This is main repository to group the my course work as part of Coursera Deep Learning Specialization by deeplearning.ai.
- Andrew Ng, Co-founder, Coursera, Adjunct Professor, Stanford University; formerly head of Baidu AI Group / Google Brain.
- Kian Katanforoosh
- Younes Bensouda Mourri
- Course 1: Neural Networks and Deep Learning
- Course 2: Improving Deep Neural Networks: Hyperparameter tuning, Regularization and Optimization
- Course 3: Structuring Machine Learning Projects
- Course 4: Convolutional Neural Network
- Course 5: Sequence Models
- Scan all the courses and review "Heros of Deep Learning" interviews. Choose an area that you are passionate about or investigating and further research interviewee, current and past research.
- For example, Geoff Hinton is considered the grandfather of Deep Learning, so looking at some of the historical context on neural networks especially back propogation algorithms help set a solid conceptual foundation. Also, he hints at his future research interests such as "Capsules Networks" and how he feels they will disrupt traditional convolutional neural network (CNN) architectures.
- DL is about "representation learning" using deep (versus shallow) networks where there is decentralized topology and control. This is very different than human curated knowledge, rules or "feature engineering".
- AN's pedagogy and teaching style as bottom-up was effective for my next intellectual jump for DL;
- NOTE: Starting to only learni bottom-up with Keras or Tensorflow to start is a distraction as you often do not understand what is happening with the DL "black box" versus layering from core programming language constructs and iterating to high level abstractions
- ANN architectures have certain "shapes" and model architecture "patterns". You get more familiar when you try to solving similar problems or tasks and evaluat and visualize outcomes.
- DL research areas such as transfer learning are very exciting (as well as new and somewhat disturbing). A specific example? Reusing cat image recognition model and tasks and actually be able to repurpose for something like radiology diagnosis!