My Experience with Deep Learning

After Completion my sabbatical period in Bennett university , Met Lot of intellectual people working on several domains but all are trying to implement those problems using Deep Learning,

few of the best research fellows i met are Mohit,(doing phd in bennett), Dr Anand natarajan (Professor in Sri Vidyanikethan),

Mohit sir taught me how to implement the scripting language(python) in real time applications/problems. Few are changing the names of the images, converting bmp images to jpg(any format of image to jpg), moving files to directories with script, later on working with him, i observed that its a matter of solving the problem with the existing solutions which are available in internet, He taught me how to implement cnn in keras, gave few useful links to learn. He supported a lot in plotting the graphs, working with Dgx, comparison of results with ML algorithms although i am not familiar with it.He only made our CNN architecture to classify the forgery images

Dr Anand Natarajan sir had a core knowledge in the field of deep learning,especially neural networks. He is using many neural networks to solve the problems, his domain is to find the breast cancer cell in the tissue marked with the blue/purple color. He showed me how to do segmentation, How to do actual work in the deep learning. One of the best suggestion he gave me is visualize each and every layer of Neural Networks output then you understood where the problem appears and what to do. I am trying on it, since i received that suggestion but due to technical reasons i am unable to cop up to it but i studied few articles about visualization of neural networks output.

After all one month experience, i found out that deep learning is a mixture of Neural Networks, Machine Learning, Image Processing, Natural Language Processing, and finally Calculus, probability, Statistics, this is going on. The best part i found is it is a diversified area of knowledge and applications, In bennett sabbatical one of the research fellow is applying cnn for control of flux in motor.

In deep learning upto my experience,(i hope it may be wrong), Hyper parameters plays a crucial role of our output,finding them is a big task,

following is the workflow of deep learning

1. Prepared a dataset for training
2. Selected a network to train
3. Trained the network
4. Tested the trained model in a training environment
5. Deployed the trained model into an application

Later will add soon…