What we have and continue to build
Key elements of Intelligent Practice
The building blocks of our adaptive practice technology
1. Generative questions
The design of questions on Intelligent Practice is rooted in sound pedagogy, expert content knowledge and skilled use of technology. Our in-house team of Educational Technologists includes both experienced teachers and content specialists from the fields of Physics, Chemistry and Mathematics. We are constantly expanding our own pedagogical content knowledge to ensure we can tailor each question we create. We pay careful attention to unpacking content in a way that puts it in context and makes it meaningful and approachable, resulting in engaging exercises that promote learning.
Rather than being limited to finite worksheets, or multiple choice questions only, we’ve incorporated diverse ways of approaching and asking questions, allowing us to address not only major misconceptions, but common sticking points and different reading levels too.
The system can handle the input and automatic marking of multiple response types, including fractions, formulae, set notation, chemical equations, spectroscopic notation and more - all instantly. Our system’s ability to accept more complex answers also means we can provide a greater range of difficulty in our questions.
The generative nature of each question results from the way the question is created. This is no ordinary question bank. Each question is actually an instance created from an exercise template. Each exercise template is built around one or more central concepts. But, the components, including images, quantities and scenarios, vary with each attempt of the exercise as different instances of the template are generated. Think of it as a computer program within a computer program, with the sole purpose of generating pedagogically-sound practice items - heaps of them. Plus, the system will produce a fully worked solution to match each and every question.
Python is a programming language. It controls how the exercise behaves.
XML is a mark-up language. It specifies how the exercise is structured.
Many versions of a question, generated from the same code, each with its own unique, fully worked solution.
2. Concept-dependency maps
Our team of Educational Technologists has created extensive, sophisticated concept-dependency maps for Mathematics, Physics and Chemistry.
These highly detailed, curriculum-agnostic maps are a hierarchy of concepts, facts, misconceptions and special cases, with dependencies between nodes and groups of nodes. They plot what prior knowledge a learner would need in order to reasonably be expected to understand the concepts tested by a particular question.
3. Adaptive learning and sequencing engine
Behind Intelligent Practice is a machine learning engine, adapting each practise session to the needs of an individual learner by changing the difficulty and sequencing of the questions they see. To ensure that learning through practice has a significant effect on learning, the practice learners do needs to be at the appropriate level of challenge. An appropriate level of challenge is neither too difficult (otherwise learners struggle too much, become frustrated, confused and possibly give up) nor too easy (otherwise learners become bored and are not pushed to improve). The goal of the algorithm is to maintain an optimal cognitive load so that learners, on average, get exercises right 70% of the time. The more they practise, the better it is able to estimate their ability. As learners progress and master concepts, the algorithm adapts so that they move through more difficult levels.
If you're a Bayesian inference geek: The generative model is a hierarchical Gaussian prior over mastery scores with a binomial likelihood function and a logistic link function between the prior and likelihood. Inference is done with a fully factored variational approximation.
Often, traditional practice focuses on a single skill or topic at a time, in what is called "blocked practice" or drilling". Recently, studies have shown that "interleaved practice" is a far more effective technique to increase your ability to learn and retain information. The sequencing algorithm behind Intelligent Practice also ensures a practice session is interleaved so that learners work on several skills in parallel.
Both of these aspects of the algorithm work together to ensure that learners stay motivated and engaged while learning.
Technology for efficient, automated content production
Bookbuilder is a one-stop shop for converting our internal content XML source code into print PDFs, epubs, HTML5 and feature-phone-ready html that we serve to our users. The idea behind it is to remove the need for people to waste their time doing something the computer could do almost instantly.
Written in Python, Bookbuilder handles things like the creation of a table of contents, splitting a book into its chapters and sections, the conversion of LaTeX equations to images for feature phones and other tweaks that make the output as close to publishable as possible for multiple formats - all in one go.
Coming up next
A small piece of our development plans
Integrating concept-dependency maps
We're currently linking all exercises on Intelligent Practice to concept-dependency maps we have developed for Mathematics, Physics and Chemistry. Once integrated, these maps will feed into the algorithm to better sequence the exercises according to a learner's prior knowledge, strengths and weaknesses.
The concept-dependency maps will provide the algorithm with information to engineer a comprehensive learning experience for any learner, the world over. Overlaying additional maps, each one dictating the specific content and order in which it is covered according to a particular curriculum, will allow us to specifically tailor Intelligent Practice to suit different countries and schooling systems.
An integrated learning experience
Our mission is to create an integrated learning experience, incorporating reading, practising, interacting, designing, programming, analysing, thinking, creating and solving, that transcends the constraints of curricula and that is truly effective and engaging
Grow with us
If you like the sound of where we're going and are interested in licensing or partnering with us, get in touch.