AI deep compression toolchain and Hybrid-fixed point CNN accelerator
Summary | Assisted by in-house AI deep compression toolchain (ezLabel, ezModel, ezQUANT, ezHybrid-M), the proposed technology supports automatic AI model design and optimization with the integrated performance of 120x model size reduction and 70x power reduction in 2D CNN model, and develops a world-first 1/2/4/8-bit CNN model realized by the developed high efficiency Hybrid fixed point CNN NPU (Hybrid-NPU), which has been verified in Xilinx ZCU102 FPGA and achieves the performance up to 2.5 TOPS(8-b)/ 20TOPS(1-b)@28nm technology running at 550MHz and 4TOPS/W energy efficiency. The proposed technology attracted Wistron to launch a four-year investment with annual amount of 10M NTD to setup Wistron-NCTU embedded artificial intelligence research center in NCTU. At the same time, the proposed technology developed in our AI project also results in a total amount of 73M NTD investment from local industry. A new start-up is under cultivation to attract the Angel round investment up to 100M-180M NTD and launch by the end of 2021. The proposed technology attracted Wistron to launch a four-year investment with annual amount of 10M NTD to setup Wistron-NCTU embedded artificial intelligence research center in NCTU. At the same time, the proposed technology developed in our AI project also results in a total amount of 73M NTD investment from local industry. A new start-up is under cultivation to attract the Angel round investment up to 100M-180M NTD and launch by the end of 2021. |
||
---|---|---|---|
Technical Film | |||
Keyword | embedded AI ADAS Self-driving system AI chip Interdisciplinary integration |
More like this
Provide the latest information of AI research centers and applied industries
-
iSleepBetter Corporation
Intelligent Portable Neurofeedback Training System: Highlights of our product: 1. High accurate EEG value (with rigorous hardware and software design and verification) 2. real-time feedback by computer screen or APP 3. can be used at home (easy to ware, easy to use)
-
DeepFIND
Artificial Intelligence Automatic Sputum Microscopy System AIASMS: AIASMS can automatically shoot images from sputum smear samples and identify Mycobacterium tuberculosis(MTB). AIASMS can find the most suitable depth for shooting images and then automatically shoot images under the microscope through auto-focusing. These images will be stored in the cloud database. Meanwhile, AIASMS detect whether the sputum smear contains MTB. AIASMS can accurately label the location of the TB. The sensitivity and specificity can reach 90% and 99% respectively. We also cooperate with three hospitals in the northern, central, and southern regions of Taiwan to conduct blind testing, and the sensitivity can be more than 90%. AIASMS can also distinguish between MTB and Non-tuberculous mycobacteria (NTM). Finally, n the expansion of technology, we have extended the identification of the acid-resistant staining to the identification of gram staining, so that our system can widely identify a variety of bacteria.
-
ALOVAS
ALOVAS Platform: ALOVAS acts as an A.I. Pathology Platform which provides high resolution pathology image viewer. Even Giga-pixel-level original images can be viewed online in real time. ALOVAS platform can be used not only on computer, but also on iPad. Users can upload images on the platform and select AI models for automated detection, and browse the detection results on the platform. ALOVAS also provides commonly used annotation tools, including hand-drawing, dots, rectangles, etc., which can be used to mark areas of interest. Embedded with the ALOVAS platform also provided several detection algorithms. We hope ALOVAS can assist physicians in rapid diagnosis, in related pathological research, and reduce the workload of pathologists in the future.
-
alpha pulse
ECG STEMI AI Model: In the past, most AI systems gave people the feeling of a black box and couldn't be trusted. The team designed a mechanism that allows doctors to adjust and observe the AI model, so that the AI model can be customized to the functions the doctor wants. We use LINE, the most commonly used communication software for doctors, to design an EKG Line Bot. Medical staff can upload an electrocardiogram to the EKG Line Bot to instantly identify whether the electrocardiogram is Stemi, so as to help doctors determine whether the patient has signs of myocardial infarction. We use this Line Bot to cooperate with doctors and ask them to communicate with the Line Bot. According to the heat map provided by the system, we can check whether it is consistent with the medical concept, and then help us correct the accuracy of our model. The system will train the correct data again.
-
NTU CSIE Medical Informatics Lab
NTU Medical Genie Precision Health Platform: The product is mainly composed of wearable devices, IoT environmental sensors, deep learning, personal health app and case management platform. It can collect and monitor user's lifestyle and environment automatically, and predict the possibility of emergency to assist medical staff in making decisions. In addition, we opened source the project to solve most clinical studies that require lots of time to build data collection tools and processes.
-
Digital Medicine Center of National Yang Ming Chiao Tung University
AI-based Brain Assessment System: Mental illness is a critical health issue in Taiwan. Psychiatric diagnosis is largely based on self-reported or symptomatic criteria without objective findings. This state-of-the-art psychiatric diagnostic platform using standardized brain imaging data, however, improves the efficiency and accuracy of psychiatric diagnosis. Our platform can easily identify the deficiency in brain regions associated with schizophrenia, provides a novel way to evaluate mental illness and its progression, with powerful visualization. This web-based diagnostic platform has international multicenter validation, scientific publication, patent-pending as well as awarded by the 17th National Innovation Award in 2020, etc. This evidence-based, AI-assisted psychiatric diagnosis platform, validated with large-scale standardized brain imaging. We believe this state-of-the-art diagnostic tool can be a new light on modern psychiatric medicine, and promote mental health in the general population.
-
An intelligent decision support system: prediction of hospitalization and length of stay, and similar medical records retrieval for proper patients arrangement in an emergency department.
The core concept of the decision support system is to establish machine learning-based predictive models and enable their interpretability to support physicians making clinical decisions in practice.
-
The application of smart watch monitoring to construct an overwork prediction and alarm model for emergency healthcare professionals
By collecting the changes of kinds of vital signs(including heart beat rate, blood pressure, steps, and etc.), we combined the results with the testers’ fatigue scores which belonged to the matched fatigue types and their basic profiles. We extracted 780 types of structural features sets by cleaning raw data, then separating into training set, validation set and testing set. With machine learning algorithm to auto-optimize regression analysis, finally we come out the correlative features between vital signs and fatigue, and the overwork model as our deliverables.
-
Artificial intelligence dynamics drug repurposing platform and application on COVID-19
e propose an AI dynamics drug repurposing platform applicate on COVID-19. This platform analyzed a large number of structures of SARS-CoV, MERS, and cross-species coronavirus 3CL protease-ligand complexes to construct uncovering six flexible active site conformations and pharmacophore clusters for SARS-CoV-2 3CL protease screening all FDA drugs and found four inhibitors within three months. Among them, JM206 had even demonstrated ten times better efficacy than Remdesivir in-vitro assay and also show the effect on the in-vivo hamster model to alleviate the symptoms caused by COVID-19.
-
MeDA Omni X-Ray (OXR): Real-time Hazard Identification System
“MeDA OXR: Real-time Hazard Recognition System,” which can automatically screen and interpret the medical images in real-time, is designed for emergency rooms to assist physicians in diagnosis, reducing medical risks, and improving overall efficiency. Combining portable X-Ray and AI algorithms, the system performs real-time and accurate preliminary screening and diagnoses of diseases, such as pneumothorax, pneumonia, and tuberculosis. It can also locate the nasogastric tube, endotracheal tube, and central venous catheter, while misplacement of that is sent to the physicians when detected.