to walk at three different speeds. As they walked, data were simultaneously collected by a motion capture system and inertial measurement units (Reseed®). The data were sent to a machine learning algorithm adapted to the wearable inertial sensor-based gait analysis device. The validity of the newly developed instrument was assessed by comparing it to data from the motion capture system. Results: At normal speeds, intra-class correlation coefficients (ICC) for the temporal gait parameters were excellent (ICC [2, 1], 0.99~0.99), and coefficient of variation (CV) error values were insignificant for all gait parameters (0.31~1.08%). At slow speeds, ICCs for the temporal gait parameters were excellent (ICC [2, 1], 0.98~0.99), and CV error values were very small for all gait parameters (0.33~1.24%). At the fastest speeds, ICCs for temporal gait parameters were excellent (ICC [2, 1], 0.86~0.99) but less impressive than for the other speeds. CV error values were small for all gait parameters (0.17~5.58%). Conclusion: These results confirm that both the wearable inertial sensor-based gait analysis device and the machine learning algorithms have strong concurrent validity for temporal variables. On that basis, this novel wearable device is likely to prove useful for establishing temporal gait parameters while assessing gait.

• The Journal of Korean Institute of Communications and Information Sciences
• /
• v.21 no.2
• /
• pp.534-540
• /
• 1996

This paper presents an Adaptive Time-Delay Recurrent Neural Network (ATRN) for learning and recognition of temporal correlations of temporal patterns. The ATRN employs adaptive time-delays and recurrent connections, which are inspired from neurobiology. In the ATRN, the adaptive time-delays make the ATRN choose the optimal values of time-delays for the temporal location of the important information in the input parrerns, and the recurrent connections enable the network to encode and integrate temporal information of sequences which have arbitrary interval time and arbitrary length of temporal context. The ATRN described in this paper, ATNN proposed by Lin, and TDNN introduced by Waibel were simulated and applied to the chaotic time series preditcion of Mackey-Glass delay-differential equation. The simulation results show that the normalized mean square error (NMSE) of ATRN is 0.0026, while the NMSE values of ATNN and TDNN are 0.014, 0.0117, respectively, and in temporal learning, employing recurrent links in the network is more effective than putting multiple time-delays into the neurons. The best performance is attained bythe ATRN. This ATRN will be sell applicable for temporally continuous domains, such as speech recognition, moving object recognition, motor control, and time-series prediction.

• Journal of Auto-vehicle Safety Association
• /
• v.8 no.1
• /
• pp.19-25
• /
• 2016

As new car technology is developing, temporal interaction is needed in automotive. Rhythmic pattern is one of the practical examples of temporal interaction in vehicle. To recognize rhythmic pattern and its input medium, dictionary learning is applicable algorithm. In this paper, performance and memory requirement of the learning algorithm is tested and is sufficiently good for use this acoustic sound.

• Clinical and Experimental Pediatrics
• /
• v.49 no.4
• /
• pp.341-353
• /
• 2006

Learning disorders are diagnosed when the individual's achievement on standardized tests in reading, mathematics, or written expression is substantially below that expected for age, schooling, and level of intelligence. Subtypes of learning disorders may be classified into two groups, language-based type learning disorders including reading and writing disorder, and nonverbal type learning disorder (NLD) such as those relating to mathematics & visuospatial skills, and those in the autism spectrum. Converging evidence indicates that reading disorder represents a disorder within the language system and more specifically within a particular subcomponent of that system, phonological processing. Recent advances in neuroimaging technology, particularly the development of fMRI, provide evidences of a neurobiological basis for reading disorder, specifically a disruption of two left hemisphere posterior brain systems, one parieto-temporal, the other occipito-temporal. The former is the reading system for beginner reading, the latter for skilled reading. Compensatory engagement of anterior systems around the inferior frontal gyrus(Broca's area) and a posterior(right occipito-temporal) system is noted in persistent poor readers in long-term follow up study. The theoretical model proposed to explain NLD's source is not right hemisphere damage, but rather the white matter model. The working hypothesis of the white matter model is that the underdevelopment of, damage to, or dysfunction of cerebral white matter(long myelinated fibers) is the source of this disorder. The role of an evidence-based effective intervention in the remediation of children with learning disorder is discussed.

