YOURIKA’s
Knowledge Prediction Engine
At YOURIKA we recently had a breakthrough innovation. We created a Knowledge Prediction Engine that is extremely accurate with results that are three to four times better than anything that exists in the market. In this post, I will share more about what we built, how it works, and why it will have a profound impact on the EdTech Community.
Paving the Way for Massive Advancements in Learner-Centric Education
Let’s talk about Numbers:
4x
more accurate than other prediction engines in the market today.
98%
of students who received instruction from a tutor outperformed their peers
85%
accuracy with prediction testing based on real students studying one subject
The YOURIKA Knowledge Prediction Engine.
It is this recognition of the challenges faced by education today, combined with our team’s extensive experience in EdTech and in the field of Artificial Intelligence (AI) that served as the impetus for Yourika’s breakthrough innovation – our Knowledge Prediction Engine – a learner-centric system that leverages advanced capabilities in machine learning and natural language understanding to directly deliver “in the learning moment" guidance to a learner that:
-
Identifies their learning gaps.
-
Provides an accurate measure of their level of knowledge (a measure of understanding) in each topic.
-
Identifies the dependencies among and between topics so their progression of learning and understanding can be charted.
-
Creates a “next generation” assessment model that no longer relies only on tests/exams to determine student understanding, but rather offers continuous activity monitoring – providing valuable feedback to the learner and the teacher.
The ability to use a machine to accurately measure learner understanding that emulates a 1:1 instructor/student relationship is quite a revolutionary concept for the order education sector. In fact, many would consider it an impossible task. But our team at YOURIKA, backed by decades of academic research into AI, machine learning and natural language processing, knew we had a technological solution to this problem.
Model Predictions vs. Real Student Performance.
Over the last several months we began applying our intellectual property (IP) that leverages Deep Learning to test and train our models on Order of Magnitude Less Data than the common machine learning techniques against existing data sets from real students. Our goal was to validate if our Knowledge Prediction Engine could predict – with high accuracy – a student’s ability to answer a series of questions correctly based on their understanding of a topic. We compared our Engine’s ‘predictions’ with real student performance. The Engine performed with an unprecedented level of accuracy.
When compared against real student performance, our Knowledge Prediction Engine performed with an unprecedented level of accuracy -- three to four times better than anything that currently exists in the market.
Prediction testing based on real students studying one subject (i.e. Math - Calculus) with anywhere from 10+ to 100+ topics (i.e. Integration, Differentiation) and 100+ question/answer pairs, resulted in from 85+% to 90+% accuracy. This is the typical number of topics in a secondary and post-secondary course.
Although a typical course wouldn’t see more than 100+ topics in one subject, we wanted to test the robustness of our model by introducing a variety of subjects (i.e. Music, Arts, English, Psychology) that you wouldn’t usually see together in a course. Introducing a variety of subjects meant that we would expect to decrease the accuracy of the model, because the more subjects we add, the harder it is for the model to distinguish between topics across subjects.
With this in mind, we tested our model with different real-world datasets as shown in the table below. When we introduced 5,000+ topics (and 50K+ question/answer pairs) the accuracy remained over 70% and held out at that level even when we increased the number of topics to 8,000+ (and 70K+ question/answer pairs). This shows how robust our model remains beyond the requirements of any typical course.
The results we achieved are three to four times better than anything that currently exists in the market.
What This Means for Education and for Learners.
At its launch, our inaugural product, ORBITS, will have the Knowledge Prediction Engine at its core, creating the world’s first learning community that puts students first. This revolutionary learner-centric platform will allow students to collaborate with peers and teachers and study smarter with personalized resources, all while gaining an understanding of their own knowledge in the ways described above.
YOURIKA's Knowledge Prediction Engine is not limited to ORBITS. It can be easily integrated into any educational content or learning management system. This means the entire educational world can benefit from incorporating knowledge prediction into content delivery models to better pace learning, sequence topics, and provide better learner feedback.
YOURIKA's Knowledge Prediction Engine easily integrates with any educational content, giving students and educators a very clear picture of an individual’s grasp of any topic and where gaps exist in that individual’s learning.
Students and educators can now access insights and analytics never before available to gain a clear view of an individual’s grasp of any particular topic and where gaps exist in that individual’s learning.
We can measure student knowledge per topic and over time and can provide recommendations to that student on what to study/learn next and feedback on where there is a shortage of knowledge. This information will help to improve a student’s learning experience, inform pathways for study, and accelerate knowledge acquisition. It also means students and instructors can build more personalized learning paths to address specific learning styles and challenges. This level of insight will help to achieve a learner-centric system that directly delivers personalized guidance to the learner.
The concept of understanding is foundational to learning success. Through our Knowledge Prediction Engine, powered by advancements in AI, machine learning and natural language processing, we have made huge strides forward in the use of technology to empirically measure student knowledge. And through this discovery, we have unlocked great potential for personalized learning at scale and paved the way for advancements in learner-centric education.
Interested in furthering the conversation about knowledge prediction? Want to know more about this innovation? Feel free to reach out to me or to the YOURIKA team to talk further.
