wwydmanski @wwydmanski@qoto.org

https://arxiv.org/abs/2004.09297

BERT got most of the early attention in the topic of foundation models, but from my experience MPNet does at least as good in most of the tasks that I've tested it on.
How does it work? Besides MLM it uses permutation learning for modeling seemingly distant relationships
🔗👇

What I love about online video courses is that most of the time you don't even have to watch them for them to be useful.
A well made one should include a content table comprehensive enough to allow you to study the topic on your own

I got to know this while writing my master thesis: Dewey Decimal Classification is a system used in libraries throughout the world.
It's main idea is that the books should be categorized according to their subject. Additionaly, we can use more digits to be as specific as we need

It's a bit ridiculous that XGBoost and Random Forest are still the best (most consistent in their performance) choices for tabular data.

I'm just getting into Notero pipeline of @zotero + @notion, but I can already see it's going to be a gamechanger.
Shoutout to a great template by @MayaGosztyla

Scientific discussions are a part of academia life which is hard to overvalue
---
RT @MichelleNewman0
I can not say this enough to those starting on their PhD journey. Find your people. Support each other. Knowledge sharing. Bouncing ideas off of one another. Emotional support. Your peers will be as important to your PhD success as your supervisors. #phdchat
https://twitter.com/MichelleNewman0/status/1560544922016616449

"Reviewer 2 symbolizes the peer reviewer who is rude, vague, smug, committed to pet issues, theories, and methodologies, and unwilling to treat the authors as peers."

Well said

TIL: "attention" term was coined a few years before the "Attention is all you need" article.
It first appeared in a 2015 paper from Y. Bengio's group, and it (apparently) wasn't even supposed to become a name of this technique.

Jeżeli wynajmujący mieszkanie sugeruje Wam że nie ma co myśleć o podatkach i umowach, to niekoniecznie znaczy że chce Was oszukać.
Niemniej, warto jest mieć świadomość z czym to się wiąże 👇

https://flathub.pl/blog/najem-bez-umowy/

For context:
https://www.analyticsvidhya.com/blog/2019/10/how-to-build-knowledge-graph-text-using-spacy/

I hoped to read an introduction to knowledge graphs algorithms, but ok I guess

The article being "Convolutional deep belief networks for scalable unsupervised learning of hierarchical representations"

#ml #datascience

https://dl.acm.org/doi/abs/10.1145/1553374.1553453?casa_token=snmcY-M-QNIAAAAA:hNmBmVum-yqXWLq-Oq7xJRInkIWrDzp8fpz1I9-qpmWfiGOS3q1Q4e13TrwoNJcJXvUYi6oqfF_g

Outpainting has come a long way, hasn't it?

I think that the publication about convolutional deep belief networks from @honglaklee, @RogerGrosse, @rajeshrang, and @AndrewYNg is the first article mentioning the problem of completing an image.

Or is there an earlier one?

Quotes from an article by Nicole Hartnett, Adam Gelzinis, Virginia Beal, Rachel Kennedy, and Byron Sharp.

"When Brands Go Dark. Examining Sales Trends when Brands Stop Broad-Reach Advertising for Long Periods"

http://www.journalofadvertisingresearch.com/content/61/3/247

"Larger, growing brands often were observed to continue to grow after advertising stopped, but almost all smaller, growing brands immediately began to decline." (not really coke's case, but still interesting)

Wondering why Coca Cola is still running ads, even though everyone already knows what it is?

"The average observed change in brand sales was –16 percent after one year without advertising, dropping to –25 percent after two years, and reaching –36 percent after three years."