# Defending malicious behavior in federated learning with blockchain

- Channel: [ZuBerlin](https://streameth.org/zuberlin)
- Date: 2024-06-18
- Duration: 28:49
- Topics: blockchain, AI, machine learning, cybersecurity, fintech
- Watch: https://streameth.org/watch/6671a02907f92b086c3bafe5
- Download: https://vod-cdn.lp-playback.studio/raw/jxf4iblf6wlsyor6526t4tcmtmqa/catalyst-vod-com/hls/e5e4cmrroy5lp5jy/1080p0.mp4

## Description

Clip The video features Jiahao Sun, the founder and CEO of Flock, discussing the company's mission and technology. Flock is a pun for federative learning blockchain, an idea stemming from a paper published in NeurIPS 2022 and further detailed in the journal "Transitioning AI." Jiahao's background includes working in traditional finance as head of AI at Eurobank Canada and as a researcher at Imperial College London.

Flock aims to solve data silo issues in industries like finance and healthcare by combining federated learning with blockchain technology. Federated learning occurs on devices like smartphones, where data stays local while only model weight changes are sent to improve collective models. However, there's a risk of centralized control and data misuse, which Flock addresses using blockchain for transparency and security.

The platform introduces an incentive mechanism where good behaviors are rewarded, and bad actors are penalized, ensuring honest participation. The system uses random selection for proposers who contribute to model training and voters who validate results. Malicious attempts are thwarted through this process, maintaining high accuracy even with a significant presence of bad actors.

Jiahao also discusses experiments with the Kaggle Learning Club dataset that demonstrate the effectiveness of Flock's approach compared to traditional federated learning under malicious conditions.