KMITL

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

Browse

Search Results

Now showing 1 - 10 of 24

Some of the metrics are blocked by your
consent settings
Self-attention hierarchical kernel reservoir state network for inland water level prediction
(2026-01-15)
Liu, Zongying
;
Xu, Xiaohan
;
Pasupa, Kitsuchart
;
Loo, Chu Kiong
;
Wei, Yang
Waterway transportation sustainably facilitates global trade through eco-efficient cargo movement, where accurate water level forecasting is critical for ensuring navigational safety and operational continuity. To develop a highly accurate prediction model, it is essential to consider the periodic characteristics of water level data, which often emerge in real-world datasets. This study introduces a novel reservoir state structure based on reservoir computing theory, the Self-attention Hierarchical Kernel Reservoir State Network (SHK-RSN). It employs three primary mechanisms. First, a hierarchical feature extraction method groups training data and extracts high-dimensional features from these groups using the kernel trick in a hierarchical manner. Second, a self-attention weight selection approach is introduced to replace the random weights in the Hierarchical Kernel Reservoir State Network (HK-RSN), improving the rationale for hidden neuron connections and enhancing the interpretability of weight selection. Third, a novel reservoir state structure is proposed to capture periodic information and extract temporal features across periods, enabling the model to capture richer temporal information and identify relationships among periods. Experiments are conducted on one artificial and five real-world time series datasets, with forecast performance evaluated over 1–7 steps. Our proposed model, SHK-RSN, is compared with models based on randomization, the kernel trick, and deep learning. The experimental results demonstrate that SHK-RSN exhibits superior forecasting ability relative to the baselines. It achieves the best Symmetric Mean Absolute Percentage Error (SMAPE) across all datasets in the 1–7 period average among baseline methods, demonstrating a relative improvement of 25.7% to 46.9% over the conventional Echo State Network.
Some of the metrics are blocked by your
consent settings
Highlight Detection in Podcasts: A Multimodal Deep Learning Approach
(2025-01-01)
Phuengpanyaloet, Wongsapat
;
Boonruengkhao, Nonpipat
;
Anchutin, Viktor
;
Pasupa, Kitsuchart
;
Loo, Chu Kiong
Podcasts have become a pervasive form of digital media, offering diverse content that often spans long hours. However, the vast volume of podcast episodes can make it challenging for listeners to locate the most engaging segments. Speech Emotion Recognition (SER) has witnessed remarkable advancements with the integration of deep learning techniques. This work proposes utilizing deep learning techniques employed in SER to discern emotional cues within podcasts, thereby enabling the detection of highlights. The task is framed as a binary classification problem, where the positive class contains examples of speech segments with high emotional activation. Transfer learning techniques from computer vision and speech recognition domains are applied, utilizing pre-trained models such as ConvNeXt, Vision Transformer, and wav2vec 2.0, which are compared with a baseline Convolutional Neural Network-Transformer hybrid. Additionally, multimodal models are introduced that learn from two distinct modalities: log mel-spectrograms and high-dimensional vector embeddings, both extracted from the raw audio data. The two modalities are combined using (i) a Simple Concatenated and (ii) CentralNet models. Experimental results demonstrate the effectiveness of combining two modalities over a single modality, achieving F<inf>1</inf>-scores of 0.6111 and 0.6270 for the Simple Concatenated and CentralNet models, respectively.
Some of the metrics are blocked by your
consent settings
Weighted error-output recurrent Xavier echo state network for concept drift handling in water level prediction
(2024-11-01)
Liu, Zongying
;
Zhang, Wenru
;
Pan, Mingyang
;
Loo, Chu Kiong
;
Pasupa, Kitsuchart
Water level holds utmost significance in maritime domains. Precise water level predictions furnish indispensable insights for safe maritime navigation, guiding ships and vessels through passages, harbors, and waterways. This paper introduces a novel approach: the Weighted Error-Output Recurrent Xavier Echo State Network with Adaptive Forgetting Factor (WER-XESN-AFF). One of the contributions of this study is the introduction of the Xavier weights selection method, which replaces random weight selection from the Echo State Network (ESN). This method not only enhances forecasting performance but also reduces uncertainty in predictions. Additionally, two modified concept drift detectors, the Early Drift Detection Method and the Adaptive Forgetting Factor, are employed to address concept drift challenges. Another notable contribution is the introduction of a novel weighted error-output recurrent multi-step algorithm. This algorithm successfully overcomes the error accumulation problem by using past forecast errors to update current output weights. This study performs extensive experiments to evaluate the effectiveness of our approach in multi-step prediction in synthetic and real datasets. It compares the performance between the conventional randomization-based models and the ESN with the new weights selection approach and also tests the ability of concept drift detectors and the weighted error-output multi-step algorithm. Empirical findings and statistical analyses demonstrate that our proposed methods achieve expected effects, and the proposed model has better prediction ability than baselines. A significant improvement rate of 75.39% in Mean Squared Error is evident within the Jiujiang water level dataset when contrasting the performance of WER-XESN-AFF against the baseline model R-ESN across the 1–5 period.
