I've been trying out leonardo.ai with a bunch of new prompts since yesterday.

The following are the best results for: "Burning the cradle at both ends" ... and no "cradle" wasn't a typo, that was the prompt I wanted 😛

#AIArt #StableDiffusion #DeepLearning #MachineLearning #CV #AI
Prompt: “Burning the cradle at …
Prompt: “Burning the cradle at …
Prompt: “Burning the cradle at …
Prompt: “Burning the cradle at …

Boosted 5 papers in the cs.CV category out of a total of 91 new (or updated) papers on arXiv.org today.

The reading of the papers is done for today and off I go to do other things 🙂

#AI #CV #NewPapers #DeepLearning #MachineLearning

I’ve been talking about trying to create a smaller, quieter home timeline and yet be able to follow interesting accounts with a large number of posts/boosts. I tried unfollowing some of these accounts and putting them into a list and, as should have been obvious, it seemed to fail since you do need to be following the accounts.

And yet, my list kept showing new posts but not at the volume that I would have expected from the users I added to the list. And only posts from some users.

I was puzzled about this and was wondering what was going on. I think I might have an answer — it looks as if you don’t need to be following the user to get their posts *as long as you follow somebody else who follows that user*. If the user and you have nobody else in common, then you won’t see any of their posts in your list.

So still not the ideal outcome, but maybe a workaround (albeit one which can have privacy implications) in case you want to go that route, I guess?

I think I’ll simply go back to my original idea of implementing homeline filtering for list members. But that’ll take a bit longer given the stuff I’m working on at the moment 😞

#Mastodon #Fediverse #Lists #Filtering

"A Picture May Be Worth a Thousand Lives: An Interpretable Artificial Intelligence Strategy for Predictions of Suicide Risk from Social Media Images. (arXiv:2302.09488v1 [cs.AI])" — Predicting suicide risk by evaluating non-verbal inputs such as social media images.

Paper: http://arxiv.org/abs/2302.09488

#AI #CV #NewPaper #DeepLearning #MachineLearning

<<Find this useful? Please boost so that others can benefit too 🙂>>

"Vulnerability analysis of captcha using Deep learning. (arXiv:2302.09389v1 [cs.CR])" — Using a Convolutional Neural Network (CNN) model to predict text-based CAPTCHAs to examine the flaws inherent in the system and to create more resilient CAPTCHAs.

Paper: http://arxiv.org/abs/2302.09389

#AI #CV #NewPaper #DeepLearning #MachineLearning

<<Find this useful? Please boost so that others can benefit too 🙂>>
Examples of different alphanume…

"Exploring the Representation Manifolds of Stable Diffusion Through the Lens of Intrinsic Dimension. (arXiv:2302.09301v1 [cs.CL])" — An investigation into the basic geometric properties induced by prompts in Stable Diffusion and how this impact depends on the layer being considered.

Paper: http://arxiv.org/abs/2302.09301

#AI #CV #NewPaper #DeepLearning #MachineLearning

<<Find this useful? Please boost so that others can benefit too 🙂>>
The output of the model at the …

"Web Photo Source Identification based on Neural Enhanced Camera Fingerprint. (arXiv:2302.09228v1 [cs.CV])" — Using a neural network to identify sensor patterns in an effort to identify the source camera for images published on the web.

Paper: http://arxiv.org/abs/2302.09228
Code: https://github.com/PhotoNecf/PhotoNecf

#AI #CV #NewPaper #DeepLearning #MachineLearning

<<Find this useful? Please boost so that others can benefit too 🙂>>
Three main stages of source ide…

"A Pilot Evaluation of ChatGPT and DALL-E 2 on Decision Making and Spatial Reasoning. (arXiv:2302.09068v1 [cs.AI])" — An evaluation of ChatGPT and DALL-E2 to assess the spatial reasoning and decision making abilities of each model.

Paper: http://arxiv.org/abs/2302.09068

#AI #CV #NewPaper #DeepLearning #MachineLearning

<<Find this useful? Please boost so that others can benefit too 🙂>>
DALL-E output for the prompts: …

A total of 91 papers in the cs.CV category on arXiv.org today — 47 new, 44 updated.

On with the reading?

#AI #CV #NewPapers #DeepLearning #MachineLearning

@KathyReid It’s sort of a thread-unravel situation, honestly 🙂 I grew up speaking Sinhalese (which has its roots in Pali and Sanskrit) and hearing my parents speaking Tamil.

When I got to learning Hindi, I realized that it too had a lot of Sanskrit (not sure about the Pali) and some Persian/Arabic words. So it was fairly easy to pick up.

Spanish had a lot of Latin in it and so by way of English, it was again easy to pick up. Written French is easy to understand for somewhat the same reasons, but spoken French is something I have a hard time with.

