Ieee federated Machine Learning White Paper
Download 0.56 Mb. View original pdf
|
e-Iraq estra.ar.en
| 12 IEEE SA learning techniques, including clustering, pattern recognition, classification, optimization, control, and recommendations. Very recently the research community has also started to consider reinforcement learning as away to learn pedagogical strategies, and deep learning to dynamically generate contents and to engage in human-AI interactions using natural languages and multimodal communication mechanisms. Typically, the application of AI in education is BC. There are at least three fundamental problems posed by these applications of machine learning to education and training a) The protection of personally identifiable data, which is regulated in general, but even more highly regulated in the educational arena, especially when children are involved. b) The interoperable exchange and sharing of the models generated by machine learning driven learning management systems, such as adaptive instructional system (AIS, many of which are expressed in terms of knowledge, skills, abilities, attitudes, and other characteristics and include learner models, domain models, pedagogical models, adaptive models and interface models. c) The ethical practice of AI, which includes verifying that the models generated are not applied in unwarranted or unwanted ways and are either not biased or transparent about their biases. FML can help address these use cases by Constructing learner models with data from multiple learning systems In this use case, multiple learning systems produce data about learners, some of whom use more than one system, but the systems are prohibited from sharing data and the identity of learners. Each system applies its own machine learning to estimate mastery, or to make predictions or estimate the effect of a particular activity as a function of the aggregation of learner states. These estimates are exchanged among multiple learning systems and a larger model is constructed using federated machine learning to improve the accuracy of each system and, if appropriate, the recommendations it makes. Using FML to aggregate and combine learner interaction data related to domain models This enables machine learning driven content analysis, ontological construction content generation, and content quality improvement for adaptive instructional system authoring. Using FML for improving pedagogical strategies Pedagogical strategies are represented in AIS in many ways. A common way is as a set of rules, which maybe an event-condition-action table, a sequence of speech or dialog acts, rules based on instructional design theories, or branching and remediation rules based on estimates of the learner’s current state. In existing AIS, these action rules are almost Authorized licensed use limited to University of Malta. Downloaded on December 24,2022 at 11:03:39 UTC from IEEE Xplore. Restrictions apply. |