FedMID: A Data-Free Method for Using Intermediate Outputs as a Defense Mechanism Against Poisoning Attacks in Federated Learning.
Sungwon HanHyeonho SongSungwon ParkMeeyoung ChaPublished in: CoRR (2024)
Keyphrases
- synthetic data
- prior knowledge
- input data
- missing data
- data sets
- learning algorithm
- bayesian methods
- test data
- data analysis
- learned models
- learning mechanism
- statistical methods
- unsupervised learning
- data collection
- preprocessing
- data structure
- image data
- training samples
- reinforcement learning
- missing values
- data sources
- knowledge discovery
- pairwise
- distributed data
- information loss
- noisy data
- prior information
- neural network
- machine learning
- decision trees
- learning tasks
- training data
- background knowledge
- feature space
- k means
- probabilistic model