Chawuthai, Rathachai

Preferred name

Alternative Name

Chawuthai, R.

Main Affiliation

School of Engineering

rathachai.ch@kmitl.ac.th

Now showing 1 - 10 of 24

Some of the metrics are blocked by your
consent settings
Assessing the Effects of Corrupted Parameters in a Large Language Model: A Case Study of LLAMA 3.2 1B
(2026-01-21)
Chawuthai, Rathachai
;
Thongsawaeng, Anon
;
Perdio, John Paul Layug
;
Zaw, Kaung Khant
;
Kraichoke, Phalat
This study explores the effects of parameter corruption in a large language model (LLM) by altering its weights and evaluating performance. Experiments involve corrupting different layers and matrix types, including Self-Attention and Feed-Forward components, with performance assessed using BERT and ROUGE scores. Testing the Llama-3.2-1B-Instruct model was performed on the GLUE-QNLI dataset. Results show that increased corruption leads to greater degradation, with Feed-Forward matrices having a stronger impact especially in the Down matrices. According to the study, later layers are more important for performance than those that come before them. These results shed light on possible future chip implementations of LLM, which may help guide the design of fault-tolerant systems by taking vulnerable parameter placement into account.
Some of the metrics are blocked by your
consent settings
Defect detection of GPS trackers through data visualization
(2019-07-01)
Chawuthai, Rathachai
;
Threepak, Thanunchai
A Global Positioning System (GPS) tracker installed in a vehicle is commonly used to improve logistics management processes and transportation safety. All GPS trackers must send data including locations, timestamps, and speeds to a server all the time. In case of a device failure, it can be checked by incomplete data; however, a device's sensor inaccuracy, which can create negative consequences to many parties, becomes a challenging issue to detect. With this reason, this paper aims to adopt data visualization to find out the defect of GPS trackers. It has been found that some defects noticed by a visualization were reported, and providers got advantage of this result to maintain their devices.
Some of the metrics are blocked by your
consent settings
Modelling an RDF Knowledge Graph with Transitivity and Symmetry for Bus Route Path Finding
(2023-12-16)
Chawuthai, Rathachai
;
Kertkeidkachorn, Natthawut
;
Racharak, Teeradaj
A key property of Linked Data is the representation and publication of data as an inter-connected labelled graph where different resources linked to each other form a network of meaningful information. A problem of path finding can be seen as searching important relationships between resources, such as, looking for chains of intermediate nodes. In this paper, we tackle this problem in the context of public transport navigation system, where we aim to find candidates of bus route path given two bus stations. We model a novel lightweight bus network as Resource Description Framework (RDF) triples of directed bus lines and walking paths between connected stations. Indeed, we demonstrate that lightweight bus network can be achieved by exploiting the sub-property of RDF Schema (RDFS) and the transitivity and symmetry provided by Web Ontology Language (OWL). We also perform a scalability test of our approach using a real-world bus network in Bangkok, Thailand. Various patterns of SPARQL Protocol and RDF Query Language (SPARQL) query statements are validated, showing the usefulness of the RDF model. The further step of this paper is to work with bus schedules and travel time analysis in order to select some proper candidates for users through an application.
Some of the metrics are blocked by your
consent settings
Assessing HDBSCAN Implementation for Traffic Congestion Pattern Estimation in Bangkok with Taxi GPS Probe
(2023-01-01)
Tony, Dio
;
Chawuthai, Rathachai
Traffic congestion is a major issue that is experienced globally in metropolitan cities. The phenomenon becomes more serious during peak hour as congestion increases and degrades the traffic networks. Each city possesses a unique traffic network, and the behaviour of its residents affects its traffic patterns. Therefore, a flexible congestion pattern identification approach is desirable. We proposed the employment of Hierarchical Density Based Spatial Clustering of Applications with Noise (HDBSCAN) to estimate traffic congestion propagation patterns through congestion length distribution. Global positioning System (GPS) probe of taxis were utilised to represent traffic pattern within Bangkok. The dataset was preprocessed into two successive timeframes, namely 'later' timeframe and 'prior' timeframe before being clustered. The identified congestion hotspots from both timeframes were transformed into a congestion area from which congestion lengths were extracted. Similarity measurements on congestion lengths distribution were conducted against Longdo Traffic's top 100 most congested roads list in Bangkok, showed encouraging results across all tests, with more than 90% similarity in one of the measurements, which indicated that HDBSCAN was feasible to make a key contribution to traffic management research.
Some of the metrics are blocked by your
consent settings
Eye Landmarks Detection using RT-DETR with Rules
(2024-01-01)
Boonnithititikul, Chatree
;
Jaknamon, Teetouch
;
Chawuthai, Rathachai
