In this class we studied 3 aspects of Deep Learning by implementing research papers on the following properties of deep learning.
Each of the 5 assignments required us to read research papers and implement them at 3 different levels of abstraction.
This was achieved using the TensorFlow framework using Keras at the high level and NumPy at the low level.
Security
Interpretability and Accountability
Privacy and Fairness
Learnings :
Adversarial manipulation of
deep learning models
Techniques to enhance robustness
Papers :
The Limitations of Deep Learning in
Adversarial Settings
Towards Evaluating the Robustness
of Neural Networks
Learnings :
State-of-the-art methods to enhance the transparency of deep learning models.
Papers :
Influence-Directed Explanations for Deep Convolutional Networks
An Evaluation of the Human-Interpretability
of Explanation
Learnings :
Debiasing language models
Basics of differential privacy
Papers :
Feature-Wise Bias Amplification
Man is to Computer Programmer as Woman is to Homemaker? Debiasing Word Embeddings
