長距無人機圖傳系統
High power microwave for knocking out programmable suicide drones
物聯網與行動通訊網路(5G/6G行動通訊，長距多媒體物聯網)
智慧物聯網與邊緣計算(Edge AI computing, AI on device)
智能合約之匿名線上交易、金融、醫療大數據分析與機器學習
心電圖機器自動判讀 (ECG auto-recognition)
機器學習模型設計與應用(Deep reinforcement learning for network research)
實驗室討論區：
Title: 以HPM (High power microwave)定向能量，擊落流氓無人機群或防禦軍事無人機群攻擊
主要研究目的是減少流氓無人機對我們生活的危害以及極端組織、毒販和有組織犯罪使用它們的後果，甚至包含防禦軍事衝突中的無人機群的攻擊。越來越多涉及改裝無人機的事件，證明現有技術在阻止非合法使用之無人機方面的不足，例如手持干擾槍、訓練鷹、射頻干擾器等。這些技術不太可能擊落流氓無人機群，而且無法阻止程式自動控制的無人機。本研究旨在探討並開發實作出使用HPM（高功率微波）的定向能量的原型機，利用電磁場強度能量損壞無人機結構或燒毀其PCB板電子元件。接著分析使用高頻微波功率進行電子攻擊，立即關閉無人機。評估高微波功率在不同距離和不同天氣條件下，干擾無人機的效果。還包括對磁控管耦合系統的錐形喇叭天線（操作頻率為2.45 GHz）的研究。最終目標仍以作戰時之無人機群攻擊的防禦用途為主。
Title: ChatGPT在焦慮、情緒相關疾病上的應用：一個12周追蹤研究。
憂鬱、情感、焦慮及壓力相關疾病，常互為因果且高度相關，亦佔精神科門診的大宗。然而，礙於門診時間有限，醫師長無法在門診時間詳盡和病患溝通病情，隨著網路與智慧型手機的普及，民眾已習慣在網路查詢醫學知識，隨著ChatGPT的開發，能應用生成式人工智慧整理與搜集資料，因此，若結合ChatGPT，將能協助就診患者後續進行相關衛教，讓患者更深入且完整地了解其疾病或治療計畫的相關資訊。目前，ChatGPT在精神醫療上的應用，目前仍多為假設預測，尚無相關前瞻性研究，亦無針對ChatGPT在精神科問題相關回覆上準確性與實用性的評估。本研究將分為2步驟：
1) 在計畫前3個月，驗證ChatGPT針對憂鬱症、焦慮症，雙極症、創傷及壓力相關障礙症的相關提問之回覆的正確性，相關問題將取自衛生福利部台灣e院精神科網站，藉由專科醫師確認ChatGPT回覆的正確性，並進一步將把這些問題分類，建立資料庫。
2) 收集200名患者（憂鬱症、焦慮症，雙極症、創傷及壓力相關障礙症），隨機分為介入組與控制組各100人。個案在基準點及第12週接受認知功能評估，並在基準點、第4週、第8週、第12週接受各項症狀評估（憂鬱症狀、焦慮症狀），健康生活品質量表(Short form-12 items health survey，SF-12)，以及用藥配合度測量表（莫力斯基八問項用藥配合度問卷(8-item Morisky Medication Adherence Scale, MAS-8) ）。控制組維持一般門診追蹤與衛教；介入組在收案基準點將邀請新增一個身心衛教Line聊天機器人(設為好友)，它是用來取代ChatGPT的網頁介面，也就是透過手機慣用的Line作為介面來輸入問題給ChatGPT，如此設計，除了滿足手機使用者的習慣性與方便性外，也方便我們進行側錄，(同時也會記錄詢問者ID、詢問的時間)，然後再加以統計與整理。設計Line機器人能讓個案，隨時提出有關病情之詢問並即時回覆。因為此機器人實際是連接ChatGPT，故個案之問題實際是由ChatGPT來回覆，只是經由Line介面方式傳達給提問患者。在12週追蹤結束後，進行分析，比較介入組與控制組在預後各項評估指標上的差異，同時我們也可能就不同診斷與問題，細部分析相關衛教在症狀上的幫忙。
我們假設，即時提問將有益疾病病程改善、預後、乃至生活品質。由於醫療資源有限，若能藉由現代科技以及ChatGPT生成型人工智慧的協助，建立一個容易取得、快速的線上身心諮詢系統，若能慢慢拓展到各項疾病，將對國家人民的心理健康，有重大助益。
This study aims to evaluate the accuracy and usefulness of ChatGPT in responding to questions related to mental health. Mental health conditions such as depression, anxiety, and mood disorders are often closely interrelated and account for a significant portion of psychiatric outpatient visits. However, due to limited time during outpatient appointments, physicians are unable to provide detailed information about patients' conditions. With the increasing popularity of the Internet and smartphones, people have become accustomed to searching for medical knowledge online.
