電腦科學與資訊工程

In Brazil, approximately sixty million people suffer from or acquire some type of disease daily. However, the average time for blood count diagnoses, used to identify many of these diseases, remains very lengthy. This can lead to the worsening of conditions and delays in care, as well as a decrease in the patients’ quality of life. Moreover, in some cases, the waiting period can result in irreversible situations and even the death of the affected individuals. In this landscape, technological tools such as artificial intelligence software can help reduce the time taken for diagnostic reporting. In light of this, the project involves developing software to assist in the analysis of blood counts and optimize medical diagnoses. For this purpose, the methodology was divided into three stages. In the first, titled ”Medical Standardization”, a survey of the standard variables related to diseases that can be identified with the help of blood counts was conducted. Among the findings, diabetes, anemia, leukemia, dengue, polycythemia, tuberculosis, leprosy, meningitis, chlamydia, schistosomiasis, spotted fever, and malaria were the main diseases detected. Furthermore, hemoglobin, leukocytes, platelets, glucose, cholesterol, ions, and hormones were the key findings concerning the primary blood indicative factors for the mentioned diseases. In the second phase, the theoretical and practical foundations of the software were developed, based on artificial neural networks. In Python, regression models were also crafted to check the feasibility of the analyses. Finally, the last stage consisted of testing with real datasets, based on 1,227 anonymized blood counts. Among the artificial intelligence algorithm models tested, Support Vector (0.02) and Multiple Linear (0.61) had the lowest performances, while Polynomial (0.97), Random Forest (1.0), and Decision Tree (1.0) showed the best results. Given that the Random Forest and Decision Tree regression models achieved an accuracy of 1.0, while the Polynomial model scored 0.97, Support Vector 0.02, and Multiple Linear Regression 0.61, it is concluded that the blood count analysis system, with Python tools like regression, proved to be highly efficient. The closer the R² value is to 1.0, the better the programming fits the model, ensuring accurate analyses. Aside from that, in order to expand the number of analysis possible to do be done we decided to use a second tool called ”classification”, with which we made a bigger dataset to be used as a model to identify blood related diseases and the behavior of complex and diverse diseases. With that in mind, we performed a second evaluation of the models by doing an accuracy test, scored 87 percentage points and with a confusion matrix. With those results, we verified that the high performance of the tests indicates that Artificial Intelligence can be avaunt-guard to the elaboration of more efficient medical diagnosis, improving people’s lives quality and, overall, lowering the number of deaths in our country.

Wibrazz is a wearable communication tool that allows the teacher, the therapist, the parent to communicate information to the child remotely using the device. Haptic (vibrationbased) feedback is becoming increasingly important in everyday life. A vibrating device that transmits information through clothing can help people with disabilities who have no or limited sensory use to live an integrated life in society without barriers.

Road safety remains a global concern, with driver-related factors like distraction, drowsiness, and medical conditions such as stroke being leading causes of accidents. In this paper, we propose a real-time ensemble learning framework that leverages transfer learning for the detection of stroke, drowsiness, and distracted driving. Our model integrates multiple Convolutional Neural Networks (CNNs) fine-tuned for each specific task, and employs a stacking method to combine the predictions of these models using a meta-classifier. Notably, the model is optimized to enhance stroke detection, minimizing false negatives— an essential aspect for timely medical intervention. Experimental evaluations on diverse datasets demonstrate the efficacy of our approach, achieving an overall accuracy of 92.5%. The results emphasize the model’s potential for real-time driver monitoring, offering critical safety features that could reduce accidents and save lives.

現今醫療中，黃疸的早期偵測對肝臟疾病的預防與治療至關重要，但多數人難以在症狀輕微時察覺。我們希望藉由智慧手機影像結合機器學習進行黃疸檢測，提升民眾自我監測的能力。Su 等人（2021）曾使用深度學習和機器學習進行黃疸預測，但其方法依賴專業色卡進行色彩校正，成本高且限制應用範圍。本研究提出以白平衡演算法中的白色補丁法與灰界演算法，搭配深度學習模型 DCCNM1和2 取代色卡，提升黃疸檢測的普及性與便利性。經黃疸偵測效果評估顯示，DCCNM2 在無色卡模型中表現最佳，雖然各指標略低於色卡校正，但其展現出優異的穩定性和準確性，證明其作為無色卡黃疸篩檢方案的可行性。本方法將能提供便捷的居家黃疸檢測途徑，尤其對偏鄉地區居民而言，不僅提升早期發現的機會，還能有效減輕醫護人員的負擔，推動大眾健康管理。

This report presents a novel AI-driven tool for aiding musical composition through the generation of chord progressions. Data acquisition and analysis are discussed, uncovering intriguing patterns in chord progressions across diverse musical genres and periods. We developed a range of deep learning models, from basic recurrent networks to sophisticated Transformer architectures, including conditional and style-based Transformers for improved controllability. Human evaluation indicates that, within the context of our specific data processing methods, the chord sequences generated by the more advanced models are practically indistinguishable from real sequences. The models are then integrated into a userfriendly open-source web application, making advanced music composition tools accessible to a broader audience.

