KMITL

Permanent URI for this communityhttps://dspace.kmitl.ac.th/handle/123456789/1

Browse

Search Results

Now showing 1 - 10 of 12

Some of the metrics are blocked by your
consent settings
Generative AI for Industrial Applications: Synthetic Dataset
(2023-01-01)
Sasiaowapak, Thanyathep
;
Boonsang, Siridech
;
Chuwongin, Santhad
;
Tongloy, Teerawat
;
Lalitrojwong, Pattarachai
The research and development of artificial intelligence (AI) techniques to enhance quality control in industrial equipment might face challenges due to the scarcity and limited privacy of actual industrial datasets. One approach to address this involves utilizing generative AI models that create synthetic data, simulating the characteristics and diversity found in crucial datasets. We present a methodology for generating synthetic datasets for industrial products such as bolts and screws by employing a segment-anything model and a stable diffusion technique for creating accurate representations. Furthermore, we propose the developed model by using a scaled-down version of DinoV2 algorithm's vision transformer (ViT-small). The self-supervised learning approach was studied to fine-tune the model to classify between normal- and defective industrial products, as well as those contaminated with dirt. By additional training dataset created through synthesis, we achieve an improvement in performance. The synthetic data leads to nearly perfect true positive results while completely eliminating false negatives. This indicates a significant advantage in terms of accuracy, recall, precision, specificity, and F1 score, all of which exceed 98%. Similarly, the model's predictions align perfectly with the area under the curve (AUC) metric. Although there is a slight performance reduction when dealing with up to six different class labels, the model retains strong capability in identifying normal products. Notably, the ViT-Small-based self-supervised learning model demonstrates superior accuracy compared to using ViT-Base, with considerations for dataset compatibility and model suitability. In conclusion, this study's contribution lies in enabling the deployment of the Dino V2 model for implementing quality control measures in industrial domains. It emphasizes the challenges that limited real-industrial data by leveraging synthetic data and innovative fine-tuning approaches, ultimately enhancing AI-powered for quality control processes.
Some of the metrics are blocked by your
consent settings
Intelligent Triage Assistant
(2021-01-01)
Duangdee, Wannarat
;
Lalitrojwong, Pattarachai
The triage of outpatient department is a vital process to screen patients. In addition, assessing the emergency case by the outpatient department can determine the quality of service. In emergency case, the patients have to be treated as soon as they arrive. If the triage of this case is delayed and the patients have major symptoms, it may develop serious complications leading to the cause of death or disability. Accordingly, assessing and making triage decisions, and assigning the level of patient acuity require being accurate, rapid and trustworthy in order to take care and treat them in time. Due to the fact that nowadays there is only one nurse on duty for the triage, he or she cannot well handle a bunch of patients arriving for treatment. According to Hospital and Healthcare Standards, 4th edition, the Healthcare Accreditation Institute (Public Organization) allows the use of technology to improve hospital services. This research aims to develop an intelligent triage assistant. To help the process by making triage decisions for all the patients promptly, assigning an acuity level and screening and assigning the patients to an area based on acuity. If unsuitable results happen, the triage nurse can be on duty promptly. Our intelligent triage assistant has been implemented using Visual Studio C#, Microsoft SQL Server for the database, and QnA Maker for the knowledge base. The application has been tested by outpatient nurses. They are quite satisfied with the primary outcome and recommend further improvements for future work.
Some of the metrics are blocked by your
consent settings
OntoPop: An ontology population system for the semantic web
(2012-01-01)
Thongkrau, Theerayut
;
Lalitrojwong, Pattarachai
The development of ontology at the instance level requires the extraction of the terms defining the instances from various data sources. These instances then are linked to the concepts of the ontology, and relationships are created between these instances for the next step. How ever, before establishing links among data, ontology engineers must classify terms or instances from a web document into an ontology concept. The tool for help ontology engineer in this task is called ontology population. The present research is not suitable for ontology development applications, such as long time processing or analyzing large or noisy data sets. OntoPop system introduces a methodology to solve these problems, which comprises two parts. First, we select meaningful features from syn tactic relations, which can produce more significant features than any other method. Second, we differentiate feature meaning and reduce noise based on latent semantic analysis. Experimental evaluation demonstrates that the OntoPop works well, significantly out-performing the accuracy of 49.64%, a learning accuracy of 76.9%, and executes time of 5.46 second/instance. Copyright © 2012 The Institute of Electronics, Information and Communication Engineers.