With the advancement of ChatGPT, artificial intelligence can be utilized to organize and gather data. As a result, when integrated with ChatGPT, it can assist patients in accessing pertinent health education following their consultations. This allows patients to gain a deeper and more comprehensive understanding of their condition or treatment plan.
Currently, the application of ChatGPT in psychiatric treatment remains largely theoretical, and there is a lack of prospective studies or evaluations assessing the accuracy and usefulness of ChatGPT in responding to mental health-related questions. This study will be conducted in two stages:
1. During the three months preceding the project, we will verify the accuracy of ChatGPT's responses to questions related to depression, anxiety, bipolar disorder, trauma, and stress-related disorders. These questions will be sourced from the Taiwan e-Clinic Mental Health website of the Ministry of Health and Welfare. Specialist doctors will confirm the correctness of ChatGPT's responses, and the questions will be further categorized to establish a database.
2. We will recruit 200 patients with depression, anxiety, bipolar disorder, trauma, and stress-related disorders, and randomly divide them into 2 groups (intervention group, N=100; control group, N=100). These patients will undergo baseline and 12-week cognitive function assessments. Additionally, they will be assessed with symptom for depression and anxiety, the Short Form-12 Items Health Survey (SF-12) and the Morisky Medication Adherence Scale (MAS-8) at baseline, week 4, week 8, and week 12. Patients in the control group will be kept with their usual clinical visits. Patients in the intervention group will be enrolled in a Line chatbot for physical and mental health education, which will replace the ChatGPT web interface. This design not only caters to the habits and convenience of mobile phone users but also facilitates side-recordings for analysis purposes. The content of patients' consultations with ChatGPT and its responses (including the inquirer's ID and the time of inquiry) will be recorded and subsequently organized for statistical analysis. The Line chatbot allows patients to ask questions about their condition at any time and receive immediate responses. Since the chatbot is directly connected to ChatGPT, patients' questions are actually answered by ChatGPT and delivered through the Line interface. After the 12-week follow-up period, data analysis will be conducted. The differences in various outcome assessment indicators between the intervention and control groups will be compared. Furthermore, we may conduct a detailed analysis of the impact of relevant education on symptoms, considering different diagnoses and issues.
We assume that the ability to ask immediate questions will be beneficial for disease progression, prognosis, and even overall quality of life. Considering the limited availability of medical resources, leveraging modern technology and ChatGPT generative artificial intelligence to establish an easily accessible and efficient online mental health consultation system holds great potential. If this system can gradually expand to encompass various mental health conditions, it could have a significant positive impact on the mental well-being of the population.
Title: Joint Resource Allocation, User Association, and Power Control for 5G LTE-based Heterogeneous Networks
Abstract: The aim of 5G wireless network to provide Mbps and Gbps data rates to end users is expected to be fulfilled by the advanced technologies such as multi-input multi-output (MIMO), carrier aggregation (CA), inter/intra-cell communication and adaptive modulation and coding techniques, which would be all realized in the Long Term Evolution-Advanced (LTE-A) heterogeneous network constituted by macrocells (MCs) and small cells (SCs) adopting these 5G advanced techniques. Given the potential of significantly increasing the network performance, the resource allocation (RA) problem involved becomes harder than ever especially when MIMO and CA are included in the complex RA problem involving multiple types of resources to be concurrently allocated for the global optimization. Facing this challenge, we develop an framework to jointly optimize the energy efficiency (EE), spectrum efficiency (SE) and queuing length for a downlink transmission with an overall and comprehensive consideration of dynamically allocating user equipments (UEs), resource blocks (RBs), component carriers (CCs), modulation and coding schemes (MCSs), and deciding user association (UA) with a power control (PC) mechanism on discrete power levels (PLs) in the heterogeneous LTE-based MIMO wireless network. Specially, for the complex RA, UA, and PC problems involved, we first conduct a mixed integer programming model to accommodate the stochastic optimization problem involved that is nonpolynomial (NP) in general. We reveal that the reduced problem could be resolved easily through linear relaxation when its coefficient matrix is totally unimodular (TUM), and show that it could be resolved efficiently as well even when the TUM property is not guaranteed through the experiments. Based on the reduced problem, we further develop distributed or semidistributed algorithms operated on two levels to approach the optimal result with lower computational complexity if the UA requirement can be relaxed. Finally, apart from the exhibition of solving the multi-resource allocation (MRA) problem with the linear programming model, the numerical evaluation show also that our approach can make a good tradeoff among SE, EE, and queue length, outperforming the greedy-based state-of-the-art algorithms.
