Discover and read the best of Twitter Threads about #diffusion
Most recents (14)
WARNING:
There are new pathogens in our midst.
My latest article about media #manipulators, velocity #hacking, and the latest Covid origin #controversy.
Maybe some #mainstream journalists & influencers take a good look in the mirror.
Read:
protagonistfuture.substack.com/p/the-rise-of-…
There are new pathogens in our midst.
My latest article about media #manipulators, velocity #hacking, and the latest Covid origin #controversy.
Maybe some #mainstream journalists & influencers take a good look in the mirror.
Read:
protagonistfuture.substack.com/p/the-rise-of-…
How much should we think about the role of information in society?
In this article, I use the recent media coverage surrounding #lableak versus #zoonosis to point toward a new threat we have not yet wrapped our heads around:
Information pathogens with high #velocity.
In this article, I use the recent media coverage surrounding #lableak versus #zoonosis to point toward a new threat we have not yet wrapped our heads around:
Information pathogens with high #velocity.
#Velocity is a metric for the transmission efficacy of information given a particular content payload, its viral packaging, and its host environment.
Basically the R0 of information, a measure of #contagiousness.
Info pathogens with high velocity outcompetes good information.
Basically the R0 of information, a measure of #contagiousness.
Info pathogens with high velocity outcompetes good information.
Anlässlich der Debatte um das 🇪🇺#Verbrenner"verbot", die von 🇩🇪 jetzt auch nach 🇦🇹 schwappt, hier etwas Hintergrund zum optimalen #policy #mix, wenn es darum geht, den technologischen Fortschritt in eine #CO2-freie Richtung zu lenken, aus umweltökonomischer Sicht 1/
1. Instrument der Wahl ist i.d. Theorie eine umfassende #CO2-Bepreisung in angemessener Höhe (z.B. = #Schadenskosten) mit sozial ausgewogener #Rückvergütung: CO2-Preis korrigiert fehlenden Preis für Umweltgüter, der #fossile Kraftstoffe zu günstig macht 2/
Aber i.d. Realität lässt sich ein angemessen hoher CO2-Preis nicht überall umsetzen: Weil a) der ökonomisch #perfekte Markt nicht existiert (weitere #Marktversagen), und b) sich aus #sozialen Gründen im Gebäude- und Verkehrsbereich keine so hohen CO2-Preise durchsetzen lassen 3/
Retrieval Augmented #Diffusion (RDM) models: Smaller diffusion models can generate high-quality generations by accessing an external memory to guide the generation. Inspired by Deepmind's RETRO.
A 🧶
Paper: arxiv.org/abs/2204.11824
Day 10 #30daysofDiffusion #MachineLearning
A 🧶
Paper: arxiv.org/abs/2204.11824
Day 10 #30daysofDiffusion #MachineLearning
If the model can rely on this external memory always, it just has to learn important details about the image generation process such as the composition of scenes rather than, for example, remembering how different dogs look like.
Setting: X is the training set and D is a *disjoint* image set which is used for retrieval. θ denotes the parameters of the diffusion model. ξ is the retrieval function which takes in an image and selects "k" images from D. φ is a pretrained image encoder.
StructureDiffusion: Improve the compositional generation capabilities of text-to-image #diffusion models by modifying the text guidance by using a constituency tree or a scene graph.
A 🧵
Paper: arxiv.org/abs/2212.05032
Day 9 #30daysofDiffusion #MachineLearning
A 🧵
Paper: arxiv.org/abs/2212.05032
Day 9 #30daysofDiffusion #MachineLearning
T2I models like SD produce great aesthetically pleasing generations for a given prompt, however, most of us never get them right on the first try. Sometimes the model ignores part of the prompt and some objects we want in the picture are missing.
Also sometimes the model gets adjectives mixed up. For example, in the figure below, the prompt is - "red car and white sheep". However, the model produced a red sheep too!
The authors address this compositionality issue in this paper.
The authors address this compositionality issue in this paper.