Some of the metrics are blocked by your
consent settings
Extracting product features and opinions from product reviews using dependency analysis
(2010-11-29)
Somprasertsri, Gamgarn
;
Lalitrojwong, Pattarachai
In web pages, the reviews are written in natural language and are unstructured-free-texts scheme. Online product reviews is considered as a significant informative resource which is useful for both potential customers and product manufacturers. The task of manually scanning through large amounts of review one by one is computational burden and is not practically implemented with respect to businesses and customer perspectives. Therefore it is more efficient to automatically process the various reviews and provide the necessary information in a suitable form. The task of product feature and opinion is to find product features that customers refer to their topic reviews. It would be useful to characterize the opinions about product. In this paper, we propose an approach to extract product features and to identify the opinions associated with these features from reviews through syntactic information based on dependency analysis. ©2010 IEEE.
Some of the metrics are blocked by your
consent settings
Mining feature-opinion in online customer reviews for opinion summarization
(2010-06-18)
Somprasertsri, Gamgarn
;
Lalitrojwong, Pattarachai
Online customer reviews is considered as a significant informative resource which is useful for both potential customers and product manufacturers. In web pages, the reviews are written in natural language and are unstructured-free-texts scheme. The task of manually scanning through large amounts of review one by one is computational burden and is not practically implemented with respect to businesses and customer perspectives. Therefore it is more efficient to automatically process the various reviews and provide the necessary information in a suitable form. The high-level problem of opinion summarization addresses how to determine the sentiment, attitude or opinion that an author expressed in natural language text with respect to a certain feature. In this paper, we dedicate our work to the main subtask of opinion summarization. The task of product feature and opinion extraction is critical to opinion summarization, because its effectiveness significantly affects the performance of opinion orientation identification. It is important to properly identify the semantic relationships between product features and opinions. We proposed an approach for mining product feature and opinion based on the consideration of syntactic information and semantic information. By applying dependency relations and ontological knowledge with probabilistic based model, the result of our experiments shows that our approach is more flexible and effective. © J.UCS.
Some of the metrics are blocked by your
consent settings
Classifying instances into lexical ontology concepts using latent semantic analysis
(2010-05-28)
Thongkrau, Theerayut
;
Lalitrojwong, Pattarachai
A lexical ontology is useful as the basic knowledge base in artificial intelligence and computational linguistics application. However, it is insufficient to recognize only existing instances for each concept. Adding new instances into the lexical ontology will expand knowledge in the system. In this paper, we propose an efficient unsupervised instance population system that classifies new instances into a corresponding lexical ontology concept. Compared to previous related works, it does not require manual preprocessing to prepare training data. In terms of processing time, it does not need to search for many concepts in the lexical ontology. Furthermore, it is able to handle an unlimited number of ontological concepts in any domain. Our system employs latent semantic analysis together with context voting to find the appropriate concept of the instance. The experiments demonstrate that this approach compared to similarity approach yields higher accuracy for instance classification. In sum, the system achieves higher accuracy when the lexical ontology contains a lot of concepts, which generally occurs in practical problems. ©2010 IEEE.
Some of the metrics are blocked by your
consent settings
Automatic product feature extraction from online product reviews using maximum entropy with lexical and syntactic features
(2008-09-23)
Somprasertsri, Gamgarn
;
Lalitrojwong, Pattarachai
