Young-Sup Hwang

Predicting protein-protein interactions from one feature using SVM

One of the difficulties of using Artificial Neural Networks (ANNs) to estimate atmospheric temperature is the large number of potential input variables available. In this study, four different feature extraction methods were used to reduce the input vector to train four networks to estimate temperature at different atmospheric levels. The four techniques used were: genetic algorithms (GA), coefficient of determination (CoD), mutual information (MI) and simple neural analysis (SNA). The results demonstrate that of the four methods used for this data set, mutual information and simple neural analysis can generate networks that have a smaller input parameter set, while still maintaining a high degree of accuracy.

Fast multi-feature pedestrian detection algorithm based on histogram of oriented gradient using discrete wavelet transform

A convergence between a natural user interface (NUI) and advanced driver assistance system is considered as a next generation technology. This kind of interfacing system technology becomes more popular in driver assistance system of automobile. Especially, pedestrian detection is an important cue for intelligent vehicles and interactive driver assistance system. In this paper, we propose a pedestrian detection feature and technique by combining histogram of the oriented gradient (HOG) and discrete wavelet transform (DWT). In the method, the magnitude of motion is used to set region of interest (ROI) for improving detection speed. Then, we employ multi-feature for a pedestrian detection based on the HOG and DWT. In last stage, to classify whether a candidate window contains a pedestrian or not, the designed multi-feature is learned by using the training data with the support vector machine (SVM) mechanism. Experimental results show that the proposed algorithm increases the speed-up factor of 27.21 % by comparing to the existing method using the original HOG feature.

An effective and intelligent Windows application filtering system using software similarity

As licensed programs are pirated and illegally spread over the Internet, it is necessary to filter illegally distributed or cracked programs. The conventional software filtering systems can prevent unauthorized dissemination of the programs maintained by their databases using an exact matching method where the feature of a suspicious program is the same as that of any program stored in the database. However, the conventional filtering systems have some limitations to deal with cracked or new programs which are not maintained by their database. To address the limitations, we design and implement an efficient and intelligent software filtering system based on software similarity. Our system measures the similarity of the characteristics extracted from an original program and a suspicious one (or, a cracked one) and then determines whether the suspicious program is a cracked version of the copyrighted original program based on the similarity measure. In addition, the proposed system can handle a new program by categorizing it using a machine learning scheme. This scheme helps an unknown program to be identified by narrowing the search space. To demonstrate the effectiveness of the proposed system, we perform a series of experiments on a number of executable programs under Microsoft Windows. The experimental results show that our system has achieved comparable performance.

Pairwise matching of spots in 2-DE images using hopfield network

Matching spots between two-dimensional electrophoresis (2-DE) images is a bottleneck in the automation of proteome analysis. Because the matching problem is an NP-hard problem, the solution is usually a heuristic approach or a neural network method. So a Hopfield neural network approach is applied to solve this problem. An energy function is designed to represent the similarity of spots together with its neighbor spots. Experiment showed that Hopfield neural network with appropriate energy function and dynamics could solve the matching problem of spots in 2-DE images.

Context-aware block-based motion estimation algorithm for multimedia internet of things (IoT) platform

Shaping video data into fast-responding transmission and high resolution output video using cost-effective video processing is desirable in many applications including Internet of Things (IoT) applications. In association with rapid development of IoT smart sensor applications, real-time processing of huge-amount of data for a video signal has become necessary leading to video compression technology. Motion estimation (ME) is necessary for improving the quality, but it has high computational complexity in video compression system. The present article, therefore, proposes a context-aware adaptive pattern-based ME algorithm for multimedia IoT platform to improve video compression. In the proposed algorithm, the motions are classified into large or small based on distortion value. Accordingly, the search pattern is chosen either small diamond search pattern (SDSP) or large diamond search pattern (LDSP) in each and every step of ME; allowing adaptive processing of large and small abstract information. Compared to conventional fast algorithms, the experimental results demonstrate up to 40 and 36% reduction in encoding time for low-delay main (LB-main) and random access main (RA-main) profiles respectively in HEVC test model 16.10 encoder with bit-rate loss of 0.071 and 0.246% for both the profiles, ensuring quality video and searching precision.

Android malware detection using convolutional neural networks and data section images

The paper proposes a new technique to detect Android malware effectively based on converting malware binaries into images and applying machine learning techniques on those images. Existing research converts the whole executable files (e.g., DEX files in Android application package) of target apps into images and uses them for machine learning. However, the entire DEX file (consisting of header section, identifier section, data section, optional link data area, etc.) might contain noisy information for malware detection. In this paper, we convert only data sections of DEX files into grayscale images and apply machine learning on the images with Convolutional Neural Networks (CNN). By using only the data sections for 5,377 malicious and 6,249 benign apps, our technique reduces the storage capacity by 17.5% on average compared to using the whole DEX files. We apply two CNN models, Inception-v3 and Inception-ResNet-v2, which are known to be efficient in image processing, and examine the effectiveness of our technique in terms of accuracy. Experiment results show that the proposed technique achieves better accuracy with smaller storage capacity than the approach using the whole DEX files. Inception-ResNet-v2 with the stochastic gradient descent (SGD) optimization algorithm reaches 98.02% accuracy.

Computing paper similarity based on latent dirichlet allocation

This paper discusses methods to compute paper similarity accurately using Latent Dirichlet Allocation (LDA). The problems occurring when we compute paper similarity based on LDA are as follows. At first, paper similarity in a paper database is hard to be calculated accurately because they are too deficient in text information, which is caused by the copyright problem and the technical limits of crawling and parsing. Secondly, it is hard to provide the inputs necessary to compute similarity based on LDA. To compute LDA-based similarity, a user should input the topic number and determine seed papers as many as the topic number. This paper proposes the following methods to solve these two problems. To solve the deficiency of text, we apply the keyword extension method to compute LDA-based similarity. The keyword extension method uses the text referred by the compared paper or text in papers referring the compared paper as text information. To select appropriate seed papers, we propose a method to utilize reference information of the paper compared. Finally, we demonstrate the superiority of the proposed method by experimenting on real papers.

A semantic similarity measure in document databases: an earth mover's distance-based approach

Measuring document similarity is important in order to find documents which are similar to a given query document from a user. Text-based document similarity is measured by comparing the words in two documents. The representative text-based document similarity is the cosine similarity. Since the cosine similarity computes document similarity by estimating the frequency of common words, it cannot reflect word similarity. To solve this problem, we propose a new document similarity measure based on the earth movers distance (EMD). The EMD is one of the most popular distance functions used to search similar multimedia contents and is known to provide good search results. To apply the EMD to compute document similarity, we have to solve two problems: (1) The EMD is too time consuming to be used in a document database, (2) the distance between words should be defined. Our proposed approach first extracts topics as new features of a document by applying the latent Dirichlet allocation, which is a generative model of a document. It can decrease the computational cost of the EMD because the number of topics is much smaller than the number of words in a document. After extracting the topics, the proposed approach calculates the distance between topics based on the relation between the topics and the words in a document database, thereby making computing document similarity based on the EMD possible. Our approach searches documents more accurately since we can consider the semantic similarity by using the EMD. Experimental results on a real-world document database indicate that the proposed approach outperforms the cosine similarity in terms of the accuracy and the performance.