Title: Use Machine Learning to Predict Bacteremia in Febrile Children Presented to Emergency Department
Abstract:
Background
Blood culture is frequently used to detect bacteremia in febrile children. However, a high rate of negative or false-positive blood culture result is common in febrile children visiting pediatric emergency department (PED). Aim of this study is to use machine learning to build a model to predict bacteremia in febrile children.
Methods
We conducted a retrospective case-control study of febrile children presented to PED from 2008 to 2015. The case group included febrile children with true bacteremia. Each patient was randomly matched with 10 febrile children without bacteremia as control group according to gender and age. Machine learning methods (logistic regression and support vector machines) were used to establish a predictive model of bacteremia. Cost-sensitive learning was applied to increase the usability of the predictive model. Patient’s gender, age, and laboratory tests were used as the predictor variables.
Results
Sixteen thousand, nine hundred and sixty-seven (16,967) febrile children with blood culture test were enrolled during the 8-year study period. Only 146 febrile children had true bacteremia and more than 99% febrile children had a contaminant or negative blood culture result. Even in the patients younger than 3 years of age, the true bacteremia rate is as low as 1.58%. Serum C-reactive protein is the most common different parameter between children with and without bacteremia either in all study population or in the younger children (less than 3 years of age). With cost-sensitive learning the maximum area under curve (AUC) of logistic regression and support vector machines to predict bacteremia are 0.768 and 0.832 respectively. Using the predictive model, we can category febrile children by the risk value into 5 classes. Class 5 has the highest probability of having bacteremia while class 1 has no risk.
Conclusions
Obtaining blood cultures in febrile children at PED rarely identifies a causative pathogen. Prediction models help physician determine whether patients have bacteremia and may reduce healthcare resources.
以NFT實現無第三方介入的同儕匿名安全電子交易設計
(The design of NFT-based peer-to-peer anonymous secure electronic transactions without third-party intermediation)
