Data Sets

One of the hardest parts of starting out with machine learning used to be finding good data. The UCI Machine Learning Repository was the most comprehensive site, and still is a great resource. The site currently has over 440 data sets. Later,  Kaggle  and similar sites became popular. Google AI is becoming a comprehensive resource for both tools and data. As of this writing, Google has made available over 50 data sets for machine learning, and I'm sure more will come in the future.

It's an exciting time to get into machine learning!





Comments

Post a Comment

Popular posts from this blog

Google is watching

Image
Google is watching you, but you knew that already. Several years ago I saw a pop-up like this: It scared the heck out of me so I just exited out of Chrome. It popped up again today and I hit 'I want to play' and this popped up: Wow. I have heard that this is how Google recruits developers. I don't know if that's true but I exited out of this. Maybe next time . . .
Read more

Your Brain on Code

 Is reading code like solving math problems? or is it more like reading in general? Neither, it turns out. MIT researcher Eveline Fedorenko and team placed coders in a functional magnetic resonance (fMRI) scanner, and showed them code in a programming language in which they were proficient, then asked them questions about the code.   Although coding languages have many things in common with human languages, like meaning units and syntax, the fMRI showed that reading code did not significantly activate the regions of the brain associated with language. These language regions of the brain, including Broca's area, are primarily in the left hemisphere.  Instead, the distributed parts of the brain known as the multiple demand (MD) network were activated across the brain. These regions had previously been shown to be activated for complex mental tasks like solving math problems, or doing crossword puzzles. However, the regions activated for the coders were not identical to...
Read more

Ethics in AI

This article from Quartz came into my phone's news feed this morning. AI conferences are beginning to think about ethics. Finally. Specifically, the premier machine learning conference NIPS requires a broader impact on society statement for presented research. One of the top natural language processing conferences, EMNLP, will now reject papers on ethical grounds. Whether these moves actually keep AI ethical or just push the dirty parts underground remains to be seen. But we all have a voice in keeping each other in check.  When I was a PhD student, our advisor had us sit in on talks by prospective faculty and arranged for the prospect to meet with just the students. I remember one young man boasting of one NLP project after the other, each more horribly intrusive on innocent lives than the last. All of these done in internships with a major computer manufacturer. Finally, when we got to questions, I asked him, "Can you describe any project you have done which makes the world...
Read more