Boonserm Kijsirikul | Chulalongkorn University (original) (raw)

Papers by Boonserm Kijsirikul

IEEE Access, 2023

Glomerulus segmentation in kidney tissue segments is a key process in nephropathology used for th... more Glomerulus segmentation in kidney tissue segments is a key process in nephropathology used for the effective diagnosis of renal diseases. This study proposes a novel and robust application of MLP (Multi-Layer Perceptron) based architectures for the segmentation of glomeruli in PAS (Periodic Acid-Schiff) stained whole renal images for effective diagnosis of renal diseases. For the segmentation challenge, the proposed unique solution uses MLP-UNet (Multi-Layer Perceptron U-Net), a novel design that evades using conventional convolution and self-attention mechanisms. Additionally, the report compares various approaches, including U-Net, and for the first time, trains the TransUNet model on the kidney WSI (Whole Slide Image) dataset. Dice Score and Dice Loss were used as the metric and loss function for training these models. Results showed that MLP-based architectures provide comparable results to pretrained architectures like TransUNet with effectively lesser parameters and no pre-training, and also produce superior Dice scores across the 5-fold cross-validation training and learn more quickly than conventional U-Net architectures.

ABSTRACT Support vector machine (SVM) is a very powerful machine learning algorithm that can be a... more ABSTRACT Support vector machine (SVM) is a very powerful machine learning algorithm that can be applied to many kinds of applications, not only computation sciences but investing tasks also. This paper presents a new algorithm combining SVM with technical analysis for investing in stocks. RReliefF feature selection is used to choose the appropriate training and trading features for SVM. The experimental results show that we can make very appreciating investments from the new investing strategy.

arXiv (Cornell University), Dec 3, 2019

The ever-growing volume of data of user-generated content on social media provides a nearly unlim... more The ever-growing volume of data of user-generated content on social media provides a nearly unlimited corpus of unlabeled data even in languages where resources are scarce. In this paper, we demonstrate that state-of-the-art results on two Thai social text categorization tasks can be realized by pretraining a language model on a large noisy Thai social media corpus of over 1.26 billion tokens and later fine-tuned on the downstream classification tasks. Due to the linguistically noisy and domain-specific nature of the content, our unique data preprocessing steps designed for Thai social media were utilized to ease the training comprehension of the model. We compared four modern language models: ULMFiT, ELMo with biLSTM, OpenAI GPT, and BERT. We systematically compared the models across different dimensions including speed of pretraining and fine-tuning, perplexity, downstream classification benchmarks, and performance in limited pretraining data.

In the medical diagnosis, endoscopic polyps are examined and discrimination between benign and ma... more In the medical diagnosis, endoscopic polyps are examined and discrimination between benign and malignant is judged via the visual observation by a medical doctor. Estimating size of the polyp is important because the larger polyp tends to become malignant if the size is more than 1 cm since it is not easy to estimate the size of the polyp except medical doctors who have several years of experience. This paper proposes reliable correspondence between two images with a slight movment of endoscope using ORB features estimating the absolute size of polyp using blood vessel information in the endoscopic images. The effectiveness of the proposed approach is confirmed via experiments.

Fine-grained visual classification (FGVC) is a task belonging to multiple sub-categories classifi... more Fine-grained visual classification (FGVC) is a task belonging to multiple sub-categories classification. It is a challenging task due to high intraclass variation and inter-class similarity. Most exiting methods pay attention to capture discriminative semantic parts to address those problems. In this paper, we introduce a two-level network which consists of raw-level and object-level networks, and we name it “Efficient Image Embedding”. Its training procedure has two stages which the raw-level is for localization by the aggregation of feature maps, and the last is for classification. The two-level use Adaptive Angular Margin loss (AAM-loss), which improve an intra-class compactness and inter-class variety of image embedding. Our approach is to identify object regions without any hand-crafted bounding-box, and can be trained in an end-to-end manner. It has achieved better accuracy on two datasets compared to the existing work, which are 89.0% for CUB200-2011 and 93.3% for FGVC-Aircraft.

