Incremental neural network calculation

[Karthik Ganesan]

Hi, I am a very new user to Caffe. As part of my research I am doing an experiment to modify a NN to run in an incremental manner. So for instance with a NN with a single hidden layer of say 6 nodes, I am trying to modify Caffe such that I can run 2 of the nodes in the first 'pass' save these output values, run the next two nodes in the 2nd pass and so on. And for each pass, re-use the values from the previous passes, so the final Softmax layer gets inputs from 2 then 4 then finally all 6 nodes. so the final run should give me the same result as running the original network. 

So my questions are:

1. As a first step, how can I modify the blobs passed to inner product layer so that it only a subset of the full weights and data is passed to GEMM?

2. Is there a way to save the values calculated between 'passes'? So calculate the value for pass #1, save it somehow and reuse it for subsequent passes?

Thank you!