In order to help ophthalmologists diagnose eye problems, it is necessary to scan for eye landmarks such as the pupil, the reflection point on the retina, and the boundary of the eye. An individual's eye landmarks on their face can be obtained via some facial landmarks' detection methods, including Haar Cascade. Two problematic aspects of the current approaches, however, are that the pupil and reflection point information is not provided, and the detection is ineffective when confronted with a picture of the upper half of the face or a person wearing a mask. In this study, we intend to develop a deep learning model for eye landmark identification using the Realtime identification Transformer (RT-DETR) approach together with our rules. As a consequence, nine landmark points-two for the eye, six for the pupil, and one for the reflection, are computed with an accuracy of 0.974. Since the focus of this paper is on eye landmark recognition, the next stage will be to build an application and a machine learning model for the diagnosis of eye disorders. - Keywords Deep Learning, Detection, Eye Landmarks, Facial Landmarks, Ophthalmology, RT-DETR
Some of the metrics are blocked by your
consent settings
A Classification Model for Road Traffic Incidents on Twitter Data
(2022-01-01)
Raksachat, Thawatchai
;
Chawuthai, Rathachai
This study aims to create a classification model for road traffic incidents in Thailand using Twitter data. The challenging issue of our work is to deal with highly imbalanced dataset of 5 classes. As we surveyed, some pieces of research solved this issue by the Markov Chains method. However, using the Markov Chains in our dataset provides low performance, so we study the Undersampling, Oversampling, Markov Chains, and Bi-directional Long Short-Term Memory (Bi-LSTM). As we use the Markov Chains as the baseline, the result of our experiment found that using Bi-LSTM provides the improvement of F1-score up to 15.44% against the baseline.
Some of the metrics are blocked by your
consent settings
U-GMo: Individual Clip Detection from a Graduation Ceremony Video
(2024-01-01)
Treesoonrat, Natee
;
Kriengchaiyaprug, Nunnapat
;
Upadhayawong, Thanakann
;
Lohapongpan, Warinya
;
Chawuthai, Rathachai
Graduation ceremonies are important occasions in life. A video in this event is usually very long due to a lot of graduates getting their degree. This study suggests a method for automatically cutting the entire ceremony video into customized segments that only include the most significant events for each particular graduate, named U-GMo (Your Great Moment). The system uses deep learning with computer vision techniques, such as YOLOv8 for posture detection, to identify graduates by observing their motions and posture during the degree ceremony. After that, the identified bits are taken out and assembled into brief video snippets for every graduate. The algorithm can detect and extract each graduate's crucial moments with high performance, according to an examination conducted on a dataset of graduation ceremonies. The personalized video clips provide a convenient way to preserve the meaningful highlights from these milestone events.
Some of the metrics are blocked by your
consent settings
A Recommender System for Trip Planners
(2020-01-01)
Chawuthai, Rathachai
;
Omarak, Prodpran
;
Thaiyingsombat, Vitchaya
Making a trip plan is a key activity for having a satisfying trip for tourists. However, as we survey, there are many users do not need to spend a lot of time to write the plan, and it becomes a pain point for users. The users prefer to have a simple way to create a whole trip plan from a few users’ constraints, and the users just customize some items for their satisfaction. Thus, this work introduces a recommender system that mainly employs the genetic algorithm for generating a trip plan. The approach accepts a few roughly input requirements from users, and then it creates a whole trip schedule and allows users to modify. To have a quality trip plan, any places and times in the plan have to correspond to places’ categories, open weekdays, times to spend, favorite daytimes and months, and possible routes. A web application for trip planner is developed to demonstrate the suitability and feasibility of the proposed recommender system. After that, the user feedback and usage statistic present the high degree of user satisfaction and opportunity to improve tourism of any city.
Some of the metrics are blocked by your
consent settings
Software Requirement Readiness Levels (SrRL)
(2024-01-01)
Chawuthai, Rathachai
This paper introduces software requirement readiness levels (SrRL), which is a definition to determine the readiness level of each software requirement of a software system. Our proposed SrRL provides the definitions of 9 levels that are applied from the concept of technology readiness level (TRL), together with scenarios and recommendations to achieve each level. The contribution of this work expects that the SrRL can be a guideline for software development teams and criteria for accepting software systems.
Some of the metrics are blocked by your
consent settings
Route Prediction from GPS Trajectory and Road Data
(2023-01-01)
Chawuthai, Rathachai
;
Kawachakul, Kampanart
;
Boonrod, Kittikom
;
Threepak, Thanunchai
This paper presents an approach to create a route prediction model for multiple vehicles from GPS trajectory and road data. Since the baseline model is designed for a single car and it provides low performance for our experiment, our approach using the HDBSCAN clustering for route data preprocessing and the prediction model based on Viterbi algorithm, which is an extension of the Hidden Markov Model, provides the better performance in terms of Hit@K where K being 3. The result of our work demonstrates the feasibility to improve the smart city technology under the scope of smart mobility as well. (Abstract)

Chawuthai, Rathachai

Filters

Settings

Sort By

Results per page