The multi-task learning provides an opportunity for identifying students that require attention
Everyday, with the world harnessing new information, the students are feeling the challenge to retain it. With the vast amount of information and subjects that demands understanding often feel getting pushed by education. While some of the students can overcome this challenge, many others need assistance to counter it. But it becomes difficult for a teacher, to selectively aid student in learning, especially when they also feel the burden of collective performance. To address this issue, an artificial intelligence is designed by the researchers from North Carolina State University, from predicting the position of educational games in inducing learning amongst students.
Cited as the Predictive Student Modelling in Educational Games with Multi Tasking, the paper describes the utilization of predictive analysis as the basis for formulating the performance of a student.
What is Predictive Student Modelling?
Predictive Student Modelling is an assessment test that determines the performance of a student based on his/her past interaction with the learning environment. It is a formulative assessment of the knowledge and skills of students, thus providing them with an opportunity of personalised support and to learn through experiences which are both effective and engaging.
It is essential for tailoring the experiences of students in a range of learning environments like intelligent tutoring and educational games. The Predictive Student Models represents the student knowledge as an aggregate of students performance.
The researchers incorporated the artificial intelligence in education with the help of creating a model known as Multi Tasking learning (MTL) model, which requires performance of multiple task at the same time.
It is observed by the researchers that the shared information across multiple variables improved, with additional data and an enhanced regularization. MTL also reduces the computation time required for comparing the outcome. It demands output based on lesser parameter models. It allows the separate modelling for individual questions with the possibility that each question can have different characteristics as compared to the underlying latent variable.
The MTL model involves assessment of a students’ performance based on only a standard neural network and not on the sequence of students behaviour in an adaptive learning environment.
The researchers deployed the Item Response Theory (IRT) which models the probability that the student will deliver an answer based on the characteristics incorporated by the test-taker and the questions. It doesn’t require assumption of considering all the questions as difficult, instead models the individuals success based on the students’ capability and difficulty of the question.
The researchers at the university determined the dataset of the model with the help of a microbiology game Crystal Island. In Crystal Island, the students are told to take up the role of a medical agent for investigating the outbreak of an infectious disease at a remote island. The data was then collected by asking the students to talk with non-player characters, explore different locations, read virtual books and microbiology posters, test hypothesis about the outbreak in virtual laboratory and findings of the virtual diagnosis sheet. The system recorded the action type, action arguments, location and game time elapsed.
The researchers observed that as the number of tasks increased, so did the performance of students. The paper also states that with the help of educational games like Crystal Island, students are more prone in gaining knowledge in an adaptive learning environment. This type of model also provides personalised support especially when the learner is deviating towards a negative outcome. This model can thus be employed by educators or teachers as an “early warning system”, which would enable the re-allocation of assistance of those students that requires attention for learning.
click here to watch making of B-AIM: