An Uncertainty-Aware Transfer Learning-Based Framework for COVID-19v Diagnosis

More and more studies are looking at the use of dynamic off-chain transportation in IoT systems that are based on payment channel networks (PCNs). Flexible routing in PCN-based IoT, where devices often have little resources, has two key challenges, however. The primary problem is figuring out how to make PCNs with variable channel capacity work efficiently for long-term transactions. The second issue is how to maximise transaction efficiency (the number of successful payments and the proportion of those that go through) by employing a lightweight routing strategy on IoT gadgets. To achieve this objective, we propose a powerful deep reinforcement training method for learning the combined flexible and lightweight routing approach. We use a proximal policy optimising strategy to attain the best effectiveness in dynamic routing problems for off-chain systems via an actorcritic learning structure to train the instructor's DRL model. By using an adaptive pruning strategy to remove superfluous network parameters from the teacher model without compromising the latter's receptiveness to new information, we are able to generate a compact and effective DRL model for the student. The effectiveness of the pupil's network is also improved by the use of knowledge distillation. Therefore, it may be possible to construct a small and efficient pupil DRL model and utilise it to optimise the long-term efficacy of operations in off-chain networks on limited-resource IoT devices. Results from computer simulations show that the suggested DRL algorithm improves PCN transaction performance compared to the baseline techniques, while using just
$10{{\% }}$
of the resources available in the original instructor model.