Baidu has announced the product launch of EZDL, a no-code platform to build custom machine learning models. The company informed that the solution is geared towards small and medium businesses. “It is gradually breaking down the barrier to allow everyone to access AI in the most convenient and equitable way,” said the company.
“EZDL is a service platform that allows users to build custom machine learning models with a drag-and-drop interface. It takes only four steps to train a deep learning model, built specifically for your unique business needs,” said Yongkang Xie, Tech Lead of Baidu EZDL .
“Even if you have had no exposure to programming, you can quickly build models on this platform with zero barriers. EZDL can help companies with limited AI experts and computing resources to quickly and efficiently complete deep learning training and deployment with only a small amount of data,” Xie added.
“Through our Global Business Unit, we are devoted to sharing Baidu's core technologies, capabilities, and resources with the global community. We seek to create a true ecosystem for AI, democratizing access to AI capabilities,” said Youping Yu, General Manager of AI Ecosystem Division at Baidu.
Yu added “Baidu officially released Baidu Brain 1.0 in September 2016, and just two years later, in July 2018, we released Baidu Brain 3.0 — encompassing a full AI “tech stack” from chips to deep learning frameworks to platforms. Through our open AI platform, Baidu provides developers with a number of leading AI capabilities such as speech, vision, NLP, and AR/VR. Baidu's open AI platform is currently the most comprehensive, open, and dynamic in China.”
Yu Informed “To date, more than 600,000 developers are using the Baidu open AI platform and Baidu's AI capabilities have been applied in more than 20 industries. Technology has no borders. We've seen in China how EZDL has lowered the barrier for small and medium businesses to use AI to solve their business problems. We are excited to see innovative use cases from the global community. It is our hope that by opening up the platform and democratizing AI capabilities, we will help developers and businesses around the world achieve greater success.”
The company said that EZDL has four characteristics: it is easy-to-use, fast, high-performance, and highly secure. It is extremely friendly to entry-level AI developers. EZDL has opened up capabilities in image classification, object detection, and sound classification.
Image classification model
Quick classification of home decorating images – Home decorating website Idcool uses EZDL to train custom image classification models that automatically identify the design style and room type from massive home improvement pictures. Specifically, the models are able to accurately classify 13 design styles and 15 room types. The accuracy rate is about 90%. Idcool saw a 50% increase in image utilization and a 30% lift in click-through-rate from their recommendation feature based on image classification.
Image recognition of Chinese herbal medicine, wild birds and frogs – A scientific research institution uses EZDL to train image classification models in vertical domains. The models can identify 42 kinds of Chinese herbal medicine, 500 kinds of wild birds and three kinds of rare frogs. The average accuracy rate is over 92%. It effectively helps spread awareness of scientific discoveries.
Industrial quality inspection to identify defective products – A keyboard manufacturing company uses EZDL to train an image recognition model with only hundreds of sample pictures, and was able to build a system which automatically detects normal, misplaced and missing assembly parts with an accuracy rate of more than 95%, which greatly reduced labor costs.
Pavement damage monitoring – A pavement management service company uses EZDL to train image classification and recognition models based on 300 sample images. The models are deployed to monitor road pavement conditions in real time and detect the risk of road distress, such as cracks, with an accuracy rate of over 88%.
Object detection model
Retail inspection – A retail company uses EZDL to train a product inspection model that automatically detects whether the items on supermarket shelves are displayed according to the requirements of the brand directly from a user's mobile phone. The model operates with 93% accuracy, which significantly improves the inspection efficiency and reduces the risk of cheating.
Medical detection for cell counting – a medical institution uses EZDL to build object detection models for microscope imagery from blood tests. These detection models help automate and improve the identification and statistics of test results, with accuracy rate of more than 95%. EZDL replaces the need to count by hand and greatly improves efficiency.
Sound classification model
Security Monitoring – Real time danger detection – This model helps to recognize the sounds from indoors vs. outdoors and abnormal sounds vs. normal sounds. These models are used for emergency warnings and real-time detection of dangerous situations.
Scientific Research – Voice recognition of different species and individuals – this model is used for field research; it detects the sounds of different individuals and can detect if they are of the same species or from different ones.