Kaggle Joins Google Cloud | Hacker News Reader

Kaggle Joins Google Cloud

blog.kaggle.com

67 pointstranscranial7 comments

Is anyone else worried that Kaggle no longer being independent will cause the diversity of contests to diminish?

I'm not sure that another Fortune 500 company would be as comfortable running their contest on a site that is part of Google (given that google may be privy to the entries or datasets in some special way)

The previous thread speculated the value-add of Kaggle to Google. From Google's side of the announcement:

> We must lower the barriers of entry to AI and make it available to the largest community of developers, users and enterprises, so they can apply it to their own unique needs. With Kaggle joining the Google Cloud team, we can accelerate this mission.

What does Kaggle do for "lowering the barrier to entry?" (And to AI specifically, as Kaggle competitions tend to be more solvable with things like xgboost, without requiring deep learning)

The Kaggle announcement mentions data storage for public datasets, but Google already has BigQuery.

for what its worth, many of the recent competitions have had a heavy computer vision component - an area of great interest in AI research.

Kaggle makes it easier to "enter" AI (by reading about competitions, looking at successful approaches, and eventually implementing your own approaches).

You mentioned that Kaggle competitions can be solved "without requiring deep learning". This benefits Google because Google doesn't need a world where every developer is a deep learning expert. Instead, Google needs a world where developers have a little experience with AI, enough that they can look at Kaggle as a friendly, non-intimidating way to get started solving problems with AI. And fundamentally, more people are interested in "solving problems with AI" than "building AI to solve problems from scratch", which will lead to more customers for Google's AI/ML based APIs.

In my limited past experience, I've liked the fact that on Kaggle, I can understand the marginal benefits of different approaches. I can try a naive approach with simple ML techniques, and achieve a score of say ~60. I can see (after the competition is complete) that an optimized version of my same ML solution can score ~70 from some clever tricks. And perhaps I can see that the winning solution added a lot of complexity to achieve a score of ~75.

It's all in the announcement:

> Making Google Cloud technology available to our community will allow us to offer access to powerful infrastructure, scalable training and deployment services and the ability to store and query large data sets.

So they will basically extend what happened in the Youtube 8M competition (being able to run your code on Google Cloud directly instead of having to spin your own EC2 cluster etc.).

Exactly - at the end of the day, the majority of Kaggle users are potential Google customers. Majority of them would probably buy Google API access outright (for things like NLP, for example). A minority is probably more interested in building their own AI-based processing pipline using Google Cloud resources.

Google's statements are PR fluff. This was an acqui-hire. They paid for a few talented engineers. I expect it to be shutdown in a year or two.

Reply on news.ycombinator.com