KMITL

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

Browse

Search Results

Now showing 1 - 4 of 4

Some of the metrics are blocked by your
consent settings
High dynamic range preprocessing, ParallelAttention Transformer and CoExpression analysis for facial expression recognition
(2025-04-01)
Zhou, Yuntao
;
Kantathanawat, Thiyaporn
;
Tuntiwongwanich, Somkiat
;
Liu, Chunmao
Facial expression recognition (FER) aims to enable computers to automatically detect and recognize human facial expressions, thereby understanding their emotional states. Despite significant technological advancements in recent years, FER tasks still face several challenges, including expression diversity, individual differences, and the impact of lighting and detail variations on recognition accuracy. To address these challenges, a high-performance FER model is proposed that comprises three key components: High Dynamic Range (HDR) Preprocessing Module, ParallelAttention VisionTransformer structure, and CoExpression Head. In the preprocessing stage, the HDR Preprocessing Module optimizes input images through local contrast and detail enhancement techniques, improving the model's adaptability to lighting and detail variations. During the feature processing stage, the ParallelAttention VisionTransformer structure employs a multi-head self-attention mechanism encoder to effectively capture and process facial expression features at various scales, allowing for a detailed understanding of subtle facial expression differences. Finally, the CoExpression Head utilizes a collaborative expression mechanism to efficiently handle and refine features across different expression states during the feature integration process. Combining these three stages significantly enhances the accuracy of facial expression recognition. Extensive experimental evaluations on public datasets, RAF-DB and AffectNet, demonstrate that the model achieves accuracy rates of 92.11%, 67.25%, and 63.40% on RAF-DB, AffectNet, and AffectNet-8, respectively, exhibiting outstanding performance comparable to other state-of-the-art models.
Some of the metrics are blocked by your
consent settings
Neural Backward Chaining Logic Algorithm Based on Dynamic Knowledge Region Segmentation in Knowledge Graph Completion
(2025-01-01)
Liu, Chunmao
;
Tuntiwongwanich, Somkiat
;
Kantathanawat, Thiyaporn
With the rapid development of the Internet, the information age has arrived in a comprehensive way. The explosive growth of information data has greatly increased people's demand for knowledge management. Knowledge graph, as an effective structured knowledge representation tool, greatly enhances the organization and retrieval capabilities of information. However, in practical applications, knowledge graphs often face problems of knowledge loss and incompleteness, which severely limit their widespread application. To address this issue, this study proposes a neural backward chain inference method based on dynamic knowledge region generation. This method overcomes the bottleneck of performance degradation of traditional static methods on large datasets by introducing a dynamic knowledge region generation mechanism, which significantly improves the completion effect of knowledge graphs. The experiment was conducted on the Unified Medical Language System dataset and the Nations dataset. The results showed that when the size of the Unified Medical Language System dataset reached 2500, the accuracy of the proposed method reached 0.85. It was 6.25%, 20%, and 51.8% higher than the 0.80 of the neural backward chain inference method generated by static knowledge regions, 0.70 of the conditional theorem prover algorithms, and 0.56 of the traditional neural theorems proving algorithm, respectively. In the Nations dataset, the accuracy of the proposed method was 0.80 at the same data scale, which was significantly better than other methods. In addition, the method based on dynamic knowledge region generation reduced the iteration time to 1.4 seconds, which was about 52% and 63% higher than the static method's 2.9 s and the traditional method's 3.8 s, respectively. The research results indicate that the proposed neural backward chain logic algorithm based on dynamic knowledge region generation exhibits good performance in completing knowledge graphs.
Some of the metrics are blocked by your
consent settings
Artificial Intelligence-driven Cognitive Diagnosis anAdaptive Learning: The Impact of Online Course Stickiness and Learning Skills
(2025-01-01)
Liu, Chunmao
;
Tuntiwongwanich, Somkiat
;
Kantathanawat, Thiyaporn
The study analyzes behavioural data that express cognitive connotations, establishes AI-driven cognitive diagnosis and adaptive learning, fills the gap in multimodal data fusion in characterizing cognitive characteristics, constructs a holistic cognitive diagnosis model, and evaluates the impact of AI-driven adaptive learning models on online course stickiness and learning skills. The study innovatively proposes four components of the cognitive diagnosis model: the core theoretical framework of the model composed of cognitive dimensions, the input variables of the model consisting of behavioural data features, the diagnostic model as a method for calculating cognitive states, and the cognitive diagnosis output. The study sorted out the core theoretical framework of the model and proposed behavioural data feature input variables, and completed the cognitive diagnosis output in three dimensions: learning momentum, effectiveness, and strategy through the XGBoost model based on the Gradient Boosting framework.An adaptive online learning model based on behavioural data cognitive diagnosis and knowledge graph is proposed, which includes six parts: input layer, feature extraction layer, cognitive diagnosis module, learning path recommendation module and output layer. The cognitive diagnosis module uses the feature weights calculated by XGBoost as input to predict the mastery of knowledge points through LSTM improved by deep learning recurrent neural network (RNN) and makes learning recommendations based on knowledge graph, cognitive evaluation matrix CEM and collaborative filtering algorithm. Experimental results show that the adaptive learning model of behavioural data cognitive diagnosis has more advantages than traditional online learning, and the adaptive online learning model driven by artificial intelligence behavioural data mining can effectively improve course stickiness, learning skills and platform experience.
Some of the metrics are blocked by your
consent settings
Enhancing Personalized Learning in Online Education: The Impact of Adaptive Learning Systems and Recommendation Technologies
(2024-01-01)
Liu, Chunmao
;
Tuntiwongwanich, Somkiat
;
Kantathanawat, Thiyaporn
The study investigates the impact of integrated adaptive learning systems and recommender technologies on the improvement of online education. A component-level quantitative evaluation was conducted, which involved measuring user interaction, content applicability, knowledge acquisition, and system usability, with support from surveys and interviews. The findings indicate that recommendation systems enhance active user participation, content relevance, and learning outcomes, while maintaining high usability rates that positively influence learners’ perceptions. However, certain limitations were identified, including the system’s less-than-ideal suitability for advanced learners and the absence of contextual information. The study concludes that, when appropriately implemented as suggested by existing literature, adaptive learning systems possess significant potential to transform online education by offering personalised and efficient learning methods. Recommendations for future developments include the integration of third-generation machine learning, ensuring equal opportunities for learners, and further refining the system to address small learner differences.

KMITL

Browse

Filters

Settings

Sort By

Results per page

Search Results