The task of product feature extraction is to find product features that customers refer to their topic reviews. It would be useful to characterize the opinions about the products. We propose an approach for product feature extraction by combining lexical and syntactic features with a maximum entropy model. For the underlying principle of maximum entropy, it prefers the uniform distributions if there is no external knowledge. Using a maximum entropy approach, firstly we extract the learning features from the annotated corpus, secondly we train the maximum entropy model, thirdly we use trained model to extract product features, and finally we apply a natural language processing technique in postprocessing step to discover the remaining product features. Our experimental results show that this approach is suitable for automatic product feature extraction. ©2008 IEEE.
Some of the metrics are blocked by your
consent settings
A maximum entropy model for product feature extraction in online customer reviews
(2008-01-01)
Somprasertsri, Gamgarn
;
Lalitrojwong, Pattarachai
Product feature extraction is an important task of review mining and summarization. The task of product feature extraction is to find product features that customers refer to in their topic reviews. It would be useful to characterize the opinions which they review or express about the products. In this paper, we propose an approach to product feature extraction using a maximum entropy model. Maximum entropy is a probability distribution estimation technique. It is widely used for classification problems in natural language processing, such as question answering, information extraction, and part-of-speech tagging. The underlying principle of maximum entropy is that without external knowledge, one should prefer distributions that are uniform. Using a maximum entropy approach, at first we extract features from the corpus, train maximum entropy model with an annotated corpus, and then use it with additional product feature discovery to extract product features from customer reviews. Our experimental results show that this approach can work effectively for product feature extraction with 71.88% precision and 75.23% recall. © 2008 IEEE.
Some of the metrics are blocked by your
consent settings
The use of computer technology by visually impaired high school students in integrated classrooms in Bangkok
(2004-12-01)
Lalitrojwong, Pattarachai
;
Chaisingharn, Namchoke
This research analyzes the use of computer technology by students with visual impairments in the integrated classrooms in Bangkok. The purposes of the research are to acquire the real facts about applying computer technology in the educational programs by visually impaired students, and to identify problems and needs of the visually impaired students in the integrated classrooms. Data is collected through questionnaires and interviews from all the 56 students of population. The acquired data is then analyzed using the SPSS program in statistic terms of frequency, mean, and percentage. The findings obviously demonstrate that the majority of students employ solely screen reading programs and speech synthesizers. In addition, instructors teaching computer technology to the visually impaired students should be effectively trained to be more skillful and to understand limitations of the way those students learn and follow the instructions in a computer laboratory class. The expensiveness of the technology is also one of the problems that ought to be taken into consideration. Moreover, even though the lack of opportunities to apply computer technology in classrooms is mainly concerned, most of the visually impaired students still need to cooperatively enhance their chance through the universal design of computer technology with general students.
Some of the metrics are blocked by your
consent settings
An Ontology-based Multi-agent System for Matchmaking
(2002-12-01)
Pothipruk, Pakornpong
;
Lalitrojwong, Pattarachai
In a multi-agent system (MAS), at least one agent matchmaking service is required to let any agent in the system find some agents that can help it complete some part of its job. This service is normally in the form of a typical agent called matchmaking agent. In this work, Distributed Matchmaking Agent Architecture (DMAA) is proposed. It is a group of matchmaking agents that provide agent matchmaking service and exploit domain ontology to enhance the matchmaking process. It is composed of a group of three-type matchmaking agents working together: root matchmaking agents (RMMAs), authoritative matchmaking agents (AMMAs), and local matchmaking agents (LMMAs). They represent agents' preferences and capabilities by predicate calculus sentences, and use taxonomic knowledge base to semantically match the preferences and the capabilities. The root matchmaking agent also uses the taxonomic knowledge base to perform ontological load balancing between matchmaking agents in the system so that the system workload from its matching process is reduced. This makes the system perform its task faster.

KMITL

Browse

Filters

Settings

Sort By

Results per page

Search Results