KMITL

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

Browse

Search Results

Now showing 1 - 10 of 17

Some of the metrics are blocked by your
consent settings
Improved Naive RAG by Integrated Advanced Techniques: A Comprehensive Framework Using Parent-Child Architecture, Hybrid Retrieval, and Contextual Compression
(2026-07-01)
Aromsuk, Tinnarat
;
Netisopakul, Ponrudee
;
Nootyaskool, Supakit
This paper presents a novel approach to enhance Retrieval-Augmented Generation (RAG) systems through the integration of three advanced techniques: Parent-Child Architecture, Hybrid Retrieval, and Contextual Compression with cross-encoder re-ranking. We implement this framework using Langchain and FAISS vector search, with Anthropic’s Claude as the foundation model. Our comprehensive evaluation across 14 diverse Wikipedia-based knowledge domains employs the RAGAS framework to measure multiple performance dimensions. Results demonstrate that our advanced framework yields significant improvements in key metrics: Context Precision 10.11%, Context Recall 2.25%,and BLEU scores 1.14% compared to Naive RAG implementations. Domain analysis reveals particularly strong performance in Medicine 8.0% BLEU, Science 4.9%, and specialized Technology domains 4.9%. While, some technical domains such as Cybersecurity (−2.4%) and Biology (−6.6%) show performance degradation. Our framework achieves these improvements with minimal computational overhead by 1.89%,offering a practical approach to implementing domain-adaptive RAG systems that optimize context quality for improved generation performance.
Some of the metrics are blocked by your
consent settings
DEEP LEARNING PREDICTING POWER COSTS FOR POWER PLANNING
(2025-05-01)
Ounsrimung, Pimolrat
;
Phuanphannid, Walaiphan
;
Nootyaskool, Supakit
The current global situation, especially in Thailand, is showing recovery following the easing of COVID-19 restrictions. With a growing population and industrial expansion, there is an increased demand for electricity. This rise in demand, along with factors like fuel costs and fluctuating currency rates, has influenced global power prices. The authors are keen on studying power’s variable price prediction using machine learning techniques. The objective is to analyze factors associated with power costs and their interrelationships. It is essential to have accurate data for strategic power planning. Research has been conducted on theories related to variable power prices, time series, traditional statistical forecasting, machine learning, and deep learning. It was found that three main factors affecting variable power prices were identified: natural gas prices, exchange rates, and inflation rates, with natural gas prices and inflation rates having a strong correlation. Traditional statistical forecasting is less efficient for highly volatile power’s variable price. Deep learning models outperform conventional machine learning for this dataset.
Some of the metrics are blocked by your
consent settings
Quantitative Assessment of Retrieval Strategies in RAG Architectures: A Comparative Study Across Multiple Knowledge Domains Using Standardized Performance Metrics
(2025-01-01)
Aromsuk, Tinnarat
;
Nootyaskool, Supakit
;
Netisopakul, Ponrudee
This study conducts a systematic quantitative assessment of four distinct retrieval methodologies within Retrieval-Augmented Generation (RAG) frameworks: baseline implementation, hybrid dual-paradigm approach, hierarchical parent-child structure, and contextual compression. Through rigorous experimental evaluation spanning six distinct knowledge domains, we employ established metrics including ROUGE [1], BLEU [2], and computational timing measurements to characterize performance profiles. Our findings reveal that sophisticated retrieval approaches deliver substantial computational efficiency gains (4-5 × acceleration) alongside varied performance patterns across quality assessment dimensions. The hierarchical parent-child methodology demonstrates superior BLEU performance (0.1046 mean score) coupled with optimal retrieval speeds (0.0124 s), whereas hybrid approaches excel in ROUGE metrics (0.0317 mean score). Domain-specific analysis indicates pronounced performance disparities: medical/health domains (COVID19 pandemic) achieve highest aggregate scores (0.1198 mean ROUGE), while specialized technical and legal domains present distinct retrieval complexities. This research establishes empirical foundations for evidence-based retrieval method selection, identifying clear efficiency-quality relationships and domain-dependent optimization strategies for production RAG deployments.
Some of the metrics are blocked by your
consent settings
Predicting Fuel Burn with Neural Network to Adjust Contingency Fuel of Airplane
(2023-01-01)
Ounsrimoung, Pimolrat
;
Nootyaskool, Supakit
;
Atchariyachanvanich, Kanokwan
;
Yooyen, Soemsak
The amount of fuel in an airplane tank is very important for flying. however, flying a short distance by adding a fuel-full tank is not energy efficient because spending a lot of tons for holding fuel weight. The flight planners who consider the amount of fuel to add to the tank by using historical data, use fuel burn calculating and adjust contingency fuel. This research presents the neural networks to predict fuel burn, which learn from historical airplane data. The experiment applied to local and international flight data and used both Airbus and Boeing. The predicted model was swapped and tested on the outbound and inbound replacements for confirmation capable of the predicted mode.
Some of the metrics are blocked by your
consent settings
Categorize Level of Crystal Sugar Making with Recurrent Neural Network
(2022-01-01)