• Korean Journal of Remote Sensing
• /
• v.37 no.4
• /
• pp.719-731
• /
• 2021

This study proposes a two-stage hybrid classification model for crop classification using multi-temporal remote sensing images; the model combines feature embedding by using an autoencoder (AE) with a convolutional neural network (CNN) classifier to fully utilize features including informative temporal and spatial signatures. Long short-term memory (LSTM)-based AE (LAE) is fine-tuned using class label information to extract latent features that contain less noise and useful temporal signatures. The CNN classifier is then applied to effectively account for the spatial characteristics of the extracted latent features. A crop classification experiment with multi-temporal unmanned aerial vehicle images is conducted to illustrate the potential application of the proposed hybrid model. The classification performance of the proposed model is compared with various combinations of conventional deep learning models (CNN, LSTM, and convolutional LSTM) and different inputs (original multi-temporal images and features from stacked AE). From the crop classification experiment, the best classification accuracy was achieved by the proposed model that utilized the latent features by fine-tuned LAE as input for the CNN classifier. The latent features that contain useful temporal signatures and are less noisy could increase the class separability between crops with similar spectral signatures, thereby leading to superior classification accuracy. The experimental results demonstrate the importance of effective feature extraction and the potential of the proposed classification model for crop classification using multi-temporal remote sensing images.

• Journal of the Korean Society of Physical Medicine
• /
• v.5 no.3
• /
• pp.395-404
• /
• 2010

Purpose : The object of this study was to examine the effect of motor learning on brain activation depending on the method of motor learning. Methods : The brain activation was measured in 9 men by fMRI. The subjects were divided into the following groups depending on the method of motor learning: actually practice (AP, n=3) group, action observation (AO, n=3) group and motor imagery (MI, n=3) group. In order to examine the effect of motor learning depending on the method of motor learning, the brain activation data were measured during learning. For the investigation of brain activation, fMRI was conducted. Results : The results of brain activation measured before and during learning were as follows; (1) During learning, the AP group showed the activation in the following areas: primary motor area located in precentral gyrus, somatosensory area located in postcentral gyrus, supplemental motor area and prefrontal association area located in precentral gyrus, middle frontal gyrus and superior frontal gyrus, speech area located in superior temporal gyrus and middle temporal gyrus, Broca's area located in inferior parietal lobe and somatosensory association area of precuneus; (2) During learning, the AD groups showed the activation in the following areas: primary motor area located in precentral gyrus, prefrontal association area located in middle frontal gyrus and superior frontal gyrus, speech area and supplemental motor area located in superior temporal gyrus and middle temporal gyrus, Broca's area located in inferior parietal lobe, somatosensory area and primary motor area located in precentral gyrus of right cerebrum and left cerebrum, and somatosensory association area located in precuneus; and (3) During learning, the MI group showed activation in the following areas: speech area located in superior temporal gyrus, supplemental area, and somatosensory association area located in precuneus. Conclusion : Given the results above, in this study, the action observation was suggested as an alternative to motor learning through actual practice in serial reaction time task of motor learning. It showed the similar results to the actual practice in brain activation which were obtained using activation of mirror neuron. This result suggests that the brain activation occurred by the activation of mirror neuron, which was observed during action observation. The mirror neurons are located in primary motor area, somatosensory area, premotor area, supplemental motor area and somatosensory association area. In sum, when we plan a training program through physiotherapy to increase the effect during reeducation of movement, the action observation as well as best resting is necessary in increasing the effect of motor learning with the patients who cannot be engaged in actual practice.

• KSII Transactions on Internet and Information Systems (TIIS)
• /
• v.12 no.7
• /
• pp.3307-3321
• /
• 2018

With the ever-increasing queries related to temporal (or time-related) information, such as the product launching time, in search engine, most web pages will be augmented with such information in the future. Meanwhile, the gradual emergence of the use of electronic textbooks (or e-Textbooks), which enrich the traditional paper-based textbooks with multimedia contents such as interactive quizzes and multimedia-based simulations, has led us to infer that e-Textbooks will be blended with temporal information to support learning. The use of temporal information helps teachers and students to understand the level of prior knowledge required to study a topic, as well as the sequence of learning activities and related sub-topics, that best attains the educational goals. This paper presents a simple yet efficient tool called TimeMap, which is based on mind mapping, to create an e-Textbook called TimeBook that takes account of time-related curriculum and the ability of students to learn via collaboration.