本計畫在設計同儕網路電子交易平台，實現無第三方介入之「去中心化」「匿名」的「安全」電子交易。同時，所有「交易記錄」必須被完整記錄並以公開資訊形式，提供給任何人查詢，且絕無竄改之疑慮，再者，這裡的「安全」電子交易是指除了買賣雙方必須「銀貨兩訖」外，還必須要確保交易過程中買方所應擁有的商品控制權不可外洩。此研究的前提假設是假設買賣雙方彼此及平台三方是互不信任。研究方法是採用區塊鏈的智能合約技術。NFT (Non-Fungible Token，非同質化代幣)是基於區塊鏈(blockchain)技術的一種虛擬代幣，目前NFT的數位版權應用是讓一枚NFT來連結一件數位商品，做為此數位商品的所有權移轉記錄證明，一旦能夠建立數位所有權人，任何數位內容的交易就能夠實現來進行。NFT標記著所連結的數位商品目前的所有人，以及過去所有權移轉的歷史交易紀錄。目前的NFT數位版權，其真正的價值在於其後面的所有人的紀錄，反而不是數位內容本身，也就是原始著作檔案並沒有被納入考量。買下NFT的時候，實質買到的是區塊鏈上的智能合約(smart contract)。合約中包含一組後設資料（metadata），這份資料含有作品名稱、作品描述以及標示作品所在的URI。這樣的NFT應用設計，與實際的電子交易需求大相逕庭。本計畫基於區塊鏈的匿名特性(使用者僅以公鑰hash過的「地址」(Address)做為識別)與智能合約技術，建置一個無需第三方介入的匿名安全電子交易平台，且數位商品不再只是圖畫與影音資料，將擴及任何數位內容(例如文件類：商業情報、產業分析研究報告、電子書等)。交易流程協定設計必須確保：貨品必須以買方電子錢包公鑰加密後交付，以確保物品本身的控制權不致於外洩，買賣雙方Addresses及物品資訊會隨著智能合約上鏈公開。基於買賣雙方及對交易平台不信任的前提下，解密的私鑰不可能交付他人，如何在無第三方介入下完成同儕直接交易，需要提供保密計算(confidential computing)機制，是平台重要的關鍵模組，也是本計畫的重點之一(因為不信任，所以買方無法提供私鑰來呼叫使用賣方或平台所提供之商品解密程式)。
本研究將採用功能較多的ERC 1155協定來打造交易智能合約，而非NFT常用的ERC 721。設計區塊鏈的應用，如何最小化上鏈費用(gas fee)是其中最重要的一環。目前數位版權應用，是每件商品必須綁定一枚NFT，做為商品所有權人註記之用，不同於本應用的電子交易情境，所以我們NFT是與賣家Address做綁定，NFT由平台智能合約統一大量鑄造(mint)備用，然後每一枚NFT只會做唯一的一次移轉給某個賣家，其用途是做為交易時的商品所在網址URI(密文，以買方公鑰加密)的紀錄之用，此設計讓賣家能用同一枚NFT重複使用於不同交易，因為它始終只綁定特定的賣家Address，所以不會有NFT ownership變更所造成的高額gas fee費用。賣家可視交易熱絡狀況需要，自行鑄造或向平台購入多枚NFT備用。交易的智能合約上鏈，買方可取得加密的商品檔案的URI來解密，因此可確保交易安全性，即物品控制權只會在買方手中(物品檔案也經買方公鑰加密，雙重加密更為安全)。以上NFT重複使用的設計，可以讓上鏈費最小化。除上述降低上鏈費外，ERC1155智能合約初始化部署時，可一次鑄造大量NFT備用，可進一步減少平均鑄造成本。此外，也可考慮於初始化時鑄造FT (Fungible Token)，來做為平台之專用代幣，取代以乙太幣來交易。
黃恆志(Brian Huang)
黃志銘(Andrew Huang)
陳世斌(Benny Chen)
張桓(Huan Zhang)
蔡智閩
洪詮盛
鄭又誠
王志文
余承翰
林彥智
王嘉慶
賴虹竹
陳炳華
謝宜甄
黃鐘萱
洪晟緯
蘇明鴻
張傑翔
曾筠惠
劉晏誠
張正昕
楊哲嘉
林品辰
劉益嘉
林宏澤
賴彥廷
黃彥欽
劉建源 ～ 2005
王國淵 ～ 2006
夏銘君 ～ 2007
黃億祥 ～ 2008
廖宏仁 ～ 2012
董光遠 ～ 2012
溫峻皓 ～ 2013
郭大維 ～ 2014
曾冠樺(Ethan Tseng) - 2020
邱義閔(Ray Chiu長庚醫師) 2019/9 ~2023/06
鄭吉詠(Jill Cheng長庚醫師) 2019/9 ~2023/06
黃玉成，鄭慶三，蔡欣懌 1996~1998
王凱明，趙尚威 1997~1999
江振在，郭偉程，劉忠青，林孟岳 1998~2000
江天賜，高啟洋，江存賢 1999~2001
楊順興，李忠來，吳建偉，黃億祥 2000~2002
夏銘君，林修緯，李文姍，張淑萍 2001~2003
張原豪，蕭鈞庭，董光遠，林慶豐 2002~2004
李俊嶧，王復民，喻浦軒，張萬餘，江建豪，黃梓城 2003~2005
徐紳益，郭以謙，梁家銘，許育偉 2004~2006
陳志欣 2005~2006
侯承志，高立功，林承諺，蔡振宇 2005~2007
毛彥凱，黃則霖，黃敦麟，郭家懿，薛懷宗，施智維 2006~2008
何思嫻，劉哲宇，吳承陽，葉柏廷，許文耀 2007~2009
蘇峻晟，吳世麟，吳旻熾，陳致毅，張書榜，徐偉致，鄭聖倫，紀裕信，楊育瑋，劉祐玟 2008~2010
王瑞陞，朱勁豪，洪健恆，蔡慶堂，陳誌翔，鄭彥暉，詹博超 2009~2011
林容瑋，黃珣，徐啟昕，邱敏鈴，廖榆恬，邱文，林平鑫 2010~2012
曾千瑜，鍾易軒，林承毅，游瑞穎，陳松毅，李昂珍，蔡昇宇 2011~2013
蔣朝勛，吳冠賢，林世宏，鄭又新，蕭銘傳，蔡政修，蔡順昶 2012~2014
黃子華 ，金翼，莊順賢，劉淳楷，謝宗霖 2013-2015
朱俊瑋 ，陳柔尹，江韋辰，張家豪，李茂耀，王耀頡 2014-2016
李宇哲，蒲宏易，歐陽劭宇 2015-2017
王躍翰，謝昀婷，連哲芙 2016-2019
張桓，余承翰，劉宇軒 2017-2019
李智傑，黃承鴻，曾文輝 2018-2020
吳典橋，張柏凱，林澤丞，黃貞豪，廖允澤，潘建廷，許茗雅 2019-2021
莊凱威，鄭鈺城，朱育承，陳立超，葉明蓁，李昱賢，黃柏健，林佳誼 2020-2022
陳守倫，洪宗啟，黃立至，王昱翔，周德昀，王楷翔，許耀元 2021-2023