Towards Human Centered Education

Towards Human Centered EducationEducation for mankind’s empowermentAndy SpezzattiBlockedUnblockFollowFollowingJun 7Photo by Ana Viegas on Unsplash — The AlpsToday’s educationEducation is at the hearth of societal, individual and collective development and is supposed to facilitate children empowerment.

Yet, in many situations, it fails to do it well, whether because of economic problems or outdated approaches.

According to USI data, about 260 million of children are out of school and a study conducted by Cindy Liu on US students concluded that 1 out of 5 students would consider suicide because of stress at school.

Education is currently not optimized, with a large level gap between students and individual differences not sufficiently considered.

Education in the 21st centurySchool defines what skills are valued by society.

Today, it mainly promotes individualism, competition and zero-sum game.

Yet, living in the 21st century, in a society of growth, resources are growing and highly valuable skills are now cooperation and communication.

In addition, knowledge is now at the core of every economy.

The more knowledge you have, the more patents, publications or innovations you have and the more powerful you are.

This is what makes Google and Facebook the giants they are today.

As a result, to redesign learning approach is key so that schools promote cooperation, self-Discovery, creativity and win-win state of mind.

Catching attention through gamification, fun and customized information is probably the most promising approach.

“I would rather entertain and hope that people learned something than educate people and hope they were entertained” Walt DisneyThe first part of this article introduces the individual differences that influence the learning process.

Then, a digital tool aimed at optimizing each individual learning potential will be introduced, leveraging Artificial Intelligence and Big Data to design a more customized platform.

Towards a creativity and cooperation based education systemHow do we learn?Skills are not fixed; their development is a dynamic process.

They partly result from genetic basis but are also shaped by relationships, environment and experiences throughout the life.

Learning styleLearning style has centered the attention of extended researches.

It has been shown to impact significantly the student attitude and engagement toward learning materials.

Studies have shown that students develop preferred learning styles, with five main profiles: visual, aural, physical, verbal and logical.

One way of leveraging learning style is through technology-enhanced-items (TEIs) such as video segments, audio files, interactive and simulation components.

TEIs also enable active interaction, participation and can deeply evaluate knowledge gain for students, by customizing the evaluation metric of learning.

Cognitive abilitiesCognitive skills are all the functionalities that your brain need in order to process and use information.

We usually have several preferred abilities that enable us to learn more efficiently.

With a better understanding of cognitive skills, we can tailor learning materials to each individual brain.

Studies identify 10 main abilities [1]: attention, working memory, flexible memory (adaptability and ability to change ones’ approach), verbal reasoning, abstract reasoning, spatial perception, verbal memory, visual memory, visual motor speed and processing speed.

For scientific subjects, different questions will require different cognitive abilities such as spatial representation, calculus and logical thinking.

PersonalityNon-cognitive personality factors act like moderators of the learning process.

Sociability, confidence, persistence, ambition, motivation and control over working habits can modulate the general learning approach of one individual.

The general frame used in the field of personality and education is the big five factors.

This is a good basis to cluster people based on their personality traits.

[2]When students are treated equally, their individual differences due to personality traits are usually considered as obstacles in teaching rather than being leveraged and developed further.

The big five factors in personalityA new approach for educationNow, I am introducing a new TEI, that I developed during a class at UC Berkeley.

It leverages the individual differences from a learning perspective, that was introduced just before, to produce a better tool for teachers and students.

At its current state, it is a minimum viable product aiming a demonstrating the benefit such tool can have on large scale customized education.

It is designed to be scalable and easy to handle.

ArchitectureThe TEI is a mobile app that interact with students through questions and recommendations.

It is then used to construct a user database that is analyzed with Machine Learning models to customize the question tests and look for relevant learning materials online.

Then, all the information collected about one student are gathered and displayed in a dashboard that is accessible for current or future teachers.

Architecture of the applicationClustering for personalizationStudent’s learning style profileWe define a profile for each student, so we can know in which way the student is learning more efficiently.

This profiling is done using learning style, personality as well as knowledge of academic subjects.

Clustering questions according to cognitive abilitiesIn order to test students on math questions and to customize the experience, we clustered the dataset according to three abilities: spatial representation, calculus and problem solving, in a semi-supervised way.

We first hand labeled three hundred questions and then used a Support Vector Machine model with L2 regularization to cluster the remaining questions.

Clustering by contentQuestions were also clustered by their content, using NLP techniques for text preprocessing.

Visualizing the feature space, we get an idea of the optimal number of clusters, that is confirmed using the elbow method.

The latent space (of dimension 500 here) is colored by cluster.

It highlights the patterns behind the data.

Dimensionality reduction visualization of the clustering by contentThe clustering can be more or less precise according to what level of information we want to extract.