Machine Learning and Operations.
Through machine learning and operations (MLOPs) we built continuous delivery and ML automation pipelines to continue to advance and enhance our Knowledge Prediction Engine’s capabilities by fine-tuning the model parameters to achieve the best results. This means that the time taken to train the model on a dataset and to generate predictions is minimal, and the process is fully automated.
Through machine learning and operations (MLOPs) we built continuous delivery and ML automation pipelines to continue to advance and enhance our Knowledge Prediction Engine’s capabilities by fine-tuning the model parameters to achieve the best results. This means that the time taken to train the model on a dataset and to generate predictions is minimal, and the process is fully automated.
Students don’t know what they don’t know.
Often students lack a good grasp of their own learning understanding. In other words, they “don’t know what they don’t know.” The educational system currently relies on strategies such as self-reflection, “I can” rubrics, and technology-guided self-paced learning paths to empower students to take charge of their own learning, but until now, there has been no true empirical measure of understanding that learners (or instructors) can rely upon.
Our learner-centric Knowledge Prediction Engine leverages advanced capabilities in machine learning and natural language understanding to directly deliver “in the learning moment" guidance to a learner.
From this, we recognize that:
-
Understanding is foundational to learning success. If we are able to achieve understanding this unlocks huge potential for greater personalization and advancements in learning;
-
Personalization (or one on one tutoring) helps drive a better understanding of topics and material and that improves individual performance and learning outcomes;
-
The educational system as a whole is currently not structured to measure understanding or to deliver personalized learning at scale.
At YOURIKA we recently had a breakthrough innovation. We created a Knowledge Prediction Engine that is extremely accurate with results that are three to four times better than anything that exists in the market. In this post, I will share more about what we built, how it works, and why it will have a profound impact on the EdTech Community.
Research on cognitive science in recent decades has significantly advanced our understanding surrounding the study of learning. Applying this understanding is strongly advocated by leading mathematics and science educators and researchers for all students, and also is reflected in the national goals and standards for mathematics and science curricula and teaching.
However, regardless of the topic at hand, measuring a student’s understanding of a topic or material is one of the most difficult tasks in education. Accomplished instructors and teachers will use as many as 50 different ways to assess an individual student’s mastery of the subject matter, and still, the determination of understanding remains somewhat subjective.
Achieving understanding unlocks huge potential for greater personalization and advancements in learning.
At YOURIKA we recently had a breakthrough innovation. We created a Knowledge Prediction Engine that is extremely accurate with results that are three to four times better than anything that exists in the market. In this post, I will share more about what we built, how it works, and why it will have a profound impact on the EdTech Community.
Research on cognitive science in recent decades has significantly advanced our understanding surrounding the study of learning. Applying this understanding is strongly advocated by leading mathematics and science educators and researchers for all students, and also is reflected in the national goals and standards for mathematics and science curricula and teaching.
However, regardless of the topic at hand, measuring a student’s understanding of a topic or material is one of the most difficult tasks in education. Accomplished instructors and teachers will use as many as 50 different ways to assess an individual student’s mastery of the subject matter, and still, the determination of understanding remains somewhat subjective.
Achieving understanding unlocks huge potential for greater personalization and advancements in learning.
Researchers agree that frequent and deliberate use of learning strategies is related to academic achievement (Boekarts, Pintrich and Zeidner, 2000) and can empower learning. Strategic learners are better aware of themselves as learners, employ different knowledge acquisition strategies, understand the specifics of task qualifications, can connect prior knowledge to new knowledge and the possible contexts where knowledge could be useful, and engage in meta-cognitive activities while learning. Furthermore, research shows that the most important bottlenecks impeding student success are the lack of students’ own goal-setting skills and abilities, and the current curriculum design, which, especially in the observed higher education institutions, provides little support for goal-setting interventions.
In 1984, research into personalized learning by educational psychologist Benjamin Bloom found that students who received personalized instruction from a tutor outperformed 98% of those who did not. In addition, the average student taught under mastery learning outperformed 84% of the students taught under a conventional form.
Class sizes and student-teacher ratios are much-discussed aspects of education and play a key role in the delivery of a more personalized learning experience. A study by Glass and Smith (1979) found that smaller class sizes can have a positive effect on achievement in later grades. But currently, with the average student to teacher ratio at 16:1 in American public schools and 18:1 in colleges and universities (with many academic institutions exceeding that number) instructors are challenged to provide students with individualized learning opportunities. Too often, teachers are forced to “teach to the middle,” meaning that high achievers and those falling behind get less personalized attention and instruction than the middle performers.
Further compounding the issues of class size, every learner has his/her own preferred learning style and set of learning challenges. These differences (personality, perception, ability, intelligence) affect students' motivation and attitudes towards the lessons and can impact the effectiveness of the learning.
Add to all these contributing factors the recent shift en masse to online learning due to the COVID-19 pandemic. The massive move to remote learning has exposed many failings with respect to personalized learning. In talking to the market we have learned that students and instructors are dissatisfied with their learning experience, and learners feel unmoored, isolated and frustrated with the current learning process.