Researchers have developed a new algorithm that significantly improves the ability to learn predictors robust to adversarial examples. This method achieves linear sample complexity in the VC dimension, an exponential improvement over previous bounds. The algorithm combines bootstrap aggregation (bagging) with robust empirical risk minimization (RERM), computing RERMs on multiple bootstrap samples and outputting their majority vote. A complementary lower bound indicates this approach is necessary for learning in this model. AI
IMPACT This research could lead to more resilient AI systems capable of withstanding adversarial attacks.
RANK_REASON The cluster contains two identical academic papers detailing a new algorithm for machine learning.
- arXiv
- bootstrap aggregating
- Breiman
- Hanneke
- Montasser
- Reconceptualizing Educational Research Methodology
- Srebro
- venture capital
- Bootstrap Aggregation for Model Selection in the Model-free Formalism
- robust empirical risk minimization
- VC classes
AI-generated summary · Google Gemini · from 2 sources. How we write summaries →