In this fast-paced digital economy, customers' judgment is based on their experience with a company’s products and services. Customer reviews become a vital source of information for companies because this information can be used to enhance their products, understand customer wants and needs, improve brand reputation, and provide a competitor’s advantage. A company can understand customer needs and wants by going through reviews. Customers are encouraged to leave not only their opinion but also their ideas for the development of the product or service. By understanding these reviews, a company can actively respond and engage with a reviewer or problem. Failure of companies who don't answer customer queries may negatively impact customer loyalty. Customers will feel neglected by these companies and will choose competing companies to handle their needs. Additionally, customers may speak negatively about a company that does not respond to reviews. The AI-based customer sentiment dashboard can help gain a company's competitive advantage by identifying weaknesses in themselves and others. Companies will be enabled to understand where they succeed and where improvement is needed compared to their competitors, leveraging businesses to address strengths and weaknesses before competitors do. Through AI-based customer sentiment dashboards, a company can analyze its competitor’s reviews and use that information as leverage to make improvements to its products and services. Customers are increasingly leaving reviews on popular apps like Google Play, Stamped.io, Yapto, and Judge.me, Loox, Qualaroo, and Yelp. The reviews are rich in customer sentiments offering valuable insights into user satisfaction and pointing out the areas for improvement that are crucial to every company no matter how big or small. Despite their value, manually processing these reviews is a challenging task due to the large volume of unstructured data. Manual processing is also vulnerable to bias and human error, leading to inaccurate information. Traditional methods such as surveys have been proven to be ineffective if the main focus is targeted feedback and have low responses compared to reviews. The advances in artificial intelligence like Natural Language Processing (NLP) help us interpret and analyze human language and generate outputs like predicting what type of sentiments are in reviews. This project proposes developing an AI-based sentiment analysis model to evaluate customer feedback on two widely used taxi applications. Natural Language Processing libraries, such as the Valence Aware Dictionary and Sentiment Reasoner (. The model aims to categorize customer reviews into positive, negative, and neutral sentiments.

1.1 Short project summary My project involves the conceptualization and development of an innovative approach to modular self-assembling robotic systems. Through its ability to form any complex configuration, the system is highly adaptable to various scenarios and environments. Before delving deeper into the details of my project, I will provide an overview of my background and motivations. 1.2 Background Ever since I first watched the movie "Big Hero 6", I felt amazed by the applications of the so called “microbots”. From that point on, it made me always wonder what would be possible in the real world. When I did the research, I stumbled upon this field of modular robotics. Initially, I was unsure whether to embark on a project focused on electronics and robotics due to my background in programming. On the other side, this year gave me a chance to see the incredible performances of various projects at different science expos. Besides, I took part in the program of CANSAT LU and learned a lot during it, such as microchips, the control of miniature robotics, and the sensors of it. Finally, at school, I took the option Electronics where we dig into similar topics. With this accumulated knowledge and experience I felt confident enough to start this project.

發展安全且可靠的身份辨識技術是當今的重要議題，而指靜脈因其高安全性及難以偽造特性成為我們的主題。本研究提出一種基於Transformer模型架構的指靜脈辨識模型稱為GLA-FD，旨在解決現有技術對指靜脈影像特徵表示與提取的局限性。透過開發特徵解耦與重建模組(FDRM)，模型能夠有效區分指靜脈的背景資訊與紋理特徵，並將其重新組合以提升辨識準確度。此外，本研究開發的全域-局部注意力模組(GLAM)能同時捕捉影像的全域與局部特徵，進一步強化模型對指靜脈特徵的理解。GLA-FD在FV-USM、PLUSVein-FV3、MMCBNU-6000、UTFVP、NUPT-FPV 資料集中的正確辨識率(CIR)達到100%、98.47%、99.75%、96.11%、99.82%，展現卓越的穩定性與泛化能力。此外，本模型在處理不同年齡層、國籍與影像模糊度的資料下，仍能保持高辨識準確度，顯示其在需要高安全性辨識的應用場景中具備廣泛的實用性。

在臺灣手語辨識，先前研究所使用的監督式學習需要大量標記樣本而限制可辨識詞彙量。為此，本研究借鑒自然語言處理領域中BERT 的遮罩想法，將未標記手語影片隨機遮蓋部分幀數，並讓模型學習預測被遮蓋的幀數以學習臺灣手語的特徵，並透過遷移學習來訓練辨識模型，此作法可克服現有臺灣手語資料缺少的問題。經過實驗，本研究訓練之詞彙辨識模型達成了242 個詞彙量，94.8%的準確率。 此外，先前研究皆未在手語句子翻譯上有成果。因此本研究基於預訓練模型，整合設計手語翻譯的系統，實驗中，系統在100 個句子的翻譯表現達到88%的準，且BLEU-4 分數取得20.98，證明自監督學習的方式在手語辨識、翻譯上是有效的。並展現出樣本需求少與辨識詞彙量可輕易擴大的潛力。

「在格狀平面中用矩形以互不重疊的方式鋪滿(2D rectangle tiling problem)」為一NP-complete問題(Dani`ele Beauquier et al ,1995)，目前多項式時間只能求出盡可能覆蓋最大面積的近似解。本研究所創的階梯演算法 stair algorithm 透過改變動態規劃紀錄狀態的方式，使狀態數大幅減少，進而改善求準確解的時間複雜度，也成功證明此演算法的正確性。本研究的演算法可被應用於平行計算中的負載平衡、積體電路設計等方面。隨後，本研究寫了一個互動展示品清楚呈現此演算法的功能。且以階梯演算法成功檢驗並比較 RTILE PROBLEM 的 7/3-approximation algorithm (Krzysztof Lorys and Katarzyna E. Paluch,2000 [4]) 與 11/5-approximation algorithm (Piotr Berman et al,2001[7])進行比較與分析。