Deep Learning about deep learning

To develop a new model from the existing models, it is important to know the flow of logic and semantics of the code. In this project; we have decided to use deep learning technologies to design a model which is capable of analyzing the given code and its behavior from two perspectives. First one is static call graph which represents how the deep learning is done at every stage w.r.t source code. It is generalization of the source code with entry points and exit points. Here all the combinations of paths from entry point to exit point are shown. In contrast to second one i.e. dynamic call graph it will show one path from entry point and exit point. This path is decided by the input program given by user. Static and dynamic call graphs are best in their own way in terms of exposure they can value add to application users.

Furthermore for any organization, it will be interesting to adapt to a system which can train people on open source projects quickly. The combination of virtual reality world and call graph scan be proven best for the user’s understanding. Users should have an application where they can observe these call graphs for any open source project in virtual reality world. With the motion of their hand like may be a swipe action, the application will reflect all the call graphs for given user program to human eyes. User should be able to see the various entry points, middle ones and the exit ones. Based upon the performance and the shortest path between an entry and an exit point User can easily learn and start making decision on what can be the best program for any particular goal.

The architecture diagram shown in Figure-2 explains how our intelligent model works. The model takes a deep learning model source code as input. The source code is analyzed to determine functionality in terms of package structure, module, class structure, and module structure. The code with similar functionality and structure is grouped together into a cluster to reduce complexity. The next step will provide an interesting visualization to these clusters in form of static and dynamic call graphs.

Video:

https://www.youtube.com/playlist?list=PLPkAMj4WkRbf_fZ_huXT1uH973WRmAkLr

Presentation

https://drive.google.com/open?id=0B1ZPKLB98Nz2R0xGOHFWM0FzVW8