deletions | additions       diff --git a/section_First_experiment_2_classes__.tex b/section_First_experiment_2_classes__.tex  index c7868c7..1552abf 100644  --- a/section_First_experiment_2_classes__.tex  +++ b/section_First_experiment_2_classes__.tex 
...
build 10 files containing the path and the target for each trait in a unique txt  file in this case as the labels are integer (randomly sampled for the training set).  With this idea we have finally some results.  \subsection{Fixed 2C}  The latest batch of 2C experiments have corrected some errors in the prototxt files (traits 7 to 10 were affected) and completed the missing ones (traits 1 to 4). Moreover, in preliminary experiments, we found out that it was possible to reach near 100\% training accuracy by fixing the first 4 conv layers, and letting only the last 3 fully connected layers (InnerProduct ones) fine-tune on the PsychoFlickr data. Unfortunately, the training tended to overfit, showing a test accuracy graph with an "elbow" path. To reduce overfitting, we increased the weight decay parameter dramatically, in order to avoid any "neuron specialization" (neural units with high weights values). Together, fixing the conv layers and raising weight decay, we achieved the best 2C results so far. See Figure 1, where the self traits are still unimpressive, but the attributed ones clearly show predictions above chance.