CLIP-guided Diffusion Model for Backdoor Generation in Sensor-based Human Activity Recognition

1. 1Implement a diffusion model to generate synthetic data for human activity recognition models
2. 2Use CLIP guidance to create backdoors in the generated data
3. 3Train a human activity recognition model with the backdoored data
4. 4Evaluate the model's performance and robustness to backdoor attacks
5. 5Apply defense mechanisms to mitigate the effects of backdoor attacks

Machine learning engineers and researchers working on human activity recognition models can benefit from this technique to test model robustness and identify potential vulnerabilities. Data scientists can also use this method to generate synthetic data for training HAR models.

💡 CLIP-guided diffusion models can generate effective backdoors in human activity recognition models with minimal data injection, compromising model accuracy.

🚨 Generate backdoors in HAR models with CLIP-guided diffusion! 🤖