Some of the metrics are blocked by your
consent settings
Optimizing echo state networks for continuous gesture recognition in mobile devices: A comparative study
(2024-03-15)
Yadav, Alok
;
Pasupa, Kitsuchart
;
Loo, Chu Kiong
;
Liu, Xiaofeng
Continuous gesture recognition can be used to enhance human-computer interaction. This can be accomplished by capturing human movement with the use of the Inertial Measurement Units in smartphones and using machine learning algorithms to predict the intended gestures. Echo State Networks (ESNs) consist of a fixed internal reservoir that is able to generate rich and diverse nonlinear dynamics in response to input signals that capture temporal dependencies within the signal. This makes ESNs well-suited for time series prediction tasks, such as continuous gesture recognition. However, their application has not been rigorously explored, with regard to gesture recognition. In this study, we sought to enhance the efficacy of ESN models in continuous gesture recognition by exploring diverse model structures, fine-tuning hyperparameters, and experimenting with various training approaches. We used three different training schemes that used the Leave-one-out Cross-validation (LOOCV) protocol to investigate the performance in real-world scenarios with different levels of data availability: Leaving out data from one user to use for testing (F<inf>1</inf>-score: 0.89), leaving out a fraction of data from all users to use in testing (F<inf>1</inf>-score: 0.96), and training and testing using LOOCV on a single user (F<inf>1</inf>-score: 0.99). The obtained results outperformed the Long Short-Term Memory (LSTM) performance from past research (F<inf>1</inf>-score: 0.87) while maintaining a low training time of approximately 13 seconds compared to 63 seconds for the LSTM model. Additionally, we further explored the performance of the ESN models through behaviour space analysis using memory capacity, Kernel Rank, and Generalization Rank. Our results demonstrate that ESNs can be optimized to achieve high performance on gesture recognition in mobile devices on multiple levels of data availability. These findings highlight the practical ability of ESNs to enhance human-computer interaction.
Some of the metrics are blocked by your
consent settings
Correlated Online k-Nearest Neighbors Regressor Chain for Online Multi-output Regression
(2024-01-01)
Wu, Zipeng
;
Loo, Chu Kiong
;
Pasupa, Kitsuchart
Online multi-output regression is a crucial task in machine learning with applications in various domains such as environmental monitoring, energy efficiency prediction, and water quality prediction. This paper introduces CONNRC, a novel algorithm designed to address online multi-output regression challenges and provide accurate real-time predictions. CONNRC builds upon the k-nearest neighbor algorithm in an online manner and incorporates a relevant chain structure to effectively capture and utilize correlations among structured multi-outputs. The main contribution of this work lies in the potential of CONNRC to enhance the accuracy and efficiency of real-time predictions across diverse application domains. Through a comprehensive experimental evaluation on six real-world datasets, CONNRC is compared against five existing online regression algorithms. The consistent results highlight that CONNRC consistently outperforms the other algorithms in terms of average Mean Absolute Error, demonstrating its superior accuracy in multi-output regression tasks. However, the time performance of CONNRC requires further improvement, indicating an area for future research and optimization.
Some of the metrics are blocked by your
consent settings
CiRA CORE: A Low Code Platform that Makes AI Work for Industry 4.0
(2024-01-01)
Loo, Chu Kiong
;
Boonsang, Siridech
;
Sasisaowapak, Thanyathep
;
Chuwongin, Santhad
;
Tongloy, Teerawat
CiRA CORE is a central hub designed to connect AI technology creation with practical application, making it easier to work with ROS (Robot Operating System) and link different systems through a user-friendly drag-and-drop interface. This approach removes the need for extensive coding, making the platform accessible to those with minimal programming experience. CiRA CORE offers a comprehensive suite of features for AI development and robot control, including algorithm creation, AI model training, and device integration commonly used in industrial settings. It supports tasks like image recognition and facilitates data storage, labeling, and integration with other systems for data-driven AI development. Overall, CiRA CORE aims to democratize AI development and robot control, simplifying AI development for Industry 4.0 applications, and leading to increased efficiency, reduced costs, and improved safety in industrial processes. This paper reports the progress of the CiRA CORE training modules funded by the SMCS TEAM Program Award. The project has completed the design of a 6-axis robot 3D training kit and simulation models for CiRA CORE training modules. The next steps involve developing 3D-printed robots and training materials. The main goal is to democratize advanced robotics and AI by simplifying integration through a visual, node-based programming interface. This approach reduces the need for complex coding, making these technologies accessible to users with limited programming experience. This initiative aims to foster widespread adoption in business and industrial settings, aligning with IEEE SMC's mission to promote professional growth and innovation in robotics and AI.
