Deep Learning Architectures and Applications (딥러닝의 모형과 응용사례)
-
- Journal of Intelligence and Information Systems
- /
- v.22 no.2
- /
- pp.127-142
- /
- 2016
Deep learning model is a kind of neural networks that allows multiple hidden layers. There are various deep learning architectures such as convolutional neural networks, deep belief networks and recurrent neural networks. Those have been applied to fields like computer vision, automatic speech recognition, natural language processing, audio recognition and bioinformatics where they have been shown to produce state-of-the-art results on various tasks. Among those architectures, convolutional neural networks and recurrent neural networks are classified as the supervised learning model. And in recent years, those supervised learning models have gained more popularity than unsupervised learning models such as deep belief networks, because supervised learning models have shown fashionable applications in such fields mentioned above. Deep learning models can be trained with backpropagation algorithm. Backpropagation is an abbreviation for "backward propagation of errors" and a common method of training artificial neural networks used in conjunction with an optimization method such as gradient descent. The method calculates the gradient of an error function with respect to all the weights in the network. The gradient is fed to the optimization method which in turn uses it to update the weights, in an attempt to minimize the error function. Convolutional neural networks use a special architecture which is particularly well-adapted to classify images. Using this architecture makes convolutional networks fast to train. This, in turn, helps us train deep, muti-layer networks, which are very good at classifying images. These days, deep convolutional networks are used in most neural networks for image recognition. Convolutional neural networks use three basic ideas: local receptive fields, shared weights, and pooling. By local receptive fields, we mean that each neuron in the first(or any) hidden layer will be connected to a small region of the input(or previous layer's) neurons. Shared weights mean that we're going to use the same weights and bias for each of the local receptive field. This means that all the neurons in the hidden layer detect exactly the same feature, just at different locations in the input image. In addition to the convolutional layers just described, convolutional neural networks also contain pooling layers. Pooling layers are usually used immediately after convolutional layers. What the pooling layers do is to simplify the information in the output from the convolutional layer. Recent convolutional network architectures have 10 to 20 hidden layers and billions of connections between units. Training deep learning networks has taken weeks several years ago, but thanks to progress in GPU and algorithm enhancement, training time has reduced to several hours. Neural networks with time-varying behavior are known as recurrent neural networks or RNNs. A recurrent neural network is a class of artificial neural network where connections between units form a directed cycle. This creates an internal state of the network which allows it to exhibit dynamic temporal behavior. Unlike feedforward neural networks, RNNs can use their internal memory to process arbitrary sequences of inputs. Early RNN models turned out to be very difficult to train, harder even than deep feedforward networks. The reason is the unstable gradient problem such as vanishing gradient and exploding gradient. The gradient can get smaller and smaller as it is propagated back through layers. This makes learning in early layers extremely slow. The problem actually gets worse in RNNs, since gradients aren't just propagated backward through layers, they're propagated backward through time. If the network runs for a long time, that can make the gradient extremely unstable and hard to learn from. It has been possible to incorporate an idea known as long short-term memory units (LSTMs) into RNNs. LSTMs make it much easier to get good results when training RNNs, and many recent papers make use of LSTMs or related ideas.
Objectives: This study aimed to compare the utility of the Psychoeducational Profile-Revised (PEP-R), Korean Wechsler Preschool and Primary Scale of Intelligence, Fourth Edition (K-WPPSI-IV), and Vineland Adaptive Behavior Scale, Second Edition (VABS-II) for evaluating developmental disabilities (DD) in preschool children. Additionally, we examined the correlations between the PEP-R, K-WPPSI-IV, and VABS-II. Methods: A total of 164 children aged 37-84 months were assessed. Children's development was evaluated using the PEP-R, K-WPPSI-IV, VABS-II, Preschool Receptive-Expressive Language Scale, and Korean Childhood Autism Rating Scale, Second Edition. Results: Of the 164 children, 103 had typical development (TD) and 61 had DD. The mean of the PEP-R Developmental Quotient (DQ), K-WPPSI-IV Full-Scale Intelligence Quotient (FSIQ), and VABS-II Adaptive Behavior Composite (ABC) scores were significantly higher in the TD group than in the DD group (p<0.001). The estimated area under the curve of the PEP-R DQ, K-WPPSI-IV FSIQ, and VABS-II ABC scores was 0.953 (95% confidence interval [CI]=0.915-0.992), 0.955 (95% CI=0.914-0.996), and 0.961 (95% CI=0.932-0.991), respectively, which did not indicate a statistically significant difference. The PEP-R DQ scores were positively correlated with the K-WPPSI-IV FSIQ (r=0.90, p<0.001) and VABS-II ABC scores (r=0.84, p<0.001). A strong correlation was observed between the K-WPPSI-IV FSIQ and VABS-II ABC scores (r=0.89, p<0.001). Conclusion: This study found that the PEP-R, K-WPPSI-IV, and VABS-II effectively distinguished DD from TD in preschool children, and no significant differences in utility were observed between them.
