This week, we ran a lot of experiments using the matrix factorisation model I was talking about last week. The experiments take a lot of time to run, so it's been a slow week. We tried adjusting the weights a lot. We're aiming to reduce the number of experiments needed, but still get a diverse range.
The consensus seems to be that we get a little bit of a positive result from using transfer learning. We don't know if it is statistically signifcant enouh to work with. It seems like we need to try something new. It's somwhat frustrating.
Next week, I get to present a paper at our lab meeting! This will involve reading, understanding, summarising and explaining an actual research paper, published by people at Carnegie Mellon.