Some of the metrics are blocked by your
consent settings
A novel online multi-task learning for COVID-19 multi-output spatio-temporal prediction
(2023-08-01)
Wu, Zipeng
;
Loo, Chu Kiong
;
Obaidellah, Unaizah
;
Pasupa, Kitsuchart
In light of the ongoing COVID-19 pandemic, predicting its trend would significantly impact decision-making. However, this is not a straightforward task due to three main difficulties: temporal autocorrelation, spatial dependency, and concept drift caused by virus mutations and lockdown policies. Although machine learning has been extensively used in related work, no previous research has successfully addressed all three challenges simultaneously. To overcome this challenge, we developed a novel online multi-task regression algorithm that incorporates a chain structure to capture spatial dependency, the ADWIN drift detector to adapt to concept drift, and the lag time series feature to capture temporal autocorrelation. We conducted several comparative experiments based on the number of daily confirmed cases in 20 areas in California and affiliated cities. The results from our experiments demonstrate that our proposed model is superior in adapting to concept drift in COVID-19 data and capturing spatial dependencies across various regions. This leads to a significant improvement in prediction accuracy when compared to existing state-of-the-art batch machine learning methods, such as N-Beats, DeepAR, TCN, and LSTM.
Some of the metrics are blocked by your
consent settings
CowXNet: An automated cow estrus detection system
(2023-01-01)
Lodkaew, Thanawat
;
Pasupa, Kitsuchart
;
Loo, Chu Kiong
Estrus detection is essential for dairy farms to take cows for artificial insemination promptly. Conventional approaches for detecting estrus cows use electronic devices attached to cows to gather data for software analysis. However, electronic devices can be costly and make a cow moody and uncomfortable while moving. In a common approach, observers detect estrus cows by observing their behaviors. However, continuous observation can easily lead to errors due to the observer's fatigue. Therefore, we designed CowXNet, an automatic estrus detection system for cows, to assist farmers to detect estrus cows. CowXNet requires only a camera attached in a pen and a computer to analyze recorded videos. CowXNet analyzes the estrus behaviors of each cow in a pen and helps farmers to identify estrus cows. To develop and evaluate CowXNet efficiently and effectively, we collected data from Chokchai Farm, the biggest dairy farm in Asia (14.65483<sup>∘</sup>N, 101.34853<sup>∘</sup>E). CowXNet has four modules: (i) cow detection uses YOLOv4 to detect cows in recorded videos; (ii) body part detection uses a convolutional neural network to estimate locations of body parts of detected cows; (iii) estrus behavior detection uses body part coordinates to extract a set of discriminative features, and a classification algorithm to detect estrus behaviors, and (iv) behavior analysis module displays estrus behavior for analysis purposes. We evaluated CowXNet for two instances: module-independent evaluation and end-to-end framework evaluation. Overall, CowXNet was promising; it correctly detected estrus behavior interval of cows 83% of cases.
Some of the metrics are blocked by your
consent settings
An Interpretable Multi-target Regression Method for Hierarchical Load Forecasting
(2023-01-01)
Wu, Zipeng
;
Loo, Chu Kiong
;
Pasupa, Kitsuchart
;
Xu, Licheng
Accurate energy load forecasting provides good decision support for energy management. Current energy load forecasts focus more on forecast accuracy without exploring the similar patterns and correlations of energy load demand between regions. Our proposed interpretable hybrid multi-target regression approach provides more explanatory abilities for each region’s energy load prediction. After combining the correlation between forecast targets and hierarchical forecast information, our model achieves a high forecast accuracy in that the mean square error is reduced by three quarters compared to LightGBM’s independent prediction for each region on the GEFCom 2017 dataset.
Some of the metrics are blocked by your
consent settings
Energy-Efficient Continual Learning for Autonomous Driving
(2023-01-01)
Ng, Qi Ding
;
Loo, Chu Kiong
;
Pasupa, Kitsuchart
;
Dilokthanakul, Nat
;
Zhang, Jie
Our work highlighted the primary challenges of Autonomous Driving (AD), namely the Catastrophic Forgetting (CF) of previous knowledge by the AD system upon new scenario encounters. Considering the infeasible model retraining with past data given computational, power, and storage constraints on the embedded device, we proposed an experiment featuring Avalanche Continual Learning (CL) training strategies to investigate which strategies excel in this task and combine the promising ones in the hope for a more balanced and efficient trade-off between performance and energy consumption. Our experiment unprecedentedly validated the candidates against a new benchmark introducing natural distribution change and time correlation between input images. We found that although a synergy of CL strategies yields higher resistance towards CF, the slight accuracy gain is not worth the additional computation when we account for energy consumption, rendering a simple Replay strategy the best solution for the Continual Learning benchmark for Autonomous Driving: Online Continual Classification (CLAD-C). Our proposal delivers a 65.80% improvement over the baseline at our proposed accuracy-power ratio metric.

KMITL

Browse

Filters

Settings

Sort By

Results per page

Search Results