Dear Dr Thuseethan,

I came across your question and thought the following link (to an online tutorial on convolutional neural nets) might be of interest to you:

http://ufldl.stanford.edu/tutorial/supervised/ConvolutionalNeuralNetwork/

Yours sincerely,

Rainer

Many thanks Rainer. I would appreciate your time.

Regards,

Thuseethan

https://www.youtube.com/watch?v=oYbVFhK_olY this tutorial series introduces CNNs fairly effectively and provides an introduction in the context of Tensorflow, which will probably be the best tool for the job. Takes about 5 hours to do justice with the tutorial, which isn't too bad considering it has a considerable programming component to get you coding away.

Dear Dr Thuseethan,

it depends on what you mean with "from scratch": if you're saying that you're unfamiliar with the theory and implementation details of CNNs but want quick results and easily perform a few tests to see if CNNs are a viable tool to solve a practical problem you have, I would suggest you to employ one of the standard Frameworks:

• TensorFlow – https://www.tensorflow.org/  – developed by researchers and engineers from the Google Brain team within Google’s AI organization, it comes with strong support for machine learning and deep learning and the flexible numerical computation core is used across many other scientific domains.
• DLib – http://dlib.net/ – C++ toolkit containing machine learning algorithms and tools for creating complex software in C++ to solve real world problems.

There are lots of tutorials for both frameworks and also many, many research projects using them and describing their results. Eric W. Bridgeford above already suggested a series of videotutorials on TensorFlow. If you search on YouTube, you will also find plenty of material on DLib: https://www.youtube.com/results?search_query=dlib+image .

On the other side, if with "from scratch" you wanted to say "what strategy should I employ to apply CNNs to a problem I have?", then it becomes more difficult. Rasmus Rothe provides an enlightening but still superficial discussion of this issue in his blog: https://medium.com/merantix/applying-deep-learning-to-real-world-problems-ba2d86ac5837 Here is also a good analysis of historical (2012 - today) material: The 10 Deep Learning Methods AI Practitioners Need to Apply - https://towardsdatascience.com/the-10-deep-learning-methods-ai-practitioners-need-to-apply-885259f402c1?gi=6de527dedc9d