I and my students are working on a paper that has associated source code in GitHub. We are planning to change the architecture of the Deep Learning Network they have used, and check to see if we would get better results. Would we be able to do any kind of publication of our work by citing the paper we had used?
Yes: You can publish work that improves upon earlier results.
Whether changing the architecture of a Deep Learning Network will achieve better results is, of course, an open problem. Whether improvements are sufficient to merit publication is also an open problem.
In theory, negative results - e.g., we changed the architecture and didn't get better results - are publishable too. But, such results aren't published as often. I think you'd need a surprising-factor, e.g., if better results could be expected from such a change, but aren't achieved, especially if the reason they aren't achieved can be identified.