Skip to content
No description, website, or topics provided.
Jupyter Notebook CSS Python HTML JavaScript Shell
Branch: master
Clone or download
Fetching latest commit…
Cannot retrieve the latest commit at this time.
Permalink
Type Name Latest commit message Commit time
Failed to load latest commit information.
CNN
Cleaning_images
Data_extraction
NCS -2
Optimised - MLP
app
.DS_Store
LICENSE
README.md

README.md

ESA - Summer of Code in Space (SOCIS)

Portable AI solutions for test results analysis - (B12)

Scope

The main goal of this project is to train, configure and deploy two deep neural network (a perceptron and a Convolutional Neural Network - (CNN)) in a portable VPU to validate its performances, behavior and integration capacity. The proof of concept will be based on pattern recognition on test results.


Duration

The project will take place between June and September 2019 and it is divided in 6 sprints of 2 weeks each.


Sprints

The sprints will be updated when the actual sprint is finished.

  • Sprint 1:

    • Technology studying and installing
      • Python 3.7
      • Tensorflow/Keras
      • Jupyter Notebook
      • Docker
    • Input data analysis
      • Analysis of input data
      • Understanding of the data and its categorization
      • Plan how the data will be used
      • Transform the data (if needed) to be used in the network.
  • Sprint 2:

    • Implementation of perceptron
      • 4 point solution (ON Graphs)
      • 4 point solution (OFF Graphs)
      • Reduction of points solution (ON graphs)
      • Reduction of points solution (ON graphs)
    • Analyze and optimize the networks
      • Analyze the 2 different solutions
      • Trade off
      • Optimize topology and parameters
  • Sprint 3:

    • Create docker image with the perceptron and the REST API
    • Docker image deployed in Thales environment
    • Analyze input data to convolutional network (graphs)
    • Study different convolutional models
    • Trade off of the study made
  • Sprint 4

    • First Implementation of CNN
      • Download the CNN model selected in the analysis phase
      • Train the network with a set of images
      • Analyse the time the network takes in training and the training results
    • Install CNN in VPU
  • Sprint 5

    • Install CNN in VPU
      • Download a big convolutional model (e.g. RestNet50) and install in the VPU
      • Execute the CNN and compare the results of executing the CNN with or without VPU. Generate a report
      • Test the VPU in MacOS and CentOS and Windows operative systems
  • Sprint 6

    • Write user manual of perceptron
    • Write user manual of CNN
    • Write user manual of VPU
    • Upload final source code and documentation to TAS-E Github for SOCIS project
    • Hold final DEMO meeting with TAS-E CTO and SW&Ground lines including AI specialists

You can’t perform that action at this time.