This project explores the use of diffusion models for anomaly detection by leveraging their powerful generative capabilities to learn the distribution of normal data and identify deviations as anomalies. Diffusion models, which progressively denoise data from noise to generate realistic samples, can be trained on non-anomalous data and used to detect anomalies based on reconstruction errors, likelihood estimates, or inconsistencies in the reverse denoising process. The goal is to develop a diffusion-based framework that is effective across various data modalities—such as time series, tabular data, or graphs—and capable of identifying complex, subtle anomalies that traditional models often miss. This approach also opens up possibilities for more interpretable and robust anomaly detection, especially in high-dimensional or structured settings.