Anti-Backdoor Learning: Training Clean Models on Poisoned Data.
Yige LiXixiang LyuNodens KorenLingjuan LyuBo LiXingjun MaPublished in: CoRR (2021)
Keyphrases
- learning models
- prior knowledge
- accurate models
- data sets
- learning algorithm
- learned models
- data collection
- data analysis
- structured prediction
- experimental data
- historical data
- data points
- hidden variables
- human experts
- data sources
- supervised learning
- bayesian methods
- data structure
- raw data
- online learning
- background knowledge
- missing data
- database
- input data
- probability distribution
- learning process
- training data
- statistical analysis
- learning tasks
- learning problems
- small number
- knowledge discovery
- probabilistic model
- active learning
- reinforcement learning
- decision trees
- learning speed
- relational models