Topics to learn in Neural Network [closed]

Question

Closed. This question is opinion-based. It is not currently accepting answers.

---

Want to improve this question? Update the question so it can be answered with facts and citations by editing this post.

Closed 5 months ago.

Improve this question

I have recently started learning Deep Neural Networks and was going through the tutorials online. Everywhere I saw that the topics post Image classification using CNN is a little hazy. No one seems to follow a guide as to what topics should one learn after learning image classification.

I would really appreciate if someone could guide me so as to learn advanced topics and become good to get started with research in this field.

The topics i have covered so far are:

• Basic Machine Learning
  • Linear Regression
  • Logistic Regression
  • SVM
  • Decision Trees
  • Optimization Algorithms
  • Regularization Methods
  • Different Loss Functions
  • Different Metrics during Training
  • Feature Extraction
  • HyperParameter Tuning
• Basic Neural Network
  • ANN
  • CNN
    • Image Classification
    • Image Augmentation before training
    • Visualizing Filters

I have started learning Object detection using CNN but I have this feeling that I am missing some topics which can strengthen my foundation.

Thanks in advance. Any guidance is appreciated.

Answer 1

You will observe a lack of blogs because post above points Libraries don't have a very simple implementation. So, need to get your hand dirty yourself.
I am assuming, you are thinking to move into Computer vision.

- Object detection
- Object segmentation
- Object recognition
- Try all of the above using some custom data.
- Face detection/Recognition
- Get the knowledge of Classical Image processing. You don't need DL every time.
- Learn OpenCV
- Video data analysis
- Follow latest research paper Or Commercial products on CV

Jason Brownlee and Adrian Roserbrock have suggestions for CV on their blogs.

To add to this I would like to point to a very beginner friendly blog which helped me to get an overall understanding -> blog
This covers the basics of computer vision from where you can have a better understanding of how to use CNN in image processing.

Answer 2

If you want to start with research, as you say, I would make sure not to skip the theory and underlying mathematics of machine learning. Knowing what model and hyperparameters to choose for a certain problem and being able to implement this quickly is very useful in corporate settings, but research mostly requires a deep mathematical understanding of what is going on.

In my opinion the book 'deep learning' by Goodfellow and Bengio has a good structure that allows you to get familiar with the mathematical foundations of deep learning. When you understand this, you will be able to read and understand state-of-the art papers if you want and do your own research. You can find the book for free online I think.

Answer 3

At least according to Andrew Ng CNN syllabus, you're doing it alright.

It is natural to study object detection after image classification.

Perhaps what you want to do is learn about transfer learning, it is one of the most important ideas of deep learning. You might also want to focus on the specifics of neural networks training:

• Dropout
• Batch normalization
• Learning rate finders
• Cyclical learning rates

For that, I recommend fast.ai deep learning course.

You might also benefit from learning pytorch, keras, and fast.ai libraries, as using the tools is as important as theory.