Convergence Problem while training

I have developed a Deep CNN network and trained with gray scale images (in total 7800 Images). While training, some 50 epochs helps only to increase the Acc by 4% while training loss decreased by very small amount like 0.6858 to 0.6325.

I am really wandering on hyper-parameters learning rate, weight_decay value. I trained the model almost for 100 epoch, but convergence is looks very crucial, and variance being higher while training for higher epoch.

Any kind of suggestion will be welcomed ? Thank you.


I have moved this post to the DLS Course 4 category as mentors of the CNN course might help you out here with this query.

Make sure if you have any course-specific doubts, explore the specialization category and post in the relevant course subcategory as course-specific mentors are actively answering the queries there and the threads in the general discussion might not come on their radar and remain unanswered at times.