If you look at the first picture, you may ask yourself a question - why do we need to treat the projection from h to z separately?

It turns out that it's a side effect of contrastive loss! z vectors contain only the bare minimum info needed to recognize the similarity. (4/4)

Show thread

In SimCLR we are going to show our neural network two different augmentations of our sample and train it to recognize that this is one and the same thing.

This way we're training the network to look past the augmentations and focus on what's important on the image. (3/4)

Show thread

SimCLR is an algorithm which depends on a common operation: augmenting the data.

However, in contrast to traditional methods, we don't use augmentation for pretending that we have more data. To the contrary! (2/4)

Show thread

Do you remember how autoencoders can be used for pretraining on unlabeled data?

Now, imagine a method that can do it much better, much faster, without the pesky decoding part.

Welcome to a short thread on SimCLR 👇 (1/4)

RT @Charlie43375818
1/ We've deployed a new model for abstract summaries in Elicit! We trained our new summarization model using reinforcement learning from AI feedback (RLAIF) similar to @AnthropicAI constitutional AI method.

To try it, go to elicit.org and type in a query!

RT @_akhaliq
InstructPix2Pix Learning to Follow Image Editing Instructions is now part of @diffuserslib

try out the @Gradio demo on @huggingface Spaces and check the diffusers docs to run the model in a few lines of code

demo: huggingface.co/spaces/timbrook
docs: huggingface.co/docs/diffusers/

RT @valentynbez
The Russian Orthodox churchc in remains active, spreading the message of patriarch Kirill and failing to denounce the .

It's unacceptable, and here is why:

The church's stance on war (translated from Russian), in chronological order 🧵 (1/5)

wwydmanski boosted

I STILL have a spreadsheet full of #Steam #game keys (all legit!) and giving them away isn't just for Christmas any more.

I'm NOT doing this to get new followers. You DON'T need to follow me to get a key. This is about generosity when the world could really use some. The only thing I'd ask you to follow is the Wheaton rule: "Don't be a dick".

To avoid keyfarmers and resellers, I have some guidelines, though.

1. No new accounts. Sorry. You've got to have been here on Mastodon & obviously human; ie. no keyfarmers (If I don't know you, I'm checking profiles).

2. One game per person, but please pick your top three preferred games that you want to play.

3. First in, first served. If your first preference is gone by the time I get to you, that's why there's a 2nd & 3rd preference.

4. *DM* me with the row number, name, and preference order of the three games (my DMs are open)

Green: My keys
Blue: Keys donated by @McCovican.
Orange: Keys donated by @tojosan

Game listed multiple times = multiple keys.

We've given away ~100 games now, and I'll be adding new keys as I get them.

If you don't mind boosting this post; this is my small way of trying to make the world & Mastodon a nice place to be.

docs.google.com/spreadsheets/d

#gaming #giveaway #MastodonGiving #Steam #PCGaming #Games

🤔
---
RT @LKierznowski
1/9 Rok temu MEiN uruchomiło "Perły nauki" - program grantowy dla studentów, będący bezpośrednią kontynuacją dotychczasowego Diamentowego Grantu, z naborem do 14 lutego 2022.

I co? I nic, szanowni Państwo, bo do dziś nie ma wyników. Co z tego wynika i czemu to wpadka? Wątek:
twitter.com/LKierznowski/statu

RT @FightCancerClub
If you are wondering how populist, self-centered views can undermine science progress - the story of 🇵🇱Ministry of Science @MEIN_GOV_PL trolling National Science Center (@NCN_PL), the main grant funding agency in Poland, is unfortunately turning into a representative example:

wwydmanski boosted

Did you ever wonder how tissues which have a very high cell turnover rate like in our #intestine control #proliferation to avoid the development of #cancer?

In this very elegant study, Guizela Huelsz Prince and colleagues at AMOLF in #Amsterdam showed in intestinal #organoids that two daughter cells from the same mother cell always follow the same trajectory. Either they both divide, or neither divides. As a result, cell growth is constant.

Article in #eLife: elifesciences.org/articles/806

Woah, Twitter told me that I've joined it 10 years ago, which gives me an average of 10 followers per year

RT @docmilanfar
A most surprising & little-known results in statistics is that the mean (μ) and median (m) are within a std deviation (σ)

|μ−m| ≤ σ

for unimodal densities bound is even tighter

|μ−m| ≤ 0.7756 σ

This beautiful results first appeared in a 1932 paper by Hotelling & Solomons

RT @AdamCzerniak
Jeżeli już sam @NCN_PL przyznaje, że konkursy są nieatrakcyjne, bo ograniczony jest budżet, to może minister @CzarnekP mógłbym coś z tym zrobić? Finansowanie badań naukowych realnie jest coraz mniejsze, a to negatywnie przekłada się na liczbę prowadzonych badań i ich jakość. twitter.com/NCN_PL/status/1600

RT @anna_korzekwa
Budżet @NCN_PL od 2018 do 2023 wzrósł zaledwie o 13%. Wskaźnik sukcesu w naszych ostatnich konkursach spadł do nieakceptowalnego poziomu < 15%. - Jeśli Polska nie zwiększy budżetu na naukę, grozi nam drenaż mózgów - mówi Zbigniew Błocki, dyrektor . wnp.pl/parlamentarny/wydarzeni

RT @valentynbez
Saving this for the record. russian propaganda machine is falling apart and officials admit they are financing separatism since 2014.
I remember these years vividly, a lot of people were fooled.
The judgment day approaches.

All you need is a little gaslighting, and ChatGPT will try to trick the user into creating chloroform

I've seen a lot of confusion about the difference between attention and self-attention, so I've decided to clarify them a bit in a medium post.

medium.com/@wwydmanski/whats-t

Show more
Qoto Mastodon

QOTO: Question Others to Teach Ourselves
An inclusive, Academic Freedom, instance
All cultures welcome.
Hate speech and harassment strictly forbidden.