Elevator Pitch
Introduce the interdisciplinary nature (learning, cognitive, and computer science) of human learning through online learning platforms. Focus on the workforce learning to illustrate how data science and machine learning can help understand human learning, with 7 recent publications to be dived into.
Description
Introduction of human learning through online learning This talk is to introduce the interdisciplinary nature (learning, cognitive, and computer science) of human learning that is increasingly happening through online learning platforms, especially after the pandemic hit us. This talk will focus on the applications from industry workforce learning to illustrate how we can use data science and machine learning techniques to understand how we learn and improve the way we learn. The techniques have been published in several learning science focused venues in the past 1.5 years and are generalizable to other learning contexts (e.g., higher education). Data science supports a better understanding about human learning and improves how humans learn Anonymous learning data collected from online learning platforms is beneficial to us in the sense of providing evidences of the cognitive learning process happening in the brain as we learn. Statistic analysis, traditional classification and regression approaches, deep learning based algorithms, Natural Language Processing (NLP) techniques, meta learning strategies, and probabilistic modeling solutions empower us to understand the underlying cognitive process during learning. Furthermore, it supports us to design and conduct experiments to interatively improve the way we learn in terms of achieving proficiency more effectively and efficiently. Outline of the talk In this talk I will present different applications and techniques for sub topics regarding human learning through online learning, especially for but not limited to workforce learning. First, I will introduce the audience about what is human learning through online learning, why it is emerging rapidly, and how beneficial developing/applying such techniques for improving human learning is to the society. After that, I’d present the work for each sub topics (Build student models of learning, Build skill models of learning, Personalized learning, Build models with little data, Collaborative learning, see in Appendix A) around the work accomplished in the past 1.5 years (part of a science team at Amazon). Takeaways from this talk Attendees will learn the interdisciplinary nature of human learning, and how data science and machine learning can help us understand how we learn and improve the way we learn. The talk will be beneficial for them in finding the relevant literature, practical applications, and the state-of-the-art techniques around online learning from data science perspective. The talk will also encourage attendees to think about how we learn in what context with what goals, to find suitable applications in both academia and industry scenarios to experiment and introduce better ways (evidence based) of learning.
Notes
The seven research papers around the topic to be introduced and dived into have been published at ACM L@S conference, AAAI Ed AI conference, in the past 1.5 years. It would be a great opportunity to share the findings with a broader women science community, on the interdisciplinary research area - human learning. It would also be interesting to share with others, from personal professional growth/learning perspective, how to step into such an area and deliver/build things that bring inner peace. Personally, I started researching in this realm around Sep 2019, and have been enjoying the process of understanding the truth of how we learn and trying to make things work better.
Appendix A. mentioned in Description
Build student models of learning A student model of learning is a computational model that describes or simulates how individuals learn through a learning experience. Knowledge tracing is the family of the techniques that aim to approximate the knowledge state of the brain, as well as how the knowledge state changes from one to another by observing the way individuals learn. Accuracy, reliability, and interpretability of knowledge tracing is the primary research focus that aims to improve the simulation quality. Some investigation is done around how to apply the SOTA machine learning algorithms to improve the accuracy, provide interpretation and reliability of knowledge tracing with learning data collected through an online workforce learning platform. Publication links: A novel approach for knowledge state representation and prediction Evaluating bayesian knowledge tracing for estimating learner proficiency and guiding learner behavior Interpretable personalized knowledge tracing and next learning activity recommendation
Build skill models of learning A skill model of learning, or a cognitive model of learning, is a descriptive account or computational representation of human thinking about a given concept, skill or domain. It includes both a way of organizing knowledge within a subject area, and an account of how humans develop accurate and complete knowledge of that subject area. Due to the fact that the cognitive processes happening in brain is hard to observe and domain experts having the expert blind spot, it is important to design experiments that are observable, analyze observations, optimize the understanding of cognitive process, and ultimately improve the way humans learn. Some research is done around building cognitive models of learning from learning interaction data, with which the instructional design could be improved. Publication under the review process.
Personalized learning Personalized learning is defined here as understanding the individual, the learning goal, the content, and the learning context, and providing targeted learning experience to achieve some desired learning outcome. The challenge is how to describe the individual and his/her goal, to represent the learning content, and to contextualize the learning experience so that the experience can be tailored for that specific learning need in its most effective way. Some research is done on this topic by understanding and modeling the individual, the content, and the learning context to provide recommendations on the next learning move and also insights on where he/she is in the learning journey, how far it is towards the end goal. Publication link: Interpretable personalized knowledge tracing and next learning activity recommendation
Build models with little data How to model and provide insights when data is too little to train? Algorithms in cold start setting is challenged when it comes to reliability and accuracy. Meta learning is outperforming by either optimizing the gradient descent of gradient descent, or leveraging external memory bank to keep track of the gradient descent procedure for reference in the later optimization. Some research is done around using attention mechanism and external memory bank, along with reading and writing operations, to provide accurate prediction with little training data. Publication link: Cold start knowledge tracing with Attentive Neural Turing Machine
Collaborative learning Collaborative learning here is defined as agent-human interactive learning. Agent refers to a system, a voice assistant, or a bot providing hints and nudging. This research topic is around how to leverage the collected data, the rich insights from the data, and provide interactive learning environment to make learning more fun, effective and additive. Some research is finished around introducing Alexa, the voice assistant, into online learning to provide in-person practice for developing some desired skills where in-person practice is a more effective approach for developing that skill. I also researched a bit into how to provide targeted feedback for the free-form text responses in achieving some learning outcome. Publication link: Introducing Alexa for e-learning Targeted feedback generation for Constructed Response Question (Link coming soon)