For languages that have no explicit word boundary such as Thai, Chinese and Japanese, correcting ... more For languages that have no explicit word boundary such as Thai, Chinese and Japanese, correcting words in text is harder than in English because of additional ambiguities in locating error words. The traditional method handles this by hypothesizing that every substrings in the input sentence could be error words and trying to correct all of them. In this paper, we propose the idea of reducing the scope of spelling correction by focusing only on dubious areas in the input sentence. Boundaries of these dubious areas could be obtained approximately by applying word segmentation algorithm and finding word sequences with low probability. To generate the candidate correction words, we used a modified edit distance which reflects the characteristic of Thai OCR errors. Finally, a part-ofspeech trigram model and Winnow algorithm are combined to determine the most probable correction.

The idea of image-to-image translation is to take advantage in certain areas such as adding the s... more The idea of image-to-image translation is to take advantage in certain areas such as adding the sharpness to images and improving the semantic segmentation. The most popular models for solving problems are generative adversarial network (GAN) [1] models such as DiscoGAN [2] and CycleGAN [3]. In training process, input images with no desired properties, and output images with the desired properties are fed into the generative model to train the model. After training, the model can synthesize the desired properties from the input images without those properties. However, in practical usage, an input image may be different from the training process because the input image may be the image with or without the desired properties. This research proposes the method of training the generative model by giving input images with and without desired properties in the same way as when the model is used. Our proposed model enhances DiscoGAN with repeated property construction to generate images with desired properties. The model can use unpaired data as the training data, which makes data preparation more efficiently and more comprehensive than paired data. The proposed model obtained approximately 8% better Fréchet Inception Distance (FID) [4] score compared to the DiscoGAN model.

A task-oriented dialogue bot helps users achieve a predefined goal within a closed domain. A neur... more A task-oriented dialogue bot helps users achieve a predefined goal within a closed domain. A neural-network based dialogue bot tracks the user intention in each action, which can reach promising performance compared to a hand-crafted baseline [1] and has a more flexible conversational flow. One such end-to-end architecture is the Hybrid Code Networks (HCNs) [2]. It uses the simulated conversation of human-bot in the domain of restaurant booking to train an LSTM to track dialogue states and predict the next bot response. This research proposes a similar architecture to HCNs with the addition of attention to LSTM [3]. The best results are obtained by our model on both original and Thai translated versions of bAbI task 5.

An endoscope is a medical instrument that acquires images inside the human body. In this paper, a... more An endoscope is a medical instrument that acquires images inside the human body. In this paper, a new 3-D reconstruction approach is proposed to estimate the size and shape of the polyp under conditions of both point light source illumination and perspective projection. Previous approaches could not know the size of polyp without assuming reflectance parameters as known constant. Even if it was possible to estimate the absolute size of polyp, it was assumed that the parameter of camera movement ∆Z is treated as a known along the depth direction. Here two images are used with a medical suture which is known size object to solve this problem and the proposed approach shows the parameter of camera movement can be estimated with robust accuracy with correspondence between two images taken via slight movement of Z. Experiments with endoscope images are demonstrated to evaluate the validity of proposed approach. Recent research (Tatematsu et al., 2013) (Iwahori et al., 2015a) proposes an approach to recover 3-D

International journal of software innovation, Apr 1, 2017

This paper proposes a new approach to recover the polyp shape from an endoscope image using a pho... more This paper proposes a new approach to recover the polyp shape from an endoscope image using a photometric constraint equation considering two light sources. The procedures are as follows. First, obtain the initial depth distributions by optimizing photometric equation obtained from two light sources. Next, obtain the surface normal vector from depth using numerical difference at each point. Then the mapping between the obtained normal vector and true normal vector is learned using Radial Basis Function Neural Network for a Lambertian sphere, and learning is generalized to another actual polyp image. Finally, optimize the depth using the obtained surface normal to recover the final 3D shape. The validity is confirmed of this method in comparison with the previous methods via computer simulation and experiments using actual endoscope images.

