The core concept of the decision support system is to establish machine learning-based predictive modelsenable their interpretability to support physicians making clinical decisions in practice. The developed technique is part of the capstone project named "Smart Emergency Department," sponsored by the Ministry of ScienceTechnology. The system comprises NTUH (National Taiwan University Hospital) EMR (Electronic Medical Record) importance analysis, accurate clinical quality indicator prediction,similar medical record retrieval. It is expected to improve the patients＇ flowalleviate emergency department crowding. It has been verified in retrospective studieswill officially enter the clinical trial phase this year.

We propose an AI dynamics drug repurposing platform applicate on COVID-19. This platform analyzed a large number of structures of SARS-CoV, MERS,cross-species coronavirus 3CL protease-ligand complexes to construct uncovering six flexible active site conformationspharmacophore clusters for SARS-CoV-2 3CL protease screening all FDA drugsfound four inhibitors within three months. Among them, JM206 had even demonstrated ten times better efficacy than Remdesivir in-vitro assayalso show the effect on the in-vivo hamster model to alleviate the symptoms caused by COVID-19.

The IPMC system can detect possible abnormal signs in advancepredict the RUL of the machine before it fails so that it can effectively avoid unscheduled downtimeprevent enormous losses. The IPMC system also exploits the advantages of cloud computingContainer technology such as rapid deployment, failover,lightweight to enhance its performanceportability.

By collecting the changes of kinds of vital signs(including heart beat / blood pressure / steps), we combined the results with the testers’ fatigue scores which belonged to the matched fatigue typestheir basic profiles. We extracted 780 types of structural features sets by cleaning raw data, then separating into training set, validation settesting set. With machine learning algorithm to auto-optimize regression analysis, finally we come out the correlative features between vital signsfatigue ,the overwork model as our deliverables.

"The primary goal of this project is to establish a complete hospital-based liver cancer database, profiles for data feature extraction,develop different cancer, prediction models. A Medical AI program to predict the poster treatment (including operationradiofrequency ablation) recurrence of liver cancer will be established. The program system will assist doctors in the medical decision, identify high-risk patients,adjust clinical follow-up programs."

In agriculture, the team has developed a “Chinese Medicinal Functional Compound Prediction A.I. System” that can predict compounds production during cultivation period. Moreover, in application, we have implemented a smart agricultural platform-AgriTalk to link the “Functional Compound Prediction System”“Smart Fertilization System” to provide a comprehensive management in farm production.

The HeaortaNet is developed by the TW-CVAI team. The HeaortaNet is a deep learning model trained by 70,000 axial images from 200 patients with verified annotations of the pericardiumaorta. It shortens the time for data processing from 60 minutes, by manual segmentation of both pericardiumaorta, to 0.4 seconds. The segmentation accuracy, as assessed by dice similarity coefficient, is 94.8 for the pericardium,91.6 for the aorta. The imaging-based Cardiovascular Risk Prediction module was constructed by analyzing data from the National Health Insurance Databank.

In the context of FinTech, we present Financial Graph Attention Networks (FinGAT) to recommend high-profitable stocks in terms of return ratio using time series of stock pricessector info. Our FinGAT can learn the long-short-term price tendency,model the latent interactionsinfluence between stockssectors without any hand-crafted effort. Experiments conducted on Taiwan Stock, S&P 500,NASDAQ stock markets exhibit remarkable accuracy of FinGAT, comparing to state-of-the-arts (by 11, 14,12 performance improvement). The tech is published in IEEE TKDE 2021.

1. CNN model architecture optimization techniques: Propose an Agilev3-L architecture, with a performance index of mAP@50:95FPS:FP16 up to 171.79 2. GOP-mode acceleration scheme for real time inference: employ GOP-modeTRACKING algorithms to enhance the processing rate. 3. Propose a rapid data labeling system HiTag.

Integrating deep learning, 3D information analysis, hyperspectral analysis, computer vision analysiscombined the multi-source images to develop fruit quality monitoring techniques, including: planting area monitoring, position monitoring, yield monitoring, harvesting time prediction, fruit maturity measurement, fruit quality testing, to achieve the goal of enhancing industrial value.

Through the technique, SamplingPrediction of Lithography Overlay Errors, the costtime of overlay error measurement can be reduced to improve the process efficiency. We identify key sampling through clusteringmachine learning models,design a new sampling algorithm in photolithography process. Due to the complexity of wafermany training factors of wafer data, we combine the clustering algorithm with incremental learning to meet customers＇ unique needsachieve the goal of optimally samplingreducing the costs.

The proposed technologies apply deep learning methodsbig clinical data for lung cancer decision support. The system consists of: 1) CT radiogenomicspatho-genomics for automatic cancer detectionprediction of EGFR mutation 2) demographics to predict survival 3) genomics to predict cancer recurrencemetastasis,4) drug response inference for the best target therapy option. The modules are constructed on an AI-based Clinical Decision Support System (CDSS-SDM). The system is expected to provide physicianspatients with personalized medication recommendations.

This technology aims to automatically extractdefine features from a data-driven perspective through a series of data engineering, machine learningensemble learning technologies via the big data collected from equipment sensorsquality characteristic measurement results. For products that have no quality measurement, a virtual metrology resultpredicted confidence score are provided as a decision-making basis for process control.

This technology＇s nanofluid/ultrasonic atomization micro-lubrication system is different from the general micro-lubrication technology that uses high-pressure air atomization. It is first to use cavitation characteristics to disperse a small amount of nanofluid into microscopic particleseffectively disperse the bondingagglomeration of nanoparticles due to the Van der Waals force. The second stage of atomization is carried out through high-pressure air. The oil mist particles of the nanofluid can easily enter the micro-machining cutting area for coolinglubrication.

PCA-BM includes two models: “Automatic BMs Segmentation AI Model”" Distant Brain Failure Prediction Model". The former one uses C2FNAS (coarse to fine network architecture search) to detect the location, size,number of brain metastases. The latter uses radiomics to extract numerous radiographic featuresemploys machine learning methods such as XGBoost to establish a prognostic model of brain metastases. PCA-BM provides more precision treatment decisions for patientsimproves personalizedaccurate overall stereotactic radiosurgery planning.

The technology is video-based blood pressure monitoring which will help people know the continuous BP status in real time. Also, the contactless measurement can make a daily health record. We have conducted clinical trials in hospitalsestablished a large database to optimize AI solution. The system joins the physicians＇ specialty to analyze the users＇ health status with AI,takes care of human life with technology.

1.Low-Power Processing-in-sensor CMOS Image Sensor 2.Implementation of Low-PowerLow-Latency Deep Learning Chip based on Neuromorphic Intelligence (collaborated with MOST 108-2622-8-007-009) 3.Development of Neuromorphic Chip based on The Fruit Fly VisualSpatial Sensory Systems 4.Hardware-Software Co-Design for Neuromorphic AI Chips