Recently, the top conference in the field of human-computer interaction and ubiquitous computing, UbiComp2020 (The ACM International Conference on Pervasive and Ubiquitous Computing), announced the competition results. The AI Mid-Platform Machine Learning Team of Gaosi Education Group stood out from more than 50 outstanding teams worldwide and won the championship with a significant advantage.

Winning the International Champion Again, Showcasing China's AI Competitiveness

The UbiComp competition is an important part of the UbiComp2020 conference. The competition is organized by UbiComp and jointly hosted by the Machine Learning and Data Analysis Laboratory of Erlangen University and the Fraunhofer Institute for Integrated Circuits.

In recent years, the number of participants and teams in UbiComp has grown rapidly, including leading Internet research institutions and companies such as Google, Amazon, Microsoft, Alibaba, and Tencent. This year, more than 50 professional teams from Erlangen-Nuremberg University, University of Duisburg-Essen, and others participated in the competition.

This competition is themed on cutting-edge exploration in the education field, requiring participants to directly use smart pen handwriting trajectory information and recognize corresponding handwriting content without relying on traditional image recognition technology. After two rounds of fierce competition, the AI Mid-Platform Machine Learning Team of Gaosi Education Group eventually won the championship, demonstrating the hard strength of China's educational AI technology to the world.

At present, OCR (Optical Character Recognition) technology is commonly used by most enterprises for text recognition, typically requiring the final result of the writing to be photographed for recognition. However, this process is easily affected by factors such as light, shadows, shooting angles, and clarity, and completely ignores writing process, movements, and habits, resulting in inaccurate recognition results.

During the competition, the Gaosi machine learning team did not rely on OCR technology at all. Instead, they used the E-RTCR model and integrated two cutting-edge deep learning models, R-Transformer and CRNN. They successfully captured the multimodal sensor signals in the smart pen, learned local drastic features and overall trend features in the writing data, thus directly modeling and identifying the writing process, including writing trajectory, angle, speed, acceleration, pauses, strokes, and other aspects.

Compared to directly analyzing images using OCR technology, this method overcomes the extreme abstraction of sensor signals, differences in writing habits among different people, and large differences in the distribution of multi-sensor signals. The Gaosi machine learning team eventually won the championship with scores far exceeding the second place.

Gaosi Machine Learning Team won the championship with scores far exceeding the second place

As people continue to pursue the improvement of educational experiences, hardware devices such as smart pens have gradually begun to be used in teaching scenarios. These devices can retain students' traditional paper writing habits and natural experiences, while ensuring that writing trajectories, angles, and other information are recorded in real time.

The independently developed and industry-leading character recognition technology based on sensor signal time series by the Gaosi machine learning team has laid a good foundation for data analysis, content recognition, and automatic grading in teaching scenarios.

Increase Research Investment, Strengthen Research Capabilities

Since its establishment, Gaosi has always adhered to the mission of "Love and Technology Make Education Better," and has been committed to enabling everyone to have access to fair and high-quality education, continuously increasing investment in scientific research and development. In August 2019, the Ministry of Science and Technology approved the construction of the National New Generation of AI Open Innovation Platform for Smart Education based on Gaosi.

Relying on the National New Generation of AI Open Innovation Platform for Smart Education, Gaosi's AI Mid-Platform continuously strengthens the construction of underlying academic capabilities. While maintaining close scientific research and academic cooperation with outstanding universities at home and abroad, it also showcases China's leading education AI capabilities on the world's highest academic stage in various technology segments.

For example, dozens of academic achievements of the Gaosi AI team have been selected for top international academic conferences such as AAAI, WWW, AIED, NCME, and won the championship in the top international computer vision conference CVPR-EmotioNet competition. At the New York International Artificial Intelligence Conference AAAI, the Gaosi AI team successfully organized the first AI for Education academic seminar, promoting international academic exchanges in the field of educational AI.

In recent years, Gaosi's AI Mid-Platform has continuously made breakthroughs in cutting-edge core technologies and accumulated multiple AI capabilities such as speech technology, visual understanding, and knowledge graph based on productized applications, creating innovative product solutions including AI classrooms, teaching process assessments, oral expression ability assessments, and homework scanning and grading, covering various teaching processes of "teaching, learning, testing, practicing, and assessing."

As of now, Gaosi has more than 100 AI capabilities for the education industry. Among them, half are industry-leading and unique models, with a weekly call volume of over 500 million. Gaosi stated that it will continue to open up its AI+ education practical achievements to the entire industry, empower industry partners comprehensively, and jointly promote the achievement of fair and high-quality education.