Intelligent Adaptive Learning System (IALS)

 
 

Introduction

The Squirrel Ai Learning Intelligent Adaptive Learning System (IALS) provides a student-centered intelligent and personalized education. It applies artificial intelligence technology in the educational process of teaching, learning, evaluation, testing and training. IALS is cost-effective, with the artificial intelligence + human teacher model, to teach students in accordance with their aptitude, effectively solves the problems of high cost of traditional education, the lack of resources for teachers, and low learning efficiency.

IALS pioneers the decomposition of nanoscale knowledge components, take junior high school mathematics as an example, 300 knowledge components are dissolved into 30000 fine-grained knowledge components, and each knowledge component is matched with learning content, including text item, animation, slides, short instructional video, etc. A parent knowledge component can be resolved into sub knowledge components that are more specific and targeted. The relationship between knowledge components is connected into a graph structure. Then according to the real learning data of the users, the relationship between the knowledge components in the knowledge graph is iterated until the knowledge graph becomes stable.

Nowadays, the modern education system has achieved great success in knowledge education, but its role in quality education is unclear. If we want to improve these capabilities of students through education, how can we really make progress? IALS solved this scabrous problem by proprietary MCM model(Methodology, Capacity, Mode of Thinking).

MCM model(Methodology, Capacity, Mode of Thinking)

We define quality education from the three dimensions. MCM are split into nanoscale just like knowledge points. The ambiguous and incomprehensible capabilities into nanoscale capabilities that could be clearly defined. In this way, we could measure the level of a student's capabilities and represent his capabilities by numbers. In the meantime, we should ensure that the capabilities can not only be clearly explained by teachers and but also be understood and digested by students.

MCM are curated and summarized, take middle school of Math as example, 500 components subdivide them into 1,000 application scenarios to make them completely definable, measurable and teachable.

When recording the show Outwitted in CCTV, six kinds of occupations were invited to take the stage to test their MCM through IALS. A lawyer is excellent in pattern exploration skills and summarization skills and their skills on analyzing 3D graphics and algorithm construction and realization skills are under-developed. In contrast to a scientist who is excellent at data analysis skill but poor at linguistic association skills. SA Beining who is a famous host in China was invited to take the stage to test his MCM through IALS. The evaluation report is shown that his observation capability, deductive reasoning capability and modeling capability were very strong, his geometric and visual capability was relatively poor, and his number sense ability was particularly poor. He said that he could only know roughly his strengths and weaknesses before but the IALS accurately discovered multi-dimension through only 18-minute test of the problems.

In our system, the users’ behavior is taken automatically into account of the algorithm parameters. Although in the initial phase, some of the parameters are set according to experience. For example, the item difficulty is set according to experts’experiences if the number of data samples is less than 25. When the number of data samples is more than 25, the item difficulty will be computed by machine learning algorithm.

IALS Modules

IALS consists of four modules, user persona, assessment, learning, and report. The system is powered by adaptive engine and the customized content tagged with fine-grained knowledge components and MCMs.

User persona module provides fundamental support to other algorithms used in assessment and learning.

Assessment module is one of the most important features in the system to utilize the adaptive engine. Within few questions, the system is able to know learners' mastery level for many nanoscale knowledge. The assessment infers learner's knowledge mastery state without even answering questions on certain knowledge components. Based on learners' persona, the assessment adjusts learners' starting point on a knowledge graph, to a large extent, increases the efficiency of assessment. Moreover, the core of the system is to identify what they have mastered and what students need to learn so students won’t waste their precious time on practicing what they mastered.

Learning module will take learners' persona into consideration and deliver pin-pointed items to them. It prioritizes knowledge components so that learners will start from those they are most likely to master. It shows learners where they are by visualizing knowledge graph. It pops up explanation automatically when the user got a wrong response to the question. And, it offers a hint to help learners learn.

The report module generates a report once learners complete a module, and offers the capability of comparing learners' learning result to others'. The student's learning data is collected during the student learning process. Through the processing and analysis of the data, the learning model of different students is established. Learning Analysis Technology mainly predicts and monitors students' test scores and proposes corresponding interventions. The system will evaluate academic performance and identify weak knowledge components for specific students. Visualize learning path, percentages of mistakes, mastery level, the order of the knowledge graph and professional learning suggestions, which are convenient for students to review.

Success Stories

Qingtai County is a poverty-stricken county in China. The average yearly income of Qing Taiping town is $1025. Squirrel Ai Learning helps the children increased their mastery rate from 56% to 89% in 1 month by using the methods of "tracing the source" for student learning. The achievement level of these rural children not only exceeded that of the children in the county but also some children's level far exceeded the average level of students in Wuhan (the provincial city of Hubei, China).

Algorithms

IALS mainly adopts more than ten kinds of artificial intelligence algorithm technologies. The followings explain how the algorithms work.

1) Clustering algorithm, such as k-means and EM is used to study how different types of students interact with systems. The algorithm is used to explore students' characteristics and preferences, help-seeking activities, self-regulation methods, wrong behaviors, data at different learning times. IALS delivers learning contents according to individual user’s behavior.

2) By logistic regression and EM algorithm, a user ability is determined by seeking the value that maximizes the likelihood function consisting of the value of users’ performance. The users’ performance includes whether the user responds correct or not, the difficulty of the item, the response time, etc. The more a user learns in the system, the more accurately the ability value is determined.

3) Take advantage of the item response theory, probabilistic theory, graph theory, bayesian network, IALS distinguishes knowledge according to difficulty levels, importance, and cognitive levels, models the knowledge system, constructs a “knowledge graph,” and sorts out the logical and cognitive relationships between knowledge components and predict the learning ability of learners when determining the time nodes for the next phase of learning.

4)According to knowledge space theory, information entropy theory, Bayesian knowledge tracing technology together with logistic regression, IALS precisely diagnose the student's knowledge state with as few questions as possible.

The algorithm model takes into account the learning objectives and the current knowledge state of the students, recommends the best knowledge components as the next upcoming learning content for students to learn, and dynamically adjusts the learning path in real time according to the changing knowledge state of the students. With Strategic priority, the key knowledge components that are essential to other knowledge components or are of high score in examinations, the difficulty of the knowledge components are all taken into account to determine priority to be recommended to certain students, therefore, low ability students may not recommended the end-positioned knowledge components. The policy will efficiently make students focus on what matter them most. Accordingly, the engine in real time aims to dynamically determine the most optimal learning path so that it can deliver appropriate learning content for maximal learning effect.

5) Source Tracing Model would find out the root cause of students' weakness in current knowledge by tracing the pre-requisite knowledge components of current learning content. It determines the recommendation priority of each knowledge component from three aspects: whether the pre-requisite knowledge component is easy to learn, whether its map position is relatively backward and the central degree. It can achieve the effect of laying a solid foundation to effectively promote the learning of current and future knowledge components.

Our mission is that every child become people who are capable and decent. When everyone has the most knowledgeable AI teacher beside every time s/he needs, then education equity is not just a slogan, but every child can realize his/her own different dream completely.