Learnability and Privacy Vulnerability are Entangled in a Few Critical Weights

Researchers discovered that privacy vulnerabilities in neural networks exist in only a small fraction of weights, but these same weights are critical for model performance. They developed a new approach that preserves privacy by rewinding and fine-tuning only these critical weights instead of retraining entire networks, maintaining utility while defending against membership inference attacks.

• →Privacy vulnerabilities in neural networks are concentrated in a very small fraction of weights.
• →The same weights that create privacy vulnerabilities are also critical for model utility and performance.
• →The importance of weights stems from their network locations rather than their actual values.
• →Selective weight rewinding and fine-tuning can preserve privacy while maintaining model performance.
• →This approach shows better resilience against membership inference attacks compared to traditional full network retraining methods.