The problem De_FL solves

• Federated Learning, as good as it is, has a primary flaw that it is prone to data breaches and if we go towards the route of encrypting data to protect user privacy, then users can inject malicious data that poisons the complete data set and lowers the accuracy of the model.
• Our aim was to solve both of these issues at once. We aimed to build a POC(Proof of Concept) on applying Federated Learning on Blockchain as blockchain ensures transparency, verifiability and anonymity simultaneously. We built a user-case
  for the case of self-driving vehicles. We implemented it by letting only the aggregator server having access to the models creating by individuals, and the key to decrypt all these local models on the blockcahin network.
• It takes the federated average of the local models to update the global model in place, and then pushes the new global model to update all the local models at nodes.

Challenges we ran into

First we thought to implement the mathematics of federated learning through flower framework but it's integration was not feasible for us and the results were very unsatisfactory, we then finally had to frame our own architecture for this. We also faced several issues with the integration of Web3.storage and filecoin in an encrypted way but again had to shift to lighthouse at the very last moment which resulted us very favorable ,then there were some difficulties with the integration of biconomy but we figured it out.