Perturbation Analysis

Perturbation analysis investigates how small changes in input data affect the outputs of complex systems, primarily focusing on improving model robustness and interpretability. Current research emphasizes applications in diverse fields, including deep learning (exploring feature attribution and model explainability in neural networks and vision transformers), linear systems (analyzing solver sensitivity to adversarial attacks and data poisoning), and other areas like single-cell genomics and orbital mechanics. These analyses aim to enhance the reliability and trustworthiness of models, leading to more robust algorithms and improved understanding of complex systems' behavior.