The Future Of Maxpool In The World of Sparse Attention

Our slack community Maxpool

Okay, the mention of sparse attention in the title is just an unnecessary pun between NLP's sparse attention, and how difficult it is to get people’s attention 🤪

No, this article is not about any deep learning topic like pooling. This is about our newly launched slack community Maxpool. Earlier we had a WhatsApp group of NLP engineers which ran out of limit(256) and it was time we scaled it in the right way.

Today I got intrigued by reading this blog by Microsoft on how to build communities - the why and how. What came out of it is the goals and metrics part!

Let’s discuss what we want Maxpool to be.

  • Why: As we grow older, our definition of friends change and we look forward to connecting with those who share a similar and equal amount of passion.

    A lot of the friends I made in the last few years came from my previous online interactions. When I couldn’t get an answer to the problem from my colleagues, I got them from my new community friends. Some of them include Ankur Bohra, Ankur Pandey, Goku Mohandas, Abhinav Verma, Phillip Vollet, Aditya Malte, Manas Ranjan, Ramsri Golla!

    This year Aditya and I worked together on an NLP problem which led to research and a publication too! I have never met Aditya in real life. I haven’t even seen his face live. (This is now feeling weird 😅)

    So the hope with the community is to share ideas without limitations of geography with like-minded people.

  • Audience: We want people from around the world to come together. The reason being work varies across cultures.

  • Goals: Discuss practical aspects of building ML and get good at it. We expect our members to have a basic level. This group is not for beginners of ML.

  • Principles: There will be situations when we have a difference of opinion or somebody might be wrong or we are wrong.

    I have been in all the situations and learnt that the best course of action is to give the benefit of the doubt. Let’s be respectful to everyone.

    Thank anyone who helps you solve your problem. I cannot stress this part enough. Thanking is the social currency that drives the community.

  • Metrics: We don’t want to optimise for the number of people but rather see engagement as the true metric. Down the line, we might also consider number of community talks or even open-source contribution.

  • Leadership team: There are no fixed community leaders right now. It’s a flat hierarchy. People who feel passionate about a niche problem like solving Indic NLP or Search or deploying transformers should come forward and help others. As they engage more with people they will naturally emerge as leaders. This applies to me as well.

  • How to grow: The best way to grow is by having meaningful detailed discussions. If we are solving great problems, the reputation will automatically take care of growth. I have a feeling we have enough critical mass/core people to sustain. Occasionally I will invite people I admire.

Cheers to making new friends 🥂

Come join Maxpool - A Data Science community to discuss real ML problems