In this thesis, we have stduied on Backpropagation and some of it‘s variants.One of thevariant’s of Backpropagation is Rprop. We have introduced two modification in Rprop mini-batch version. First one is making mini-batch uniform which means that in every mini-batch, number of sample from different class should be same and the ratio of these numbers should be equal to the class ratio (ratio of the numbers of data present from every class) of whole data. Second, we have introduced a changing momentum. Our modifications have improved the clasification accuracy for large datasets. We have compared the result with Rprop emperically. Numer- ical evidence shows that our modification outperforms Rprop in case of large datasets. (github link)