[Pycon] [new paper] "Albertus Kelvin" - Source Code Generation Based On User Intention Using LSTM Networks

[info a pycon.it]

Title: Source Code Generation Based On User Intention Using LSTM Networks
Duration: 45 (includes Q&A)
Q&A Session: 0
Language: en
Type: Talk

Abstract: Computer programming is the process of constructing executable code from fragmentary information describing some functionalities of a program. When it is done by a machine, the process is called as automatic programming. In this case, the only thing that the users have to do is to provide the description of the problem with natural language, starting from the specifications for the input and the output as well as their constraints until the main body of the problem, and then the machine will provide the desired source code as the solution.

In the field of software engineering, such system has been the dream since a long time ago as it provides automation for the process of understanding the problem and searching for the relevant solution. It allows the programmers to focus on more proactive, creative and strategic activities.

This talk is based on my experiences in developing such automatic programming system. There are two goals of this talk. The first one is developing a simple automatic programming system that receives a problem statement written in natural language and then generates the relevant source code. While the second one is examining the performance of several variants of LSTM networks, such as the Encoder-Decoder, Normal Sequential LSTM, and the combination of both of the models via two experiments. In addition, the development process used Keras framework with Theano as backend.

Since the topic primarily relates to Deep Learning (LSTM Networks) and Natural Language Processing, any knowledges on those subjects would be useful. Moreover, any knowledges or experiences in using Keras framework and Theano would be useful too.

Tags: [u'nltk', u'code generation', u'scikit-learn', u'Data Mining', u'Machine Learning', u'nlp', u'Python', u'computer-science', u'Keras', u'word_embedding', u'programming-magic', u'text-analysis', u'Deep-Learning', u'Theano', u'data-science', u'Text-Mining', u'Big-Data', u'computational-linguistics', u'recurrent-neural-network', u'Artificial Intelligence']