Objectives : Since the awareness of autism spectrum disorders(ASD) is growing, as a result, it is increasing numbers of infants and toddlers being referred to specialized clinics for a differential diagnosis and the importance of early autism spectrum disorders detection is emphasized. This study is to know the difference between ASD and intellectual disability(ID) from comparison of the demographics, clinical characters and obstetric complications. Methods : The participants are 816 toddlers who visited the developmental delay clinic(DDC) in National Health Insurance Ilsan hospital. The number of toddlers diagnosed as ASD and ID was 324 and 492. 75 toddlers out of 114 who returned to DDC were diagnosed as ID at the first visit but 7 of them had changed diagnosis to ASD at the second visit. After compared ASD with ID from the first visit, we analyzed characters of toddlers who had the changed diagnosis to ASD at the second visit. Results : As a result, the comparison between ASD and ID at the first visit shows that the boys have higher ratio, lower obstetric complication and lower language assessment score in ASD. The toddlers who had the changed diagnosis at the second visit were all boys and they had more cases of family history of developmental delay and had lower score of receptive language developmental quotient. Conclusions : These findings suggest that sex, language characteristics and obstetric complication could be useful in the early detection of ASD.
This study investigated the effects of response prompting through 3-steps compliance training to reaction time for child with Asperger's syndrome(AS). The participant was 3 and 8 year-old boy who was diagnostic As with non-compliant, delayed receptive language. Study design was multiple-baseline across behaviors. Target Behaviors were hands-up, following direction, and answering behavior. Dependent variable was latency reaction time during compliance training. This results mean that reaction time was increased raise hands-up behavior, compliance behavior and response ask questions. During intervention, the participant improve the rate on-task behavior as well as reduce off-task behaviors.
Purposes of this study were to explore the process of experience that science teachers go through when participating in peer coaching meetings to improve teaching ability and to find out factors that affect each process of experience. The data were collected through recording of peer coaching meetings, videotapes of science class, and interviews. All the data were analyzed after transcription. The results of the study showed that even though Teacher K broke the ice and formed consensus among the peers by developing Content Representation (CoRe) at the beginning of the meetings, he became self-defensive rather than receptive of peers' opinions on the recorded class at the discussion session. But as the peer coaching went on, he realized that peer coaching was not about evaluation but rather on improving his teaching ability. In turn, he was able to look at his teaching in a more objective point of view and accepted suggestions from peer coaching discussion. The self-reflection of Teacher K acted as the key factor in the efforts to improve his teaching ability. He sought the concrete alternatives through the class analysis with fellow teachers and showed major changes in his teaching practice from the language habits, pronunciation, and speed of his speech to the interaction with students and class design. However, there was little change in knowledge of curriculum and assessment due to his strong orientation to improve students' grades as an academic high school teacher. Likewise, it was found that while peer coaching exert a strong influence on instructional methods and strategies of Teacher K, his strong orientation to improve students' grades hinders a balanced development of subcomponents of PCK.
