Archive for May, 2018

Although AI, Machine Learning and Deep learning and actually sub parts of each other but each one of them has a specific use for specific requirements. Generally, the AI is how a robot is taking a decision on making an action, Machine Learning on prediction based on the historical data and deep learning is mainly about making the Computer to understand the content of a picture, also based on the data provided. Data is the fuel of any AI project. In this blog post I will be focusing on Deep Learning.

If you have worked before or if you have learned about Neural Network, Deep Learning is the next level. Neural Network is mainly composed of input layer, output layer and the middle layer. So, in general the input layer is the data entered in the application and the output is for the data getting out the system. Now for the middle layer, it is mainly for the process that will be applied on the input data.

Now Deep Learning is the main Concept as the neural network but the main difference is it has multiple middle layers between the input and the output layers. In the following link you can find a simulation of Neural Network for an XOR function with all the math equations for the Neural Network available in the webpage whether it is a forward pass or backpropagation pass. The simulation also shows you the different values that are being changed to optimize the network for the best possible output based on the training dataset provided. Go ahead and play with the simulation. Try it out with one step forward, one iteration forward and 1000 iterations so you can see the difference in the weights that are being changed based on the data provided.

If you wanted to start with Deep Learning, some cloud vendors have already implemented their own deep learning toolkit that you can use like the CNTK from Microsoft (Code Sample), TensorFlow (Code Sample) from Google. There is also multiple platform designed for specific services that can run on top of these toolkit like the Keras platform which actually can runs on top of CNTK, TensorFlow, Theano and others. Caffe that is a newly designed deep learning designed by the

Remember the most complicated issue for any deep learning or data driven projects, is mainly the dataset, this part will take most of the time of the project, to gather the right data, normalize the data and cleanse the data. The implementation for the project won’t take much time from you as much as the work on the data itself.

You can find different projects that has been implemented on the different deep learning toolkit in the following links:

Also you can check the following link for the different deep learning projects that are actually dominating and will dominate this year.

As there has been a lot of progress in the business acceptance of the AI and the deep learning lately and it is being continuously pushed harder by the technology vendors, Ethics to work on such platform is so important to work on these services.

In the following YouTube video, a great way Microsoft is presenting the power and the impact of the AI. It all depends on how you will use this unlimited power.

So basically, why Deep Learning Now? with the explosion of the Cloud Computing providers and the different vendors that are providing their solution on top of it. the cost of running such an environment is becoming increasingly cheaper than before, especially with the newly design chips that help running such solutions from NVIDIA GPU or vendors’ designed machines like the TPU from Google and FPGA from Microsoft. Second main reason is the availability of the Data, with the sophisticated data that are being generated by the business different applications or by the Social Media, Internet of things… etc.