14 13th February, Tuesday

I've been working towards improving the black box attack today, trying to map 3 similar images to the same output for each synthetic data point. Although I have been successful in doing this I have found that the increase in performance has been only very slight. Tomorrow I will be looking to do some more write up and will think a bit more about potential ways of making the attack more practical.