Federated Learning: Schemes, Privacy, and Security by Md Morshedul Islam, Suraj Neupane скачать в хорошем качестве

Federated Learning: Schemes, Privacy, and Security by Md Morshedul Islam, Suraj Neupane

BSides Edmonton 2025 This video was captured using a locked-down, unmanned camera. As a result, there may be moments when speakers are not fully in the camera shot. Additionally, the audio quality captured by the podium microphone is dependent on the proximity of the speaker to the mic. This means that variations in audio clarity may occur if the speaker moves away from the microphone during their presentation. We appreciate your understanding of these technical aspects. ___________________________________________________________________________ Federated Learning: Schemes, Privacy, and Security by Md Morshedul Islam, Suraj Neupane The demand for artificial intelligence (AI) has significantly increased over the past decade, driven by advancements in machine learning (ML) techniques. Typically, ML developers aggregate data on a central server to train a generalized ML model and offer ML as a service. However, centralized ML faces several challenges that hinder the collection of sufficient data on a central server. These challenges include data protection regulations that restrict data transfer across geographic boundaries, social concerns such as personal data privacy, and technical difficulties in aggregating data from diverse sources. In response to these limitations, Federated Learning (FL) has emerged as a promising alternative, providing a privacy-preserving, distributed ML framework. Instead of gathering raw data on a central server, FL enables data owners to collaboratively train a global model under the coordination of a central parameter server, typically using an aggregation protocol. FL has shown significant potential across various fields, such as healthcare, agriculture, cybersecurity, the Internet of Things (IoT), and more. Despite its promise, the general FL framework often converges poorly under heterogeneous conditions, which can degrade the performance of the global model. Furthermore, although FL enhances privacy by keeping data localized, it remains vulnerable to system security threats and data leakage. This research talk aims to explore these challenges and discuss potential solutions.