Ounsrimuang, Pimolrat
;
Nootyaskool, Supakit
This research presents the study of recurrent neural networks to predict industrial crystal sugar making. The recurrent neural network trains on six parameters consisting of liquid in the pan, Brix levels, vacuum in the pan, liquor temperatures, water steam supplier, and current for mix-motor agitator. The input variables were the trained model to predict by categorizing data in three levels high, middle, and low which the data came from human control the sugar boiler machine. The trained model for the future can be extended to make an experience meter to indicate the ability of workers to control the machine.
Some of the metrics are blocked by your
consent settings
Gold Investment Model on RNN and Finding Best Investment Strategy on PSO
(2022-01-01)
Kanchanakantikul, Pakamas
;
Nootyaskool, Supakit
Nowadays, Algorithm trading in community and stock is interesting research, while gold is also an investment option. This research presents two steps. Three inputs sequence consists of the gold price(sell), gold spot and crude oil. Output has an order sequence indicating buy, sell, and wait for the signal. Firstly, finding the best strategy from historical data by particle swarm optimization (PSO) compared with random search (RS). That will get buying, selling, or waiting signals in the gold trading market Secondly, creating gold investment by recurrent neural network (RNN) model. The experiment result showed RNN trading model based on PSO is better than RS, which has a profit of 79.667 percent.
Some of the metrics are blocked by your
consent settings
Smart collar design to predict cow behavior
(2020-11-04)
Nootyaskool, Supakit
;
Ounsrimung, Pimolorat
Productivity from agriculture and farming is one major to drive Thailand's economy. Low price sensors and easy to make the Internet of Thing devices for making the data collector applied in various fields such as industry, communication, and transportation. However, smart application in agriculture does not apply widespread usage. This research proposed the detail design of the smart cow collar to use monitoring cow healthy. Data of walking and scraping of the cow sent to collect on the server and report to the user known the cow healthy by relating to the behavior. The purpose of the device helps the cow keeper reducing the monitoring of cow behaviors and it can predict illness and vital of the cow occurring in the future. For predicting part, design by the data from the gyroscopes after preparing send to Hidden Markov Models to predict cow behavior, however, this paper mainly focuses on the detail building the cow collar and creating the prototype device, then testing on the real environment.
Some of the metrics are blocked by your
consent settings
Reinforcement learning applied to scrum team towards large-scale global optimization
(2019-02-02)
Nootyaskool, Supakit
;
Ounsrimuang, Pimolrat
Large-scale problems have size of problem over a thousand dimensions in finding a best solution that uses long computation times. In this work, we use an idea of scrum methodology that is a well-known in software development companies, to create an optimization algorithm. The scrum methodology describing about the team organization likes as rugby team management that player have expert in game. The proposed algorithm is developed based on concept of the evolutionary computation by this work added agent specifics in leaning environment of the problem. The specific of the agent is reinforcement learning by taking an action and getting reward. The proposed algorithm was experimented on a large-scale global optimization finding optimum point of numerical function, comparing between with and without reinforcement learning. The experiment result showed that the usage of reinforcement learning has good results.
Some of the metrics are blocked by your
consent settings
Introducing scrum process optimization
(2017-11-14)
Ounsrimuang, Pimolrat
;
Nootyaskool, Supakit
Scrum is a well-known methodology in software development describing as organization in a developer team. A scrum team is a small size with expert members, rich communication, sharing knowledge, self-organization, and self-planning. Scrum process in this version developed by an attempt to create a self-organization team, which an action-reward function integrated in the proposed algorithm. This paper proposed a scrum process creating an optimization algorithm in the class of evolutionary computation. The proposed algorithm experimented on 30 numerical functions by the benchmark coding from the CEC2017 competition problems. The experiment results indicate that uses action-reward function to organize operation planning can help the proposed algorithm finding the best result.
Some of the metrics are blocked by your
consent settings
Roulette wheel selection applied to PSO on numerical function in discrete and continuous space
(2016-07-22)
Ounsrimuang, Pimolrat
;
Nootyaskool, Supakit
Particle Swarm Optimization (PSO) successfully finds a solution as shown in various literatures. In some problems creating on discrete space, adjustment control-parameter may be difficult to modify a reach of optimum solution. The paper proposes an approach applying roulette wheel selection to PSO, which can help PSO escape from a local solution. This approach tested on both continuous and discrete space by finding solution of 12-numerical functions and an engineering-problem. The experiment result showed that the proposed technique can help PSO getting the best result both problem spaces, the performance improvement but also maintain easily to implementation.

KMITL

Browse

Filters

Settings

Sort By

Results per page

Search Results