deletions | additions       diff --git a/section_First_experiment_2_classes__.tex b/section_First_experiment_2_classes__.tex  index 9d11435..c7868c7 100644  --- a/section_First_experiment_2_classes__.tex  +++ b/section_First_experiment_2_classes__.tex 
...
file in this case as the labels are integer (randomly sampled for the training set).  With this idea we have finally some results.  In Table \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 can see the found out that it was possible to reach near 100\% training  accuracy obtained form 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,  thefine-tuning for  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 testing phase. 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.