cho.sh

researchmini

3,249 notes

260722260722 I prefer TUIBB9A08 MarkRight8B0501 에이전트 런타임 서버 셋업1983DC Bun을 Rust로 재작성하다2BDED3 260721260721 Agent Runtime Server SetupC38C2C 260719260719 AutoBuilder63849A 260718260718 Slack Bots are the new CLIs3E662F 260717260717 WristcomputerD98BDF 260716260716 SalpakanA2940F AGENTS.mdDFE522 260714260714 260713260713 260712260712 260711260711 인간의 보완재와 AX41F1E5 260710260710 tokenmaxxing cli635294 260709260709

Qiang Yang et al. Federated Machine Learning Concept and Applications · cho.sh

Qiang Yang et al. Federated Machine Learning Concept and Applications

Qiang Yang et al. Federated Machine Learning Concept and Applications

Warning

This post is more than a year old. Information may be outdated.

Federated Machine Learning: Concept and Applications

Intro

Data exists in isolated islands
More security required

First FML framework by Google, 2016

Horizontal FL
Vertical FL
Federated Transfer Learnings

Traditional Data Processing Models

aka Simple Data Transaction Model
3 parties
- Data Collector
- Data Sanitizer
- ML Trainer
Privacy concerns

Overview

aim to extend to all privacy-preserving decentralized collaborative machine learning techniques.
simple definition
- parties federate their data without exposing them to each other to attain performance closely comparable to the model trained as if all the information were gathered.

Backlinks2

Project Florence Literature Review 230828

Comments

N

N

Secure Multi-Party Computation. jointly compute over their inputs while keeping those inputs private.

Differential Privacy. Add random noise to the data, making identifying any individual's data in the aggregate results difficult.

Homomorphic Encryption. allows computations on encrypted data without requiring decryption

Indirect Information Leakage

Considered Blockchained FL architectures

Categorization of Federated Learning

Feature
Horizontal Federated Learning. sample-based FL
- Deep Gradient Compression to share data efficiently
- assumes honest participants and security against honest-but-curious servers.
Vertical Federated Learning. feature-based FL
- assumes honest-but-curious participants
Federated Transfer Learning
- Security measures are similar to VFL

Architecture

Horizontal Federated Learning

Participants compute locally, send to server
Server aggregates global model, distributes to participants
Participants update their local model

Vertical Federated Learning

We need an intermediary collaborator.
Step 1: Collaborator C creates encryption pairs and sends the public key to A and B.
Step 2: A and B encrypt and exchange the intermediate results for gradient and loss calculations.
Step 3: A and B compute encrypted gradients and adds additional mask, respectively, and B also calculates encrypted loss; A and B send encrypted values to C.
Step 4: C decrypts and sends the decrypted gradients and loss back to A and B; A and B unmask the gradients and update the model parameters accordingly.

Privacy-preserving Machine Learning

Privacy-preserving machine learning is designed to perform learning while keeping data private.

Techniques

Federated learning. A decentralized collaborative machine learning method.
Secure multi-party computation (SMC): Provides privacy guarantees.
Secure multi-party decision trees, k-means, Naive Bayes classifier, etc.: Algorithms for various machine learning tasks with privacy preservation.
Homomorphic encryption. Allows computation on encrypted data without decryption, ensuring privacy.
Yao's garbled circuits: Another privacy-preserving computation method.

Federated Learning vs. Other Concepts

Distributed Machine Learning

Distributed ML focuses on distributed storage and operation.
Uses tools like "Parameter Server" to efficiently store and compute.
Differences with Federated Learning:
- Data privacy emphasis in Federated Learning.
- Distributed ML centralizes control, while Federated Learning decentralizes control to data owners.

Edge Computing

Federated learning serves as a protocol for edge computing.
To optimize learning, focus on determining the best trade-off for local updates and global aggregation.

Federated Database Systems

These systems integrate multiple databases.
Differences with Federated Learning:
- No privacy mechanisms in federated database interactions.
- Federated learning aims to create a unified model across different data owners with privacy.

Applications of Federated Learning

Smart Retail

Personalize services such as product recommendations.
Challenges: Data privacy, security, and heterogeneity across different entities (e.g., banks, social networks, e-shops).
Solution: Federated learning can train models without sharing raw data, thus overcoming privacy barriers.

Finance

Detect multi-party borrowing which is a risk to the industry.
Federated learning can help find malicious borrowers without exposing user lists.

Smart Healthcare

Challenges: Sensitive medical data scattered across isolated centers.
Solution: Federated learning combined with transfer learning can share model insights without sharing patient data.

Federated Learning as a Business Model

Traditional Approach

Aggregate data, use cloud computing to compute models, then use results.

With Federated Learning

Data stays where it is; only model insights are shared.
Privacy and data security are prioritized.
Offers a new paradigm for big data applications.
Can use blockchain for profit allocation in a data alliance.
Calls for establishing standards for federated learning for faster adoption.

Comments0

No comments yet. Be the first to share your thoughts.