Join the 
Investigating various properties of neural networks
This is a tiny project to see what happens if we tweak trivial neural network architecture, like use of different activation functions and subsequently present the results in a readable way.
This is more on the research side than software side, and would need more perseverance than skill.