My Projects
Score-based Diffusion Generative Classifier
Discriminative models often face a trade-off between data likelihood and classification accuracy. In this study, we investigate the use of score-based generative models as classifiers for medical images, with a focus on mammographic images.
Multi-View Mammogram Classification with Swin-Transformer
We present a novel multi-view network for medical image analysis, built entirely on the transformer architecture. To address the challenge of combining misaligned data, we developed a "Multi-headed Dynamic Attention Block" (MDA).
Enhanced Mass Segmentation Using Optimized U-Net
We present ConnectedUNets+ and ConnectedUNets++, two novel versions of the Connected-UNets architecture designed for improved breast mass segmentation, incorporating residual skip connections and an enhanced encoder-decoder.
Unsupervised Anomaly Detection for Multivariate Time Series
This project presents Unified Unsupervised Anomaly Detection (U2AD), a novel framework for anomaly detection in multivariate time series using a score-based diffusion model with a unique, time-dependent score network.
Conditional Diffusion Model for Semantically-Aware 3D Point Cloud Generation
This project explores a novel diffusion-based approach to generating realistic and part-aware 3D point clouds. By integrating conditional variables for each point, the model enables fine-grained control over structure-specific features, addressing the irregular and unstructured nature of point cloud data. Our method significantly improves the quality and precision of generated 3D shapes, as demonstrated through extensive experiments comparing guided and unguided diffusion processes.
Open Set Recognition with Diffusion Probabilistic Models
This work proposes a novel diffusion-based approach for open-set recognition (OSR) of radio frequency (RF) signals. The core idea is to train a conditional diffusion model to reconstruct signals from known classes. The reconstruction error is then used as an anomaly score to distinguish between known and unknown signals.