GSoC Project Idea 3: Web-based GUI for deep learning continuous workflows

maven
json
jquery
java
rest

#1

Our laboratory is focused on Electroencephalography (EEG) and event-related potential (ERP) experiments. The processing of EEG/ERP data consists several steps: Data reading, channels selection, pre-processing, classification and results visualization. Individual steps are implemented in various signal processing methods usually hardcoded in a computer program.

In order to make the work with the signal processing method as much flexible as possible in the previous year of GSoC we have developed a workflow designer. It is a graphical user interface for designing signal processing workflows. Workflows are built from component blocks that have configurable inputs, outputs and properties. The blocks can be combined and rearranged at runtime without making any modification to code.

The current design follows the standard workflow paradigm supposing that one block starts immediately after the previous one terminates. However there are use cases that require continuous running. E. g. once data is streamed by some streaming protocol (e. g. Lab Streaming Layer) it requires a continuous flow of data through all the block and the results in a result block are continuously updated.

Aims: The aim for an applicant is to modify the existing workflow designer to enable also running of continuous workflows.

We will provide the existing workflow designer, existing methods implemented in the workflow designer, workflows itself and testing data. The workflows are described in JSON format and can be imported and exported to the workflow designer server (a graphical tool for designing workflows). The signal processing methods are java codes that are annotated and processed by the workflow designer. The student is requested to extend this annotation feature and JSON format to enable creating of continuous workflows. Also redesigning existing workflows and modifying existing blocks or creating completely new workflow blocks needed for continuous processing is a part of the task. Finally, testing on real data is needed.

Skills: Java, Maven, JSON, REST, JQuery, Javascript, CSS, HTML, GitHub, optionally Hadoop

Mentor: Petr Je┼żek, U of West Bohemia, Czech Republic.

NB: A small change was made to the project title on March 8.