occam (@occam) profile picture

Given two equally predictive theories hack the stupid one to pulp.

Joined May 2008
Photo by occam (@occam)Photo by occam (@occam)Photo by occam (@occam)
The state of free speech in America - read every word. nytimes.com/2020/10/13/magaz…
Registration is free.
Apachecon goes virtual and the Universal Recommender will be there. Check out the Mahout track: apachecon.com/acna2020/track…
Yay. Tracking every place I go forever and pestering me to rate businesses is, at very least, creepy and annoying. When do we get GMail that doesn't track everything else?
Coming soon to @googlemaps, when you turn on Incognito mode in Maps, your activity—like the places you search or get directions to—won’t be saved to your Google Account. Just tap from your profile picture to easily turn it on or off. #io19
If you've seen Tensor diagrams this is a quite nice explanation using examples from linear algebra. If, like me, you grok LinAlg from the old days, this might help upgrade your brain. math3ma.com/blog/matrices-as…
1
Asked by the NYT why one would take social media seriously, Ricky Gervais responded perfectly; "You don’t. If you ignore it on Twitter, it didn’t happen. It’s like going into a toilet stall and arguing with graffiti." nytimes.com/interactive/2019…
1
1
Now a good explanation for the "never enough data to make Deep Learning perform" effect and how to turn NNs trained on truly huge data into feature extractors for use with more traditional ML. Highly recommended article! basilica.ai/blog/the-unreaso…
Am I the only one or does something sound wrong about Quantum Mechanics or at least how we interpret its weirdness? Nope not alone... quantamagazine.org/frauchige…
1
Awesome and they have the Behavioral Search idea but no idea why they complicate it by doing matrix factorization first. We just use the CCO model to augment the item search index. The query is terms and user-item vectors-et voila.
1
Bought 1g inet from #comcast got 400m, downgraded account to pay for 400m, got 250m. Same wires, SW, modem, all #xfinity #comcast HW. Shame on Comcast for rather obviously throttling inet speeds. Can't wait to hear how they explain that away :-)
ML/AI algorithm explanations sometimes meet blank stares, people who took linear algebra but don't recall much of it. Here is a great refresher that even touches on the mystical tensor: “Basic Linear Algebra for Deep Learning” by Niklas Donges towardsdatascience.com/linea…
1
2
The Universal Recommender v0.7.3 released. Now with cross-recommendations; "viewed this bought that" Use it to help a user find the target of a browsing or search session. groups.google.com/forum/#!to…
3
6
People, please think before using OAuth2 for single sign-on. Read what you are giving access to. I have virtually given up on SSO and opt for site specific ones instead. thisisinsider.com/google-all…
Cambridge Analytica may truly be bad guys but the root of the problem is that OAuth2 has no trust mechanism. Remember the GDoc debacle? Google blocked the fake OAuth2 GUI after the fact. There is no trust required to use OAuth2. How many abuses before we see the obvious flaw?
1
1
Close, more like, "What's everyone's history of drink orders?" nitter.cf/tniebnieb/status/96721…
Replying to @occam
@occam Your Universal Recommender walked into a bar. The bartender asked, “What would you like to drink?” UR replied, “What’s everyone else having”?
Our paper on UMAP, a faster alternative to t-SNE, is now up on arXiv! The paper provides a more detailed account of the theoretical underpinnings of the algorithm, as well as performance benchmarks. arxiv.org/abs/1802.03426
1
PredictionIO graduates to a top level project at Apache. The Mahout based Universal Recommender runs on it. globenewswire.com/news-relea…
1
5
6
A personal story of obsession and the gravitational attraction of Apache thru Mahout and PredictionIO blogs.apache.org/foundation/…
2
5