Nicola Romanò @nicolaromano@qoto.org

Bots Are Better than Humans at Solving CAPTCHAs https://www.schneier.com/blog/archives/2023/08/bots-are-better-than-humans-at-solving-captchas.html #machinelearning #academicpapers #Uncategorized #captchas

Reusing other researcher's work is inherent in science. Licensing plays a central role for reuse, so we should know how it works.

♻️

Open licenses you probably need for:

- Texts, data, media, ...: CC0 or CC-BY
- Software: MIT license

More on licenses:
https://heidiseibold.ck.page/posts/choosing-an-open-license

Please take two mins to watch this video from the TUC depicting how political decisions since 2010 have all but destroyed our once-great NHS, as waiting lists have soared from 4 to 7.5 million.

Wow! I never knew that Beethoven's "Für Elise" was actually written for a woman named Therese - and the letters these two names have in common:

E S E

may have inspired the alternating sequence of notes that starts this piece:

E E♭ E

Yes: in German E flat is called "Es" and pronounced like the letter S.

I also didn't know that Beethoven never published "Für Elise" in his life: he gave the score to Therese Malfatti, and it was transcribed and published by someone else after his death, and hers. Beethoven almost published an alternative more complicated version of the tune in 1822... but at the last minute decided not to.

I learned all this stuff from this video, which also includes a performance of the alternative version of "Für Elise". If you like the famous version, which is rather simple, this one may seem ludicrously fancy.

https://www.youtube.com/watch?v=jblFQ1whX5s

On that story about scientists leaving X and coming here. Please, please lean into the feature of hashtag following and tag your stuff. A lot of us would love to find you and hashtag discovery is the easiest way here.

#scientists #scientist #mastodon #followers #discovery #news

In the era of personalised medicine could the suggestion that a treatment is tailor-made to a patient boost the placebo? Find out from Dasha Sandra in the eLife Podcast 🎧 https://elifesciences.org/podcast/episode88?utm_source=mastodon&utm_medium=social&utm_campaign=organic

The #Wasserstein distance (#EMD), sliced Wasserstein distance (#SWD), and the #L2norm are common #metrics used to quantify the ‘distance’ between two distributions. This tutorial compares these three metrics and discusses their advantages and disadvantages.

🌎 https://www.fabriziomusacchio.com/blog/2023-07-26-wasserstein_vs_l2_norm/

#OptimalTransport #MachineLearning

Well it had to happen eventually. #Scripps is retiring FLIP (FLoating Instrument Platform). This is an amazing piece of engineering (and soooo weird on the inside - everything pivots, so walls become floors). The #ship to be towed out to a location, and it would literally flip, sinking most of the ship directly down to give a *very stable platform for #oceanography research. Launched in 1962.
https://maritime-executive.com/article/world-s-strangest-research-vessel-heads-for-scrapyard-after-51-years

Doing the #TidyTuesday early. Spam emails are more likely to contain certain characters compared to non-spam.

🔗: https://github.com/borstell/tidytuesday/tree/main/2023/2023-08-15/

#RStats #tidyverse #ggplot2 #dataviz

My book "Building reproducible analytical pipelines with #rstats" is done! Buy a drm-free epub or pdf @leanpub : https://leanpub.com/raps-with-r/

Physical https://amazon.com/dp/B0C87H6MGF (check your local market place)

Read for free forever: https://raps-with-r.dev

#RStats #datascience

@caffinepwrd @davidtoddmccarty Over here, Tories and #brexshitters and the like appear to view the world as a negative sum game, and their aim is to make sure that everybody else other than them are the losers, even if they can't win themselves.

#AI Model Fit Obsession Disorder (MFOD) : A #pathology afflicting a #datascientist or #machinelearningengineer in which they think their job is to call the function model.fit, and that's it. The disorder is characterized by a lack of curiosity about whats inside their #blackbox, a belief that only good things happen from calling the model.fit function, and disdain for understanding #ai by means other than the mean square error. #aihype #datascience #machinelearning

has anyone seen a really good analysis of the problems with git's command line UI? Would love to read it. for example:

* `git checkout` is dangerous and has too many different jobs
* for a tool that's supposed to make changes easy to undo, you actually need to learn a LOT of different ways go back to the previous state

(not looking for git tutorials or explanations of why you think git's UI is actually good, just an analysis of the problems)

Blog4: Annotate image.

Arrows, letter code, colors, and more - they all need to be explained to your audiences! Ideally in a way that they are visible too!

https://helenajamborwrites.netlify.app/posts/image_how_annot/

#scicomm #BioImages #Microscopy #dataviz #images #Image #ScienceMastodon #Science

PS Prize question, how many annotations do you see in the image?

Reminder that you should absolutely never use the website called “sci-hub dot st” to get free access to research papers instead of feeding the greed of for-profit publishers.

That’s basically communism, which is of course illegal. 👩🏻‍🏫

There are some *beautiful* lessons on building ML-based classifiers in biology in the recent Gihawi et al. paper (Major data analysis errors invalidate cancer microbiome findings)

Classifiers will always look for weird things and then report way-too-good-accuracies. At first glance no classifier in biology should be highly accurate: by now that should be a red flag.

I wonder whether there's a fun 'signal' in studies including randomness: so many tutorials, notebooks, and repos use 42 as a random seed.
I wonder whether by now we have a weird bias because of that number somewhere :)

Today I'm seeing multiple posts alerting me to serious problems with an update to Zoom's terms of use (variously stated as dated last March or April) according to which
- They have the right to feed all your Zoom calls and associated data to an AI learning system
- They own this data forever
- There is no opt-out
- If you show anything copyrighted on a Zoom call and the copyright owner goes after Zoom for copyright infringement for using this data, then you are personally on the hook for all legal damages

For example see https://the.webm.ink/not-using-zoom, via https://news.ycombinator.com/item?id=37034145

Even if you are somehow ok with training AI to look like you and talk like you, I think the last point makes it no longer safe to use Zoom for any purpose. That's a big issue for online collaboration, conferencing, and online office hours with students (note high likelihood of incorporating copyrighted textbook material into those office hours).

There's a good discussion of alternatives at https://mathstodon.xyz/@MedievalMideast@hcommons.social/110838371676478198 and probably starting up at many other places as well.

Just discovered there is an extension of #Inkscape to quickly edit scientific figures (change axis ratio, font sizes, etc) 🤩

https://github.com/burghoff/Scientific-Inkscape