iScience, Mar 1, 2022

Summary Deep learning models have become increasingly used for image-based classification. In cri... more Summary Deep learning models have become increasingly used for image-based classification. In critical applications such as medical imaging, it is important to convey the reasoning behind the models' decisions in human-understandable forms. In this work, we propose Pyramid Localization Network (PYLON), a deep learning model that delivers precise location explanation by increasing the resolution of heatmaps produced by class activation map (CAM). PYLON substantially improves the quality of CAM’s heatmaps in both general image and medical image domains and excels at pinpointing the locations of small objects. Most importantly, PYLON does not require expert annotation of the object location but instead can be trained using only image-level label. This capability is especially important for domain where expert annotation is often unavailable or costly to obtain. We also demonstrate an effective transfer learning approach for applying PYLON on small datasets and summarize technical guidelines that would facilitate wider adoption of the technique.

Procedia Computer Science, 2018

This paper proposes a new defect classification method of electronic board using Dense SIFT and C... more This paper proposes a new defect classification method of electronic board using Dense SIFT and CNN which can represent the effective features to the gray scale image. Proposed method does not use any reference image and effective keypoints are detected using Dense SIFT on the defect candidate region. Removing the feature points except defect region and Bag of Features are used to represent the histogram features. Dense SIFT and SVM are used to judge defect or not. CNN is further introduced to classify true or pseudo defect. Classification accuracy was evaluated and effectiveness of the proposed method is shown.

Sensors

Network security is paramount in today’s digital landscape, where cyberthreats continue to evolve... more Network security is paramount in today’s digital landscape, where cyberthreats continue to evolve and pose significant risks. We propose a DPDK-based scanner based on a study on advanced port scanning techniques to improve network visibility and security. The traditional port scanning methods suffer from speed, accuracy, and efficiency limitations, hindering effective threat detection and mitigation. In this paper, we develop and implement advanced techniques such as protocol-specific probes and evasive scan techniques to enhance the visibility and security of networks. We also evaluate network scanning performance and scalability using programmable hardware, including smart NICs and DPDK-based frameworks, along with in-network processing, data parallelization, and hardware acceleration. Additionally, we leverage application-level protocol parsing to accelerate network discovery and mapping, analyzing protocol-specific information. In our experimental evaluation, our proposed DPDK-b...

Electronics

The well-being of residents is a top priority for megacities, which is why urban design and susta... more The well-being of residents is a top priority for megacities, which is why urban design and sustainable development are crucial topics. Quality of Life (QoL) is used as an effective key performance index (KPI) to measure the efficiency of a city plan’s quantity and quality factors. For city dwellers, QoL for pedestrians is also significant. The walkability concept evaluates and analyzes the QoL in a walking scene. However, the traditional questionnaire survey approach is costly, time-consuming, and limited in its evaluation area. To overcome these limitations, the paper proposes using artificial intelligence (AI) technology to evaluate walkability data collected through a questionnaire survey using virtual reality (VR) tools. The proposed method involves knowledge extraction using deep convolutional neural networks (DCNNs) for information extraction and deep learning (DL) models to infer QoL scores. Knowledge distillation (KD) is also applied to reduce the model size and improve rea...

Engineering Journal, May 31, 2022

We propose a new technique for support vector machines (SVMs) in tree structures for multiclass c... more We propose a new technique for support vector machines (SVMs) in tree structures for multiclass classification. For each tree node, we select an appropriate binary classifier using data class centroids and their in-between distances, categorize the training examples into positive and negative groups of classes and train a new classifier. The proposed technique is fast-trained and can classify an output class data with a complexity between O(log 2 N) and O(N) where N is the number of classes. The 10-fold cross-validation experimental results show that the performance of our methods is comparable to that of traditional techniques and required less decision times. Our proposed technique is suitable for problems with a large number of classes due to its advantages of requiring less training time and computational complexity.

Sustainability

Understanding the quality of life related to transportation plays a crucial role in enhancing com... more Understanding the quality of life related to transportation plays a crucial role in enhancing commuters’ quality of life, particularly in daily trips. This study explores the spatial effects of built environment on quality of life related to transportation (QoLT) through the combination of GIS application and deep learning based on a questionnaire survey by focusing on a case study in Sukhumvit district, Bangkok, Thailand. The Geographic Information System (GIS) was applied for spatial analysis and visualization among all variables through a grid cell (500 × 500 sq.m.). In regard to deep learning, the semantic segmentation process that the model used in this research was OCRNet, and the selected backbone was HRNet_W48. A quality-of-life-related transportation indicator (life satisfaction) was implemented through 500 face-to-face interviews and the data were collected by a questionnaire survey. Then, multinomial regression analysis was performed to demonstrate the significant in posi...

