I am a faculty member in India (Computer Science and Engineering). I am trying to work on deep learning. I have some good ideas to improve existing algorithms. I try these on small datasets which is like 1000-3000 images but this is not sufficient for research purpose. The main problem is I don't have GPU to implement algorithms. The cost of GPU is high and there is no possibility that I or any other member of the institute will have a GPU in the future.

I have tried research on Google Collab and the pro version of Google Collab but it is not working well, I mean it works only for 8000-9000 images, which is not sufficient for research. I have one or two friends but I can only ask them 2-3 times in a month for simulations. One possibility is to work on analysis and less on experimentation.

I want to research deep learning but resources are limited. Please suggest to me how to continue my research without resources.

  • As a side note, you say for "improving existing algorithms" "small dataset of 1000-3000 images" are not sufficient. I do not know the type of data you work with, and I know image classification can need a lot of data, and there's huge dataset nowadays, but you do not necessarily need that much data depending on the problem you tackle. Most DL models applied to medical imaging are trained on small datasets, with hundred to at best thousands images, and sometimes there's still small dataset of 40 or so images. It is not the best, but it is what is available
    – JackRed
    Sep 25, 2023 at 16:55
  • Thanks JackRed for the comment
    – Rma
    Oct 1, 2023 at 5:39

3 Answers 3


As you mentioned, you can't perform your research without the type of resources necessary for the particular kind of research that you are doing.

This is precisely the type of situation in which researchers apply for research grants. They come in various forms and are awarded by different organizations, depending on where the research is to be performed.

For India in particular, it's a bit hard for an outsider to know what exactly the opportunities are, but it looks like there is some development on getting a bigger research funding body in place. There also appears to be the Science and Engineering Research Board which may be suitable for your work, and if you are willing to collaborate with institutions abroad, there are additional opportunities, such as with the NSF. For GPUs in particular, it's a pity that NVidia discontinued their GPU Grants program.

Also check if there there are computing centers run for educational institutions that you can make use of - you will need to find this out on the local level. As an example for such a case from elsewhere, the northern states of Germany have a computing center with GPU servers.

  • Thanks. Do i need to have published research papers to apply for the grants you have mentioned? I am new to this field.
    – Rma
    Sep 20, 2023 at 5:37
  • @Rma The rules of every funding body are different. If you don't have any published papers, you may want to focus on the funding sources that take the specific situation of a higher educational institution having staff doing research on their own without prior publications into account - on the international level, that is uncommon. Your peers at your local institution may have some insight. The cloud service providers mentioned in another answer may indeed be better suited for this situation because it's possible that they give out small grants to those without a publication history.
    – DCTLib
    Sep 20, 2023 at 7:51

Cloud providers often offer grants in the form of compute resources:


Tackle the theoretical problem before even starting doing anything on a computer. 8/9000 images looks fine to me, if you have a solid understanding of your approach.

When you have a solid understanding of the project, you may apply for a research grant, or approach bigger companies (Nvidia has offices in India) asking if they can lend you some GPU time.

You must log in to answer this question.

Not the answer you're looking for? Browse other questions tagged .