Adversarial Example Generation & CNN Analysis
Budget: £10 – £20 GBP
I'm working on a facial recognition project using a custom CNN architecture trained on a Kaggle dataset. I need help generating adversarial examples using the Untargeted Fast Gradient Sign Method (FGSM) on the test set.
Key Requirements:
- Use the FGSM to create adversarial examples
- Incrementally plot model accuracy against epsilon values from 0 to 0.3 in increments of 0.05
- Visualize how robustness changes as epsilon increases
Ideal Skills:
- Proficiency in Python
- Experience with Convolutional Neural Networks (CNN)
- Familiarity with the Fast Gradient Sign Method (FGSM)
- Ability to produce clear, informative visualizations
I need to clearly see how the model's robustness changes as it encounters more noise.
Key Requirements:
- Use the FGSM to create adversarial examples
- Incrementally plot model accuracy against epsilon values from 0 to 0.3 in increments of 0.05
- Visualize how robustness changes as epsilon increases
Ideal Skills:
- Proficiency in Python
- Experience with Convolutional Neural Networks (CNN)
- Familiarity with the Fast Gradient Sign Method (FGSM)
- Ability to produce clear, informative visualizations
I need to clearly see how the model's robustness changes as it encounters more noise.