This article introduces${\sf FedRLChain}$, a novel framework for blockchain-based secure federated deep reinforcement learning, which allows users to securely and collaboratively train a Deep Reinforcement Learning (DRL) model by plugging appropriate aggregation and verification algorithms for specific problems. Unlike existing systems,${\sf FedRLChain}$adopts 1) a novel verification algorithm to prevent malicious clients, 2) an aggregation weight scheme from preventing the global model from getting biased toward any client, and 3) a variant of traditional FedAverage algorithm to accelerate the convergence process. We perform a rigorous experimental evaluation of${\sf FedRLChain}$considering the classic cart-pole problem, and we show a significant improvement in the number of epochs and time required for model convergence w.r.t. the state-of-the-art frameworks – DDQL, BAFFLE, and BASE-PIoT.

The worldwide breakout of the novel COVID-19 has resulted in one of the worst epidemics in modern times since World War II. Although various vaccinations are being produced, their efficacy remains a considerable hurdle. This is especially true when new virus strains emerge. The main challenge to combating this pandemic is diagnosing and isolating COVID-19 positive cases as early as possible. As a result, COVID-19 needs to be detected early and accurately to prevent its spread. This paper proposes a computer-aided automated COVID-19 detection tool based on Computed Tomography (CT-scan) images of lungs. The proposed approach applies an ensemble technique based on Sugeno Fuzzy Integrals with convolutional neural networks (CNNs) as the base model. The lack of COVID-19 data makes it challenging to train a standard CNN from scratch, so we use a transfer learning approach instead of training the base classifiers, VGG-16, InceptionResnetV2, and Xception. We apply the gained knowledge in the target domain of small CT-scan data, considering ImageNet dataset as the source domain. We have also adapted image pre-processing techniques to remove noises so that the model can only focus on specific features. Our proposed framework achieves 98.99% accuracy on a publicly available dataset and outperforms the existing state-of-the-art methods. Experimental results and comparative analysis with baselines establish the need and effectiveness of our proposed model.