There’s a lot of promise around artificial intelligence for education to personalize learning; however, there has been very little research regarding Artificial Intelligence (AI) applications in fields with very few resources for its implementation. This paper describes a proposed AI-based adaptive learning system that aims to personalize STEM education in a low-resource school environment in Kenya. This research addresses the numerous challenges associated with such a system, such as irregular internet access, limited computer hardware in situ and no previous teacher background in both AI and education. To address these issues, an adapted reinforcement learning algorithm will personalize the content shown to students, and a modified liquid neural network is used for the prediction of student success, while not being computationally expensive. As compared to traditional adaptive systems, this adaptive learning platform supports edge computing and offline updates in order to operate in a consistently low connectivity environment. In addition, by continually adjusting the difficulty, format, and rate of delivery of STEM topics to fit the style, prior knowledge, and attention level of the individual student, this platform has been seen to improve the educational experience. An 8-month case study was done with 6 Kenyan low-resource schools and 6 comparison schools located in the city. In this case, we report a 31% increase in the level of understanding of students’ key STEM subjects, 27% reduction in student drop rate from STEM topics, and 43% increase in teacher efficiency over traditional methods. We were also able to predict the level of performance of students to a 89% success while occupying a low 1.9MB memory, making it feasible to be employed on budget Android devices. This study presents evidence that the utilization of AI for personalized adaptive learning technologies in order to minimize the disparities in the provision of STEM education in low-resource settings worldwide is possible. 

The proliferation of massive IoT networks has created an environment where distributed AI can be achieved. At the same time, it introduces serious privacy and security challenges. Federated learning (FL) allows training local models on IoT devices and aggregating them without sharing data, but still suffers from problems such as gradient inference attack, Byzantine model poisoning attack and the failure in single point of failure centralized aggregation point. In this paper, we propose QFL-BC, a framework combining Quantum Key Distribution (QKD) and a permissioned blockchain to holistically tackle the problem. Using the BB84 protocol with decoy states, QKD generates a One-Time Pad key to encrypt the model update and achieve information-theoretic security with provable security against a quantum attacker. The central aggregator is replaced by the permissioned blockchain with a smart contract, which ensures an immutable audit trail and distributes the orchestration of FL training decent rally, as well as imposes a penalty on malicious participants by automatic reputation score maintenance. The experiments with MNIST and CIFAR-10 on 100 IoT clients under Non-IID conditions show QFL-BC obtains an accuracy of 96.8% against 41.5% for classic FL under 10% poisoning attack (133% relative improvement). We have tested its robustness across adversary percentages of 10%-40% with accuracy above 87.3% and measured scalability up to 500 clients, showing good degradation, communications overhead of 5.84 MB per round, which is only 12.3% higher than the classic FL and analysed latency and energy to evaluate its feasibility on resource-constrained IoT devices.