BML can help enterprises and developers to realize various business scenarios such as image recognition, video analysis, voice recognition, recommendation and prediction and empower many industries such as finance, education, medicine, security, retail and industrial manufacturing.

Typical customer case:

Internet lending risk control

Establish a risk control model for banks’ Internet lending business to predict whether the Internet lending applicant will breach the contract or not. Based on the data in the bank, the sample size is 60,000, the ratio of positive and negative samples is 10:1, and there are more than 50 feature fields. With the GBDT algorithm, the accuracy rate and recall rate of the Internet lending risk control admission model have been greatly improved. BML facilitates the intelligence of the finance industry, builds a risk prevention and control system based on artificial intelligence and big data, controls business risks, improves capital returns and actively adapts to the development trend of new finance.

Precision marketing

Predicting potential customers of financial products for precision marketing is a typical classification problem. When the ratio of positive and negative samples is extremely unbalanced (1:5000), you can sue the GBDT algorithm and AutoML parameters for automatic tuning, let the models to learn the optimal parametric solutions, and achieve a high accuracy rate on the test set. BML uses artificial intelligence technology to support the rapid implementation of various innovative businesses in Internet finance and assists companies to quickly enter the market segment with high-efficient and personalized financial product, to expand the asset scale.

Industrial routing inspection

Have regular inspections of equipment (such as transmission lines) to ensure the safe production of power systems. Since the lines pass through various areas, some areas have bad environments and it’s difficult for staff to access to it, routing inspection by unmanned aerial vehicles is adopted. The unmanned aerial vehicle flies according to the predetermined satellite positioning route, shoots transmission lines and returns images to the backend. Through recognition and analysis of the images by BML, suspicious failures can be discovered and then reported to the staff for further judgment and analysis. Quality of the photos is judged, and photos of low quality are filtered out. Algorithms of object detection and segmentation are then applied to detect components such as wires and insulators and other parts where problems may occur to position and alarm. During training, image data enhancement and parameter optimization are used to improve the training effect.

Information extraction

Extract key elements required by customers from credit granting proposals, such as company name, credit limit, credit validity period and credit type, for subsequent business approval and handling. It is a typical series annotation matter in NLP. With the ERNIE pre-training model based on Baidu data training, the accuracy of recognizing various information has reached a top level, which greatly improves the efficiency of business review.