Deep Learning

Textual Advertisement Generator: Given any limited specifics of any product, AI Advertisement Producer can automatically generate tons of top-quality descriptions and advertisements for the product in just one second. And not just one copy is produced. With deep learning and natural language processing technologies learned from millions of existing samples, our AI model can produce various styles of advertisements at the same time for users to select. It will be a big helper or a virtual brainstorming partner for any brands or advertisers to create their advertisements.

Recently, Deep Reinforcement Learning (DRL) has been applied to many AI applications. One of the successful achievements is the AlphaZero, called the Zero method in this project, was presented to learn without human knowledgesurprisingly surpass all the human playersall the AI programs.

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.

Based on deep learning camera/radar object detection and tracking technology, the proposed road side unit (RSU) system has achieved over 95％ vehicle detection accuracy within 100m detection range in the processing performance of 10fps under nVidia Jetson Xavier platform. Compared to the 32-beam lidar based RSU, the proposed RSU achieves 97％ reduction of sensor cost that exhibits high competitiveness in deployment cost. The proposed RSU system has been verified in fields and we are now cooperating with an industry partner to deploy the RSU system in both Tainan and Tao-Yuan cities.

Machines need annotations to learn, but human babies learn human languages with almost no annotations. Can machines do the same thing? To allow machines to learn human languages with only observations like human babies, a research team at Taiwan has partnered with the speech research groups in Meta, CMU, MIT, and JHU to develop a brand new self-supervised speech processing evaluation framework, Speech Processing Universal PERformance Benchmark (SUPERB).

PANCREASaver contains a “Pancreas Cancer (PC) automatic segmentation model” (image segmentation) and a “Pancreas Cancer (PC) analysis AI model” (image classification) that can read the DICOM format of postcontrast CT images directly for the automatic analysis process. After conducting prep-processing with image processing algorithms, C2FNAS is employed to illustrate the tumor position prior to the diagnosis conducted by CNN. The results can be provided to the physician for diagnostic reference so as to reduce early omissions and increase the detection rate of pancreatic cancer.

The application of voice assistants is becoming more and more popular, however, due to the inefficiency of artificial intelligence-based technology, current products are mostly built by using rules-based methods. Therefore, in this project, we would like to propose some corresponding solutions for different components of the dialogue system to improve the data efficiency and work efficiency of each component.

Tool wear and health condition monitoring during the processing: The tool wear monitoring technology developed by our researching team is specifically designed to analyze whether the tool is broken, collapsed, etc., and to estimate the remaining useful life(RUL) of the tool according to the working conditions of mass processing. By acquiring the vibration signal data with three-axis accelerometers installed on the machine tool, this technology could determine whether the current tool cutting vibration has exceeded the safety range by plotting a control chart. Once it exceeds the safe range, the current tool processing state will be assumed as abnormal. It gives users a reference to replace the broken tools immediately to prevent continuing processing, which causes vast loss such as poor quality of workpieces. In addition, this technology allows users to build models for distinct working conditions to predict the RUL of tools. It could allow users to evaluate the current health condition of tools and schedule the time to change the tool.