Objectives:BPI was developed for assessing young children's perceptions. Using an interactive techniques for interviewing children, the BPI blends structured and clinical interviewing technique. Present study was to evaluate the reliability, validity and the clinical utility of Berkeley Puppet Interview(BPI) for young children aged 4 to 7. Methods:Subjects consisted of 37 children(boy20, girl 17) between the age of 4 and 7 who visited the child-psychiatry outpatient division of children's hospital in Seoul and Chungju. The measures used in this study BPI-S(symptomatology) and BPI-PC(parent-child relationship). BPI was translated into Korean by three clinical psychologists. To examine the reliability, Chonbach's alpha were calculated and to examine the validity, correlation coefficients were calculated on BPI-S & K-CBCL. Results:BPI-PC's Cronbach's alpha was 0.86 and BPI-S's Cronbach's alpha was 0.74. Correlation between the internalizing scale of BPI-S and that of K-CBCL was 0.477 and correlation between the internalizing scale of BPI-S and the externalizing scale of K-CBCL was -0.431, suggesting the validity of BPI-S. Conclusion:These results show that BPI-S & BPI-PC may be useful tool for young children's diagnostic interview.
Objective : The purpose of this study was obtaining data on the efficacy and safety of risperidone in child and adolescent psychiatric patients. Method : Thirty one children and adolescents (males n=18, females n=13, age ranged from 5.4 to 17.3 years) treated with risperidone were selected among child and adolescent psychiatric inpatients of Seoul National University Hospital from January, 2001 to June, 2002, and charts for them were reviewed retrospectively. Results : The primary psychiatric disorders treated with risperidone were schizophrenia and other psychosis, bipolar I disorder with psychotic features, Tourette's disorder, autism spectrum disorders, mixed receptive and expressive language disorder, attention deficit-hyperactivity disorder, conduct disorder and obsessive-compulsive disorder. twelve of these had comorbid mental retardation. Primary target symptoms of risperidone were psychotic symptoms (n=13 or
The wall shear stress in the vicinity of end-to end anastomoses under steady flow conditions was measured using a flush-mounted hot-film anemometer(FMHFA) probe. The experimental measurements were in good agreement with numerical results except in flow with low Reynolds numbers. The wall shear stress increased proximal to the anastomosis in flow from the Penrose tubing (simulating an artery) to the PTFE: graft. In flow from the PTFE graft to the Penrose tubing, low wall shear stress was observed distal to the anastomosis. Abnormal distributions of wall shear stress in the vicinity of the anastomosis, resulting from the compliance mismatch between the graft and the host artery, might be an important factor of ANFH formation and the graft failure. The present study suggests a correlation between regions of the low wall shear stress and the development of anastomotic neointimal fibrous hyperplasia(ANPH) in end-to-end anastomoses. 30523 T00401030523 ^x Air pressure decay(APD) rate and ultrafiltration rate(UFR) tests were performed on new and saline rinsed dialyzers as well as those roused in patients several times. C-DAK 4000 (Cordis Dow) and CF IS-11 (Baxter Travenol) reused dialyzers obtained from the dialysis clinic were used in the present study. The new dialyzers exhibited a relatively flat APD, whereas saline rinsed and reused dialyzers showed considerable amount of decay. C-DAH dialyzers had a larger APD(11.70
The wall shear stress in the vicinity of end-to end anastomoses under steady flow conditions was measured using a flush-mounted hot-film anemometer(FMHFA) probe. The experimental measurements were in good agreement with numerical results except in flow with low Reynolds numbers. The wall shear stress increased proximal to the anastomosis in flow from the Penrose tubing (simulating an artery) to the PTFE: graft. In flow from the PTFE graft to the Penrose tubing, low wall shear stress was observed distal to the anastomosis. Abnormal distributions of wall shear stress in the vicinity of the anastomosis, resulting from the compliance mismatch between the graft and the host artery, might be an important factor of ANFH formation and the graft failure. The present study suggests a correlation between regions of the low wall shear stress and the development of anastomotic neointimal fibrous hyperplasia(ANPH) in end-to-end anastomoses. 30523 T00401030523 ^x Air pressure decay(APD) rate and ultrafiltration rate(UFR) tests were performed on new and saline rinsed dialyzers as well as those roused in patients several times. C-DAK 4000 (Cordis Dow) and CF IS-11 (Baxter Travenol) reused dialyzers obtained from the dialysis clinic were used in the present study. The new dialyzers exhibited a relatively flat APD, whereas saline rinsed and reused dialyzers showed considerable amount of decay. C-DAH dialyzers had a larger APD(11.70