Self-Regularization with Sparse Autoencoders for Controllable LLM-based Classification

Keyphrases

• Sparse Autoencoder 100%
• Large Language Models 100%
• Model-based Classification 100%
• Self-regularization 100%
• Classification Model 50%
• Latent Space 50%
• Text Classification 33%
• Privacy Concerns 16%
• Classification Methods 16%
• Disease Diagnosis 16%
• Regulatory Compliance 16%
• Regularizer 16%
• Task-irrelevant 16%
• Real-time Task 16%
• Effective Representation 16%
• Regularization Framework 16%
• Reward Model 16%
• Interpretable Features 16%
• Contextual Embeddings 16%
• Model Embedding 16%
• Classifier Weights 16%
• Model Training 16%
• Task-specific Features 16%
• Engineered Features 16%

Computer Science

• Regularization 100%
• Large Language Model 100%
• Sparse Autoencoder 100%
• Controllable LLM 100%
• Classification Models 60%
• Generalizability 60%
• Text Classification 40%
• Regulatory Compliance 20%
• Privacy Challenge 20%
• Chat 20%
• Classification Method 20%