Development of Oil Collecting Submarine using AI and hydrophobic solution
Such as the plastic waste and industrial discharge that permeate our oceans, it is the insidious and infamous nature of oil spills that demands our immediate attention. These spills, with their far-reaching ecological ramifications, pose a profound danger to our marine ecosystems, demanding urgent action and a heightened awareness of the true menace that is caused by this oil
EIPCA : Electrocardiogram Interpretation Pattern for Cardiovascular Abnormalities Prediction
Cardiac Arrhythmia is one of the conditions in the group of heart and blood vessel diseases that can lead to sudden cardiac arrest (sudden death) and other conditions if not diagnosed quickly and accurately. According to research, heart and blood vessel diseases are the most common diseases and have a mortality rate of one-half of all non-communicable diseases. According to WHO statistics in 2012, it was found that there were 7.4 million deaths from heart and blood vessel diseases, and in 2017, the number of deaths increased to 177 million people, or about 94,444 people per day. Diagnosis of heart and blood vessel diseases can be done by measuring the electrical activity of the heart, and after the examination, a specialized physician will read and analyze the graph to find abnormal patterns. Currently, the shortage of qualified heart specialists to read the graph and screen for heart disease is a medical position shortage, which requires transferring data to hospitals with specialists, resulting in delays in diagnosis and treatment and even death. The project "EIPCA: Electrocardiogram Interpretation Pattern for Cardiovascular Abnormalities prediction" is an application program that assists in screening for fatal diseases that arise from abnormal heart rhythm. It employs artificial intelligence to aid in the screening and analysis of the electrical waveforms generated by an ECG machine, thus reducing diagnosis time and addressing the shortage of cardiology experts. EIPCA is comprised of two systems: (1) a system for screening and analyzing ECG waveforms using artificial intelligence to solve the problem of a shortage of specialized cardiology physicians, and (2) a system for risk assessment of fatal diseases by analyzing the ECG waveform data. The target group of the project is Rural hospitals, as well as health-related agencies. The project team hopes that the development of this project will significantly improve the efficiency and speed of screening for heart-related diseases, ultimately reducing the mortality rate from these diseases in the future.
Instruction-Tuning 在法律對話模型上的影響之探討
本研究探討 Instruction-Tuning 對法律領域語言模型的影響,我們使用 ChatGLM-2 6B 作為基礎模型,先以台灣法律文本進行 Continual Pre-training,再以和律師的 Q&A 數據集,分別採用 Supervised Fine-Tuning(SFT)、Reward Model 及 Proximal Policy Optimization(PPO)等 Instruction-Tuning 方法進行微調。結果顯示,僅經過 Pre-training 及 SFT 的模型,其產生的回覆較符合法律專業風格;但考量模型對法律知識的掌握,則以 Pre-training、SFT 及 PPO 整套 Instruction-Tuning 的結合效果最佳。本研究證明, 針對單一領域的語言模型, 不同的 Instruction-Tuning 方式會對其回覆風格及知識掌握造成不同影響。我們的研究為未來單一領域語言模型訓練提供了參考。