Learning how to learn deep learning（学习笔记）

Learning how to learn deep learning

You are a software engineer who works with code every day, building complex things, turning business requirements into application logic and shipping mostly on time. You have tried your hand at different programming languages and have chosen your primary weapon. You are confident enough in what you do and you are ready to learn something entirely new, something to give you new powers and to keep you relevant in the profession.

1. Dont  fret
   数学很重要，但不要忘了你个coder也可以用code的方式来训练模型。
2. Remember it's still code
3. Find a partner in crime
4. Avoid cognitive overload
   too hard or too easy is not good.
   1. The trick is to bite precisely as much as you can chew.
   2. Sticking to one course and not trying to learn things in parallel. (People, after all, are notoriously bad at multi-tasking.)
   3. Pace yourself, avoid multi-tasking and take small steps.

       (作者推荐的学习顺序）

      1. fast.ai
      2. the Andrew Ng’s latest Coursera offering
         It consists of five courses: from the introduction to neural networks and deep learning, through discovering the convolutional neural networks (that are essential to working with image data), to sequence models (speech recognition, music synthesis, any time series data).

       

5. Set your sights
   pick areas where using deep learning techniques provides results that are more satisfying to you.Working with something you can relate to will keep you motivated. You need a feedback loop, a way to get tangible results from your experiments.

   1. If you are passionate about visual arts (cinema, photography, video, fine arts), dive into Computer Vision. Neural networks are used to classify objects on images, highlight areas of interest (anomalies on MRI scans or pedestrians on the road), detect emotions or age on portraits, transfer artistic styles, even generate original artworks.

   2. If you are more into sound, you can compose music with neural networks, classify genres and recommend new tracks as Spotify does. Or, you can explore voice style transfer and pretend to speak in another person’s voice.

   3. If you are into video games, you definitely should take a look at Reinforcement learning.

   4. If you are passionate about user experience and customer support—look into Natural Language Processing and сhatbots, this way you can automate (to a certain extent) interacting with your customers: deduce intent from a message, categorize support tickets, offer immediate answers to most common questions.

6. Be competitive

7. Stay in the loop
   The blessing and the curse of deep learning is the pace at which the field is evolving.
   加入一手的俱乐部得到最及时的消息。有时候参加培训学校也挺好，记得筛选。

8. Use your programming chops
   所以啊！编程是基础！！

9. Finally, brush up your math

   1.  ML's 'bible':
      “Deep Learning (Adaptive Computation and Machine Learning)” by Ian Goodfellow, Yoshua Bengio and Aaron Courville, known as just the Deep Learning Book.

 

ps:
Useful resources:请见大佬的网页。