Introduction to Deep Neural Network

- DL Introductory Session:
* Attendees will gain the following skills after the session:
1- Know the history of AI and how DL evolved.
2- Aware of the general theory about how neural networks work
3- Can start practicing with "Supervised Learning" approach
- A portable Windows 64-bit setup will be provided

Transcript

1. DNN Intro .. Why is it important? Now? -- Tarek Eldeeb
2. What is Deep Learning? Read more: https://en.wikipedia.org/wiki/History_of_artificial_intelligence Turing Test Chess/ Games Basic NLP Expert Systems: -Basic Dialog -Medical Analysis 1965: First NN 1996: IBM’s Deep Blue beats Kasparov 2011: BIG DATA 2012: Fast GPU implementations reached human-level accurate! End-to-End Learning
3. Applications of Deep Neural Network • Automatic Speech recognition • Image Recognition • Natural Language Processing • Art Generation! • Advertising & Recommendation • Autonomous Driving • Bioinformatics (+wearables) • .. • .. and the list goes on ..
4. ~80 95 Human Level End-to-End Learning
5. Problems and Drawbacks • No theory, it’s only a good practice! ◦ You can never know: Is it trained well enough? ◦ What are the KPIs for my implementation? • Un-debuggable Errors! ◦ Corrupted/random inputs may be classified with confidence! ◦ Cannot trace the error, it’s only a little percentage! • Hackability ◦ Minor human-indetectable changes may lead to miss-classifications
6. DNN Approaches Supervised Learning • Labelled Dataset (un/structured) • Predict Outcome or future • Classification: ◦ Computer Vision ◦ Customer Retention • Regression: ◦ Weather forecast ◦ Marketing/Adv Unsupervised Learning • No Labels • Find Hidden Structure • Clustering: ◦ Recommendation System • Dimensionality Reduction: ◦ Big-data visualisation ◦ Feature elicitation Reinforcement Learning • Reward System • Decision Process • Skill Acquisition: ◦ Game AI ◦ Robot Navigation
7. Lets Go! What’s Next? 1. Continue the series 2. Setup DNN Tools 3. Try it yourself!
9. Playground .. • Build and try a visual NN https://playground.tensorflow.org • Host and share your notebooks https://www.kaggle.com • Complete Courses: ◦ Top-down: http://course.fast.ai ◦ Bottom-Up: https://www.deeplearning.ai (also on Youtube) • A list of awesome Deep Learning tutorials, projects and communities (here)