Journal of Advances in Information Technology

Crude oil plays a vital role in the global economy and forecasting crude oil prices is crucial fo... more Crude oil plays a vital role in the global economy and forecasting crude oil prices is crucial for both government and private sectors. However, the crude oil price is high volatility, influenced by various factors and challenging to predict. Thus, various machine learning techniques have been proposed to predict crude oil prices for decades. In this study, we propose an Artificial Neural Network (ANN) with different combinations of Convolutional Neural Networks (CNN) and Long Short-Term Memory (LSTM) to improve the trend forecasting of crude oil prices for better trading signals compared to traditional strategies. As the crude oil price is a time series data, it is appropriate to apply CNN and LSTM for forecasting. The concept of our model is that CNN could detect features or patterns in different locations of time series data, while LSTM could maintain both short-term and long-term memory along with time series data. The collaboration of their abilities could help the neural network model understand complex relationships of historical data and trends of crude oil prices. Our study found that the combination of CNN and LSTM could significantly enhance trading performance in the long run. Index Terms-crude oil trading, machine learning, deep learning, trading signal, technical analysis, artificial intelligent I. INTRODUCTION Crude oil is a commodity that significantly impacts the world economy because 30% of the overall energy supply in the world uses crude oil as the source of energy [1] Products of refined crude oil are used in various economic activities such as power generation, raw material for the petrochemical industry, and transportation by vehicles, ships, and airplanes. Therefore, crude oil price possesses direct influences on many industries that are related to these economic activities. In addition, the crude oil price is a vital factor in the global economy for inflation forecast, monetary policy, and fiscal policy by the government sector [2]. However, the crude oil price is very volatile. It depends on the dynamic condition of demand and supply, including the growth of economic activities, technology, alternative energy like accepted May 26, 2022. natural gas, coal, renewable energy, black-swan event like the coronavirus disease of 2019 (COVID-19).

Proceedings of the 59th Annual Meeting of the Association for Computational Linguistics and the 11th International Joint Conference on Natural Language Processing (Volume 1: Long Papers)

Lifelong learning (LL) aims to train a neural network on a stream of tasks while retaining knowle... more Lifelong learning (LL) aims to train a neural network on a stream of tasks while retaining knowledge from previous tasks. However, many prior attempts in NLP still suffer from the catastrophic forgetting issue, where the model completely forgets what it just learned in the previous tasks. In this paper, we introduce Rational LAMOL, a novel end-to-end LL framework for language models. In order to alleviate catastrophic forgetting, Rational LAMOL enhances LAMOL, a recent LL model, by applying critical freezing guided by human rationales. When the human rationales are not available, we propose exploiting unsupervised generated rationales as substitutions. In the experiment, we tested Rational LAMOL on permutations of three datasets from the ERASER benchmark. The results show that our proposed framework outperformed vanilla LAMOL on most permutations. Furthermore, unsupervised rationale generation was able to consistently improve the overall LL performance from the baseline without relying on human-annotated rationales. We made our code publicly available at https://github. com/kanwatchara-k/r_lamol.

2019 8th International Congress on Advanced Applied Informatics (IIAI-AAI)

Cytology directly examining cells in the early detection of cancer plays an important role in the... more Cytology directly examining cells in the early detection of cancer plays an important role in the medical diagnosis, but this diagnosis depends on the experience and technology of a pathologist. Problem is that it costs time for the examination and its objectivity is poor in the less experienced pathologist. Although there are some previous researches to detect nuclei, this paper proposes the automatic classification of benign or malignant of cell nuclei based on the characteristics that nucleolus has the features of malignant cell nuclei and appears frequently in the malignant cell with cancer. Classification of benign or malignant cell nuclei is performed by detecting nucleoli and counting the number of nucleolus detected.