Responsibilities include:
- Co-authored a CVPR-submitted paper on mitigating data exploitation risks in deep learning, targeting real-world impact, and reaching 10,000+ professionals.
- Designed a novel defense method using diffusion autoencoders to secure DL models against data exploitation, processing 100,000+ data points to encode latent representations and obscure data patterns effectively.
- Achieved a 60% reduction in model vulnerability, significantly enhancing security and resilience against data extraction and reverse-engineering attacks such as adversarial training.