We can also visualize the most frequent words by clusters, as in the bubble plot below.

Bubble plot of five clustersResultsInteractive dashboardThe students and teachers will interact trough aninteractive dashboard that track the student’s progress, display the recommendations and enable to visualize specific metrics.

Dashboard of the webapp (developped with Flask)RecommendationsAfter gathering questions according to what we want to test and to the learning preferences of students, a recommendation engine is developed:Recommendation of learning content: the algorithm webscrapes several educational websites to find the most relevant articles and online classes.

The first version was only based on wikifier API, a programming interface using ngrams to select relevant Wikipedia articles.

Building on that, a learning content researcher webscrapping edX, academia.

com and other online education platforms provides a more powerful search tool.

Evaluation metricsFor every Machine Learning application, we are wondering, how do we evaluate success?.How do I know if the use of the mobile app has the effect we want on students?.As stated in the introduction, we want our new learning approach to promote the following values: creativity, cooperation, and to enable students to learn more efficiently.

Our goal is, therefore, to define a metric that could evaluate each one of those skills.

Creativity: it is usually correlated with passion at work and pleasure resulted from what you are doing.

When excited about a subject, you will have more ideas and be willing to spend more time on it.

As a result, we can measure creativity with curiosity, that can be measured using the time spend on the app.

[3]Cooperation: it can be evaluated through the interactions with other students on the platforms and the correlation of the amount of interaction with the overall performance.

Learning performance: we also want to measure the overall progress of students.

This can be done by subjects and by cognitive abilities.

Yet, for the sake of simplicity, we compute a general score that evaluates the difference between the current mean score by areas with the means score a month ago.

Then, we can weight each metric to get a unique score.

The proposed weight are the following.

The proposed weightsIt is then used to track students and improve personalization.

Evaluation dashboardApplicationsThe first application is for students lacking an easy access to educational institutions, as in developing countries, mainly in Asia and Africa.

Usually because of inequalities coming from health, cultural differences or sex, those children are kept at the margins of the educational system, and as a result, their personal development suffers a lot from it.

Because of that, more than 72 million children around the world remains out of school [4].

To tackle this situation, our tool can be used as a first education basis, that could be leveraged later on by a futur teacher that would have access to the student data on the app.

Share of the population older than 15 with completed tertiary education, 2010.

[Source: ourworldindata.

org]With a more general perspective the app can be leveraged by every teaching institution as a support to the traditional approach.

It gives a way to easily learn from children, adapt and track progress more efficiently and during the whole life of each individual.

The future of educationNeuroergonomics to optimize how we are using our brainOn the path to an optimal education system, neuroergonomics will be keys to understand how our brain is behaving.

This new area of research involves the collection of physiological data to gather important information such as attention level.

The goal of neuroergonomic is to leverage the cognitive functions of the brain.

It analyses how the brain is interacting with the outside world, monitoring the brain function continuously.

Two tracked capacities important when facing with complex cognitive tasks are working memory and attention.

Using imagery techniques, we can also visualize the evolution of brain activities associated with expertise development (brain using less taxing strategies and cognitive automation) that can be used to evaluate the different learning approaches.

[5]This is used to provide direct insight on peoples’ mental states and intentions.

It can help to tackle challenges like optimizing workload level in order to estimate how to ensure continued engagement with a high-performance level.

[6]“We should adapt our school to our brain” Idriss AberkaneThe raise of more powerful NLP modelsA lot a learning material involves either written documents or vocal ones.

Both could be analyzed using NLP techniques, as have been shown in this article.

With the recent development of high performing NLP models, more accurate and efficient learning tools can be developed.

At the end of February, OpenAI published a new transformer-based model, GTP-2.

This model trained on billions of parameters learned a very complex embedding for words.

Along with BERT, those state-of-the art models could now be leveraged, with transfer learning, to bring more insights for customized education.

[7]Photo from PexelsReferences[1] Betsy Hill.

Cognitive Skills and Differenciation[2] Boele De Raad, Henri C.


Personality in learning and education: a review[3] Wei-Shan Liu, Yu-Tzu Wu, Ting-Ting Wu.

The Study of Creativity, Creativity Style, Creativity Climate Applying Creativity Learning Strategies — An Example of Engineering Education[4] Humanium, Right to Education : Situation around the world.



org/en/right-to-education/[5] Céline McKeown.

Neuroergonomics: a cognitive neuroscience approach to human factors and ergonomics[6] Adrian Curtin, Hasan Ayaz.

The Age of Neuroergonomics: Towards Ubiquitous and Continuous Measurement of Brain Function with fNIRS[7] AlecRadford, JeffreyWu, RewonChild, DavidLuan, DarioAmodei, IlyaSutskever.

Language Models are Unsupervised Multitask Learners.

. More details

Leave a Reply