・ Those who are considering AI operation in the education industry ・ Those who want to know examples of AI operation improvement in education ・Those who want to improve their online learning
Time needed to finish reading this article
5 minutes
Introduction
Now that remote work has become commonplace, I feel that it is an inevitable trend that online education will become the norm. Since there are more opportunities for children and adults to learn various contents online, real-world education is gradually becoming popular.AImade use ofDXis progressing, and the time has come when "students have their own tablets to study", which was unbelievable in the past. So this timeAImade use ofDXFocusing on "education" as an example, and giving hints on how it can be used3Let me introduce you.
1. adaptive learning
Now that online learning has become commonplace, there is a need for "adaptive learning"that optimizes learning content, such as by preparing teaching materials to suit the learner, and realizes more efficient and effective learning. There are many use cases for adaptive learning around the world, and we will introduce the latest examples from among them.
Learning support that presents the right learning questions at the right time
“Repetition is important to retain what you have learned! 』I'm sure you've heard of it. To do this, it is necessary to present learning items to each student at appropriate intervals and provide a learning schedule that retains memory over time.
The figure below shows theRLTutor: Reinforcement Learning Based Adaptive Tutoring System by Modeling Virtual Student with Fewer Interactions] proposed in the paper,A framework that provides optimal learning strategies based on reinforcement learningis. This framework grasps and memorizes the knowledge state of the learner"internal model"and acquire optimal reinforcement policies and teaching strategies through reinforcement learning“Teaching Model (RLTutter)”It has two structures of
The internal model isKnowledge Tracing(TK: a "virtual learner" constructed from the learning history, based on the technique of modeling a learner's knowledge over time. RLTuteris not the learner, but indirectly optimizes the strategy through interaction with this internal model.
Reinforcement learning optimization usually requires model learning with a lot of trial and error. But as I said earlier,RLTuterimplements reinforcement learning using internal models.
In other words, even if the number of learners' learning feedback is limited, it is possible to repeat trial and error with an internal model that grasps the learner's knowledge from the limited learning history.In fact, it is possible to optimize the educational policy without using much learning history.
Source: RLTutor: Reinforcement Learning Based Adaptive Tutoring System by Modeling Virtual Student with Fewer Interactions Caption: Figure 1: Illustration of the difference between the usual reinforcement learning setting and the proposed method. https://arxiv.org/pdf/2108.00268v1.pdf
2. TA: Teaching Augmentation
In recent years, there has been growing interest in the Teaching Augmentation (TA) system, which is a tool that expands and complements the teaching abilities of teachers. TA systems are emerging in a variety of fields, taking many forms, including ambient displays (virtual displays), wearables, and learning analytics dashboards.
AI tutors supporting teachers
Introduced in “Designing for human–AI complementarity in K-12 education,”Lumilois an application for smart glasses that enables teachers to recognize students' learning status, metacognition, behavior, etc.
As teachers survey the classroom, mixed reality icons such as emojis and question marks appear above students' heads. With such icons appearing, it becomes easier to grasp the situation of all students during learning, and it is possible to judge whether each student needs or does not need learning support regardless of the teacher's experience. Become.
In the mockup example shown in the figure below, when AI determines that a student is struggling with a specific skill during a lesson, Lumilo displays the results of that determination and specific examples of recent student mistakes in similar problems. display. By utilizing this function, teachers are expected to be able to provide more specific learning support.
Source: Designing for human–AI complementarity in K-12 education Kenneth Holstein and Vincent Aleven Caption: Fig. 1. Design mock-ups based on findings from low- to mid-fidelity prototyping sessions. https://arxiv.org/ftp/arxiv/papers/2104/2104.01266.pdf
3. Speech to Text
Speech-to-text technology is called "Speech to Text technology", and it has received particular attention recently because it can be used for online learning via real-time live captions.
This technology can be used to support deaf people, dyslexic children (who can write with their voice using text-to-speech technology), and students who have trouble keeping up with their teachers.
Until now, it has been said that Japanese is more difficult to convert speech into text than other languages such as English.
However, there are still challenges, and there are not enough datasets on which to train the model.
A large corpus of Japanese speech
In 2021, a new large-scale Japanese speech corpus was presented for training an automatic speech recognition (ASR) system.
A corpus is a collection of language structures and linguistic information used in natural language processing.
This corpus contains over 2,000 hours of speech and was based on Japanese TV programs and their subtitles. To confirm the usefulness of this corpus, an evaluation dataset constructed based on TEDx presentation videos in Japanese (https://github.com/laboroai/TEDxJP-10K) to evaluate the model.
A model trained on this corpus is the Corpus of Spontaneous Japanese, CSJ:https://ccd.ninjal.ac.jp/csj/), it seems that the performance is better than the model trained with.
We introduced three examples of AI being used in educational settings.
Now that both adults and children have more opportunities to learn, there are many possibilities for support using AI. In fact, our company, Macnica, is also working on initiatives that can be used in this day and age, such as calculating concentration levels using web cameras.
The provision of content that is optimized for each individual learner and the technology that is close to unique students are making rapid progress.
In addition, the hurdles related to Japanese localization when providing these latest technologies are also beginning to shine. I am pleased to say that we are now in an era where we can prepare several solutions for education improvement approaches by utilizing existing AI technology.
■ Sources of content and papers introduced on this page / References
Yoshiki Kubotani, Yoshihiro Fukuhara, Shigeo Morishima, “RL Tutor: Reinforcement Learning Based Adaptive Tutoring System” by Modeling Virtual Student with Fewer Interactions ”,Figure 1: Illustration of the difference between the usual reinforcement learning setting and the proposed method., https://arxiv.org/pdf/2108.00268v1.pdf
Kenneth Holstein, Vincent Aleven,“Designing for human–AI complementarity in K-12 education ”,Fig. 1. Design mock-ups based on findings from low- to mid-fidelity prototyping sessions., https://arxiv.org/ftp/arxiv/papers/2104/2104.01266.pdf
Click here for examples of papers
Case of SPACE Co., Ltd. Throwing a stone into the education industry in the online age ~ A new place of learning created and nurtured with AI ~
Case of Aisin AW Co., Ltd. AI automates visual inspection of car navigation systems
Case of Aisin AW Industries Co., Ltd. In-house development of quality control AI for production lines
Related article
* Tech Blog AI Women's Club * JSAI2020 Pick up from about 900 papers! -Explanability and systemization of AI-
* Tech Blog AI Women's Club * Posture estimation model and application examples that can be understood in 5 minutes
* Tech Blog AI Women's Club * [Transportation industry x AI] Reduction of monitoring load! Various automatic monitoring AI cases
Latest Information
Case Study
Blog
Document List