Influence Estimation

Influence estimation aims to quantify the impact of individual data points on a machine learning model's predictions, facilitating model understanding and improved trustworthiness. Current research focuses on developing computationally efficient methods, particularly for large models like LLMs and diffusion models, often employing techniques like gradient-based approaches, influence functions, and novel sampling strategies to address scalability challenges. This work is crucial for enhancing model transparency, debugging biases, detecting adversarial attacks, and improving data quality in various applications, ranging from news recommendation systems to combating misinformation.