"Squirrel AI Shines at IROS 2019 Conference, AI + Education Era Approaching"

2019-12-18

Yesterday, the 32nd IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS 2019) was held in Macau. IROS is the flagship international conference on robotics technology and intelligent systems, and it is one of the top international conferences in the field of intelligent robotics technology and automation, as well as the most influential international top-level academic conference on robotics in the world.


Dr. Cui Wei, Chief Scientist of Squirrel AI, was invited to attend the IROS 2019 conference and delivered a wonderful speech. Squirrel AI Learning is the first artificial intelligence company in China to apply AI adaptive learning technology in the K12 education field. It has successfully developed the first AI adaptive learning engine with China as its core, complete independent intellectual property rights, and advanced algorithms.


At the technical level, Squirrel AI fully utilizes more than ten artificial intelligence algorithm technologies such as knowledge space theory, Bayesian theory, logistic regression, genetic algorithm, and deep learning, creating nano-level knowledge point decomposition and MCM decomposition system (thinking model, ability model, and methodology), thus clearly and accurately understanding students' knowledge gaps.


In the past, students often had to do a lot of homework to eliminate blind spots in knowledge points. Squirrel AI Learning scans students' knowledge maps to accurately locate their knowledge blind spots. Students do not need to consolidate their knowledge in a massive amount of test questions. The system will quickly help them based on their weak areas of knowledge, maximize their learning efficiency, and ensure steady improvement in their learning performance.


Dr. Cui Wei introduced that currently, Squirrel AI Learning mainly applies several major AI technologies:


First, evolutionary algorithms, logistic regression, and neural networks. The algorithm model takes into account students' learning goals and knowledge status, and dynamically adjusts the learning path. The system gradually draws students' learning habits, interests, methods, and other multi-dimensional student portraits, and continuously optimizes the logic of content recommendation.


Second, machine learning, deep learning, and natural language processing technologies. This technology can recommend the most suitable learning content based on different students' personal preferences, learning habits, and learning styles. Different students use natural language processing technology to automatically generate learning content tags to adapt to different learning atmospheres and difficulties; use deep learning technology to analyze students' learning content and automatically select learning content suitable for students.


Third, Bayesian theory and Bayesian networks. Dynamically view problems based on experience and information. The system will draw knowledge points and probability distributions and predict students' learning abilities. By comparing the relevance between different knowledge points and students' mastery levels, the Squirrel AI intelligent adaptive learning system can infer the relevance between knowledge points without logical relationships.


Fourth, graph theory, knowledge space theory, and information entropy theory. Squirrel AI Learning distinguishes knowledge points based on difficulty, importance, and cognitive level, models the knowledge system, constructs a "knowledge map," and clarifies the logic and cognitive relevance between knowledge points.


From the perspective of measurement, information can be quantified. By using information entropy theory, it is possible to quickly reach the level of students' knowledge state by detecting some important knowledge points, and then calculate around this basic level for repeated refinement, in order to effectively and accurately diagnose students' knowledge gaps and states.


Fifth, knowledge tracing theory. The system evaluates students' proficiency levels in each knowledge point, sequentially analyzes the proficiency levels of knowledge points and related knowledge points, and finally accurately masters each nano-level knowledge point and updates students' ability value changes after learning, in order to accurately recommend the most suitable learning path and learning content for the current student situation. It can not only understand students' mastery of current knowledge points but also reflect students' potential, which is a kind of prediction.


Sixth, educational data mining and learning analytics technology. Educational data mining refers to the quantitative analysis of the learning process and learning behavior, and collects students' learning data, including learning time, stay time, and knowledge accuracy, during the learning process. Different students establish learning models through data processing and analysis. Learning analytics technology is mainly used to predict and monitor students' exam scores, providing detailed student data and information for system and teacher to improve teaching methods.


Seventh, under research, dialogue-based dialogue-driven HUI and VPA engine-based user interaction can achieve real-time voice interaction between virtual teachers and students. The main technologies are natural language processing, speech recognition, and semantic analysis. During the learning process, students can ask their virtual teachers about their learning situation and learning tasks at any time and provide feedback on questions.


Eighth, MIBA is another project in cooperation with SRI and CMU. MIBA refers to multimodal integrated behavior analysis. It collects students' physiological data and behavioral data, including facial expression data, blood changes under the skin, body movement data, and brainwave data, through cameras, measuring rings' brainwaves, and other devices. Combined with learning data generated during the learning process, it analyzes students' learning status, including their level of concentration and learning engagement. The teacher terminal system used by teachers can receive warning signals and implement timely personalized interventions to make learning more effective.


Finally, Dr. Cui Wei stated that traditional education currently faces many deep-rooted problems: the uneven development of educational resources, with high-quality educational resources tending to be concentrated in developed areas, while in underdeveloped areas, there is an imbalance in teacher quality. Squirrel AI Learning hopes to realize the educational concept we have advocated since ancient times by empowering AI technology: teaching students according to their abilities, in order to reform and innovate traditional education.

RECENT AI NEWS

RECENT AI TOOLS