Blog: VisionX Progress Report #8
- Marketing & Collaboration
(1) The VisionX team is working to update the VNX white paper and the road map in collaboration. The updated version will enhance the blockchain technology and VNX based token economy.
(2) VisionX launched the DAPP development team and has begun to design and develop a distributed AI training framework in cooperation with DBC and Asus research team. The innovative AI training framework is primarily based on VNX federated learning framework for visual inspection. VNX will extend to a distributed AI solution framework for general tasks. VNX will also open the source codes to the VNX community and invite users and developers to use and enhance the framework.
2. Tech Development
(1) Distributed AI Framework:
● VNX created a new team to design and develop a novel AI training framework, based on decentralized DBC AI cloud.
● The initial prototype design has been finished, mainly based on the Federated Learning (FL) framework for visual inspection.
(2) Defect Detection (Dataonomy):
● In order to protect raw data of customers (especially for enterprise users) and encourage data contribution, the visual inspection team has successfully developed feature extraction and distributed training technology in the Federated Learning (FL) framework. The useful features from the raw data will be extracted in the client side before contributing to the cloud server. The base models trained from the features can reach the same performance as from raw data.
The above figure shows raw images of the texture defect samples (shown in the 1st column of the plot) and its extracted features from 4 different feature channels of the third convolutional layer of the pre-trained Inception V3 model.
● VNX has developed the initial version of AI module to integrate into Hubrain’s Hu-Dra visual inspection system and confirmed the detection performance. Hu-Dra is a general-purpose visual inspection system.
● With the request of the LED light chip vendor, VNX has developed the defect detection AI model with more than 99.9% accuracy and less than 5ms for each detection.
(3) Visual Anomaly Inspection:
● With the certified pieces of equipment from CCTEG, VNX issues the first AI solution for safety monitoring and regulation enforcement in coal mining sites.
● Continued to improve the product auto-sorting system for a medical implant supplier and enhanced multiple object-identifying software via a new AI model.
● Optimized the accuracy and speed in OCR for product series number recognition.
(4) AI-based prediction and preventive maintenance:
● Environment surveillance and warning: obtained and cleaned 4 years of gas and gas associated data from more than 100 underground coal mining sites; started to develop the prediction model of the gas index and explore the association with other features.
● Smart AI finance prediction: developed stock price prediction AI model, which can predict the closing price of the PS500 index with an accuracy of 97%. The AI model is using long and short-term memory (LSTM) of finance timeline and Deep Learning framework, which will provide smart financial investment advice in the near future.
● Bio-health big data: designed the Multi-Task Deep Neutral-Network (MT-DNN) for multiple labeling task in order to improve individualized breast cancer patient’s treatment plan
VisionX, industrial AI chain, is a cross-industry high-performance AI collaboration platform, and the first AI+Blockchain project incubated by the DeepBrain Chain Silicon Valley lab. It has the single largest and quality image dataset and uses a unique taxonomy algorithm to offer a cross-industry AI solution fit for all. VisionX uses product surface defect visual detection as a breaking point. Then it will gradually branch off to machine arm grab and pick, preventative maintenance, intelligent speech, supply chain, and other fields. In the future, it will serve thousands of industries offering AI solutions and data worth millions.