[Lecture] How to build a recognition system (Part 1): best practices
Hello world!
Recognition systems have a lot of practical applications. And many companies need to create such systems to optimize their business processes, not only the industry giants like Google, Baidu, Facebook or Dropbox.
For examples, in healthcare some companies develop automatic field extractor for different patient’s forms, including insurance forms to input relevant data into a database. Other companies are focused on recognizing number plates and so on.
Fortunately, we at DeepSystems have experience in building image recognition systems. But it was very difficult to dive into this field because the lack of information on the internet. Through long research and reading many papers we have developed an understanding of main principles behind creating effective recognition systems.
And today we are going to share our understanding with you and explain how it works in plain language.
This lecture consist of two parts. Part1 covers the concepts of how to design neural network architecture.
Part2 (coming soon) covers the detailed explanation of how to train such systems, i. e. how to do forward and backward phases of CTC Loss. We feel that it is extremely valuable, because it is impossible to find nice and simple explanation of how CTC loss works.
Here is the link to slides.
Part1:
Thank you! Stay with us. Feel free to ask questions in comments.