Malayalam and Telugu again have similarities to Tamil and so again, easy to understand if you look for the connections.

The trouble with most of these is the written part since except for the European languages, everything else has it’s own alphabet and I can read in only some. Though again, a lot of them have similar letters which can help you figure at least some of it out …

@AngelaPreston I thought it was somebody else but my memory’s so bad that I might be totally wrong 🙂

@AngelaPreston This is the second time I’ve seen somebody mention Scholomance on here and mention how they liked the magic system. I love magic systems which are done well … So might have to give it a try and see if that will get me back on reading 🙂

@KathyReid Speak? English, Sinhalese, Tamil, Hindi … Understand to some extent — Spanish, French, Telugu, Malayalam, Punjabi, Arabic … Learnt but mostly forgot …Malay

#Garden in Winter

How many native plants to plant? Consider:

Mass plantings better attract and serve pollinators, which are losing habitat where you live.
Pollinators are beautiful and calming to watch, which you need more of in your life.
The cheap seeds can be sown directly in the ground and pots for those unmotivated due to lack of experience with multicolored butterflies fluttering around.

Help yourself and the #environment. Plant a bunch.

More in alt text.

#Nature #Gardening #Flowers

@at I think I know of only a couple of clients that work on the Linux side, and as far as I know, none of them have that feature, sorry.

Elk might be one to try since it’s web-based. I haven’t tried it since I think it doesn’t work with Akkoma either 😛 But this whole not supporting Akkoma thing is stopping me from testing out a bunch of things and so I might have to set up a secondary Mastodon account just so that I can be sure of the features available with the various clients …

@at I don’t know how to do that with the Mastodon web-client (or if that’s even possible) but on macOS/iOS, the following apps support it for sure: Tusker and IceCubes.

There might be other clients around which does too, for example, I have a vague recollection of Ivory maybe supporting it but I used Ivory only for a short time before switching to Akkoma and Ivory does not support Akkoma and so I can’t verify now 🙂

@wavesculptor I misread what you wrote originally, I thought you were saying that that’s how lists worked, but I realize now that you were saying that that is how it *should* happen. And you are right, that’s how it should have been implemented.

But looks as if it wasn’t implemented that way and so lists are basically a pointless feature at the moment since it duplicates content and is generally annoying 🙂

Will somebody fix it from the server end? I have no idea. But if I can find a good JavaScript open source client (not one with lots of libraries that require yarn or webpack) which is fairly feature complete, I might attempt the task on the client-side using that. Haven’t found such a client yet though 😞 Which was why I was thinking of implementing it on the client that I use now but that would limit the number of people who can use it to people running iOS or macOS …

@wavesculptor Personally, I wasn’t seeing non-followed people on a list as a rights violation but more of a privacy violation since those people might not know that you are monitoring their posts and they have not explicitly agreed to you seeing those posts …

I know that there are arguments for both side on this but I can see how this can be misused to stalk people without their knowledge. So I do see the issue there.

What the Mastodon issue I linked to in the previous reply discusses is around the fact that you can require following to add a user to a list but that if the user is on a list, their posts are filtered out from the main timeline. That seems like a good compromise to me, but unfortunately, I have no idea if they’ll actually implement that or not 🙂

@wavesculptor To be fair, most people won’t be able to engage since they might not know what to do about it either 🙂

There is an open issue for Mastodon for this feature but I’m not sure if it’s being actively worked on:

https://github.com/mastodon/mastodon/issues/6982

If it’s not, then some of the things that can be done would be:

1. Implement it yourself (I do realize you’ve said that’s not an option for you but just listing it for the sake of completeness …)

2. Get a build from somebody who implemented it (If you are on macOS I’d be happy to share my build if you want it … provided I get it implemented right)

3. Lobby your favourite Masto/Fedi client developer(s) to implement this feature for you …

I originally thought that this didn’t work for unfollowed Akkoma accounts but I’m not absolutely certain now since I’m getting some posts on the list I created for accounts that I’m not following. But I’m not sure if this is Akkoma, a fluke, or some sort of federation delays. So I’ll have to give it a few days and then see what might need to happen. But if it works for Akkoma, then you also have option #4 — set up your own instance of Akkoma or move to an Akkoma server since that might resolve the issue for you.

Short #introduction post.

I work as a #data scientist / data science consultant mostly on natural language processing #nlproc and tabular data (though getting more into the #mlops space lately).

My passion is #running, primarily #trailrunning and #ultrarunning. Also a beginner in #mountaineering, #scifi reader and interested in all things #climate.

#datascience #machinelearning
#maths #mathematics #fieldhockey #python