Next was an incredible talk by Yang Song on #diffusion and score-based generative models at MIT Brain and Cognitive Sciences. This is hands down the most thorough talk I've heard on the ins and outs of how to implement generative models. Highly recommend https://www.youtube.com/watch?v=wMmqCMwuM2Q (4/8) #GenerativeAI #AI #LLMs

Question for those of you attuned to this field - are GANs still utilized or have diffusion models largely replaced them? Not just in the image domain either, I'm talking about in general.

If GANs are still utilized in a particular field, then I ask - for what? What kinds of GANs?

#gan #diffusion #deeplearning #machinelearning #ai

https://www.sudradio.fr/bercoff-dans-tous-ses-etats/karl-zero-trafic-enfants-sound-of-freedom
Le film "Sound of freedom" sort aux États-Unis. C’est une fiction basée sur des faits réels. Son sujet : la traite d’enfants. Karl Zéro cherche actuellement à faire venir ce film en France.
#SoundOfFreedom #enfant #enfants #abus #trafic #humanité #film #films #sortie #censure #pouvoir #affairescriminelles #esclavage #droitshumains #enfance #mondial #Migrants #migrant #vol #tousensemble #cinéma #KarlZéro #SudRadio #émission #radio #diffusion #étatsunis #boxoffice #histoire

Here is a deeper dive on the relationship between filtrations and our information on stochastic processes:

https://freeenergy.blog/2023/FiltrationAndInformation.html

#math #probability #diffusion #stochastic

I started to learn about stochastic calculus because many of the most interesting breakthroughs these days (e.g., diffusion probabilistic models for generative AI) involve stochastic processes. There are a lot of new and confusing concepts: Lebesgue integrals, measure theory, sigma algebras and Borel sets, martingales, rigorous probability theory, etc. Here is a post on a few learnings related to martingales:

https://freeenergy.blog/embracing%20transitory%20confusion/probability%20theory/math/2023/05/29/Filtration.html

#math #probability #diffusion

Bon, pendant que l'illuminé de l'Elysée pousse la chansonnette avec des fafs subventionnés par la DRAC (fatiiiigue), ça me rappelle qu'il faut que je vous dise qu'on cherche un.e chargé.e de production et de diffusion pour 2 compagnies de danse strasbourgoises et leurs spectacles de qualité (eux, na !) pour 20h/mois en CDDU. Réseaux (au pluriel😉) Jeune public, Rue, et hors les murs. Venez, on se met de côté et on en discute 😉
#quonseledise #recrutement #danse #diffusion #production #partagez!

I'm trying to understand the Sohl-Dickstein 2015 paper (https://arxiv.org/pdf/1503.03585.pdf) on generative diffusion probabilistic models, and for the longest time was hung up on how the reverse process could work. Clearly, a reverse random walk is very unlikely to go back to the original spot. Then I realized that we just need to "fine-tune" the mean and stdev to make it go backwards. Here is a simple example that brute forces that: https://colab.research.google.com/drive/151u9xjBIF-bxbASym9tbvlcWWT83rMjq?usp=sharing

#generative #ml #diffusion

Next was a great talk by Greg Ver Steeg on the inner workings of #diffusion models at the #USC Information Sciences Institute. This talk provides solid theoretical groundwork for grasping this approach, highly recommend https://www.youtube.com/watch?v=8azWhUug7mY (6/11) #AI

This is acknowledgement that #Diffusion models extract the expression from input training data, store it (like a time-delayed photocopier) then output that expression on demand.

Since the prompt, by definiton, can't contain the expression, it must have come frm Copyrighted data.

There is no direct legal requirement for companies to respect opt-in, but indirectly, many customers have requirements for IP that definitely depend on opt-in!

The games industry is quickly revising its outsourcing contracts to ensure no infringing #Diffusion systems are used...

Adobe is distributing 248M stock images. I presume the total of public domain and creative commons images are less than that, so upper bound is 500M images for a #Diffusion model of that quality.

(Technical analysis only, no comment on legality or ethicalness.)

One reason #Diffusion-style models are here to stay, even if GANs make a come-back in quality: they are more flexible.

A #Diffusion model has an encoder and decoder, so you can feed in images and get useful stuff back out. In-Painting, Out-Painting, style transfer, blending...

When training #Diffusion models, you know how the loss value is significantly differently calibrated for different timesteps...

Does anyone normalize these? I've seen papers where they blur the image, I presume that helps calibrate the losses better?

RT @kadirnar_ai
Prompt-to-Prompt Image Editing with Cross-Attention Control

Demo: https://huggingface.co/spaces/kadirnar/prompt-to-prompt_stable-diffusion

#huggingface #stableDifusion #paper #diffusion

AudioLDM: Text-to-Audio Generation with Latent Diffusion Models, a TTA system that is built on a latent space to learn the continuous audio representations from contrastive language-audio pretraining (CLAP) latents.

https://audioldm.github.io

#deeplearning #diffusion

Dreamix: Video Diffusion Models are General Video Editors

https://dreamix-video-editing.github.io

#deeplearning #DIFFUSION

First was an interesting talk by @danielromero on the influence of #networks, identity, and #geography on linguistic innovation #diffusion at the #NICO. New terms seem to diffuse differently for urban and rural areas, and this work points out the need for models to incorporate socioeconomic factors https://www.youtube.com/watch?v=MSYlFl_V14w&t=2s (2/6)

@benbloodworth I haven’t seen a move from my bunch after that first, small, wave came over. I just checked with one of those third-party apps and did a word search, too. I think some are sticking around there, or just went elsewhere.

This will make for an excellent #diffusion of #innovation #study for someone’s dissertation. (If they could ever discover this post via the miserable #search.)

((In addition to your excellent list, I’m going to soon need a better app experience.))

Definitely not a grid... Genuary #18

some nice stills with different variations

#genuary #genuary2023 #genuary18 #reactiondiffusion #digitalart #newmediaart #creativecodeart #nature #rendering #blue #white #simulation #creativecode #generativeart #generative #genartclub #codeart #creativecodeart #proceduralart #processing #generativedesign #algorithmicart #artxcode #abstractart #reaction #diffusion

Definitely not a grid... Genuary #18

Reaction Diffusion system in #webgl #threejs

#genuary #genuary2023 #genuary18 #reactiondiffusion #digitalart #newmediaart #creativecodeart #nature #rendering #blue #white #simulation #creativecode #generativeart #generative #genartclub #codeart #creativecodeart #proceduralart #processing #generativedesign #algorithmicart #artxcode #abstractart #reaction #diffusion