Design and Development of a Graphical Based Machine Learning Pipelines Designer Web Application

Nowadays, lots of machine learning algorithms have been invented, giving scientists and engineers a rich toolbox of algorithms to choose from. However, the choice is becoming a difficulty for data scientists, as each algorithm has its own unique set of parameters that require proper tuning depending on the use case. Also, the data fed to each of the algorithms has to respect a given data structure for compute optimizations. With these constraints, data scientists have to spend hours writing code, training, testing, and evaluating machine learning models in order to find solutions that satisfy their use case requirements.

In order to minimize the time and cost of developing machine learning models, Integration Objects aims to provide a code-less editor for designing and executing machine learning pipelines. The editor should be web based providing ability to design a directed acyclic graph that represents the machine learning pipeline. On top of the basic features of the editor, the system should include an intelligent recommendation sub-system which recommends blocks to use at each step. The recommendation system eases the process of choosing algorithms for non-expert users.

• Perform a state of the Art study.

• Architect and design a solution.

• Implement the solution or a proof-of-concept.

• Perform tests, benchmarks and provide perspectives.

Deliver a finalized product with source codes, and documentation:

• Mathematical models & Heuristics.

• Source codes.

• Machine Learning Models.

• Test & development tools used during the internship.

• Documentation: project report, design documents, and additional documentations.

Recommendation Systems, Supervised Machine Learning, Deep Learning, Deep Re-enforced Machine Learning, Hyper Parameter Optimization, Heuristics.