InstructPix2Pix: Edit an image using text guidance using a single forward pass. Why use any inversion or other stuff,just create a dataset using inversion techniques and train a new model.
A 🧶
Paper: arxiv.org/abs/2211.09800
Day 8 #30daysofDiffusion #Diffusion #MachineLearning
A 🧶
Paper: arxiv.org/abs/2211.09800
Day 8 #30daysofDiffusion #Diffusion #MachineLearning
It should be fast when you want to edit an image in real-time. Models like textual inversion or prompt-to-prompt optimize during inference which makes them slow.
In this paper, the authors cleverly use such techniques to generate the training data and then finetune Stable Diffusion to perform edits in a single forward pass. They use 2 pretrained models, GPT-3 Davinci model and the SD model to generate the data.
DreamBooth: Assign a rare sequence of tokens as the subject's identifier and fine-tune the diffusion model on the small set of images with the "subject". A 🧵
Paper: arxiv.org/abs/2208.12242
Day 1 #30daysofDiffusion #Diffusion #MachineLearning
Paper: arxiv.org/abs/2208.12242
Day 1 #30daysofDiffusion #Diffusion #MachineLearning
The authors use the Imagen model in this paper which uses T5-XXL language model to encode the text guidance to generate small 64x64 image first and then use a super-resolution model to blow it up to 1024x1024.
The authors observed that fine-tuning all the modules (including SR module) results in the best performance.
In NVIDIA's new paper on #Diffusion Models, they show how more denoisers (for each stage) and more embeddings (text, image) helps with quality!
TL;DR: If you buy more GPUs, you get correct spelling too.
deepimagination.cc/eDiffi/ #AI #ML
TL;DR: If you buy more GPUs, you get correct spelling too.
deepimagination.cc/eDiffi/ #AI #ML
With so many different labs rushing to research and deploy this kind of technology, this will quickly turn into a race for more efficiency as different providers compete on costs too.
The paper is a bit evasive on the dataset (LAION?) — I presume for legal reasons. But the good news is that it's "only" 1B text/image pairs... although they are highly filtered.
IMHO there's much more room to improve quality with the current datasets.
IMHO there's much more room to improve quality with the current datasets.
NEW: This Post Was Written By #AI: sjtylr.net/2022/10/08/thi…
From idea to content and published post in under 30mins, using free tools. This tech is developing *fast* and will have implications for teaching, learning and our graduates.
#EdTech
From idea to content and published post in under 30mins, using free tools. This tech is developing *fast* and will have implications for teaching, learning and our graduates.
#EdTech
☝️Using tools from @playground_ai @peppertype_ai @AiWritesonic @copy_ai and inspired by @Suhail @kaifulee @FryRsquared @DeepMind @daniel_eckler and more. Follow them all for super-interesting news. #AI
Diffusion models is the new trend and everyday new enthusiasts approach these tools to make art and create awesome applications.
There is a ton of beautiful pictures made with #generativeart but have you ever tried and failed?
Me, many times😁
A thread to improve your prompts⬇️
There is a ton of beautiful pictures made with #generativeart but have you ever tried and failed?
Me, many times😁
A thread to improve your prompts⬇️
1. Sometimes less is more. Writing hundreds of characters is not the right way to produce beautiful images. Consider that there are "classes" like 4k or high-resolution which embeds a lot of characteristics.
These are already available in @StableDiffusion or #discoart @JinaAI_
These are already available in @StableDiffusion or #discoart @JinaAI_
2. More detailed, more precised. If you use a specific jargon to obtain a certain style or effect, it is quite sure that you will be impressed from the result!
Try something very specific using technicalities like
infrared photo or sticker illustration, also tattoo works!
Try something very specific using technicalities like
infrared photo or sticker illustration, also tattoo works!
I discovered a bug in my own Diffusion + CLIP pipeline and suddenly the samples are unreal.. 🤯
Here's
"Just a liquid reality..."
#AIart #notdalle2 #Diffusion #clip
Here's
"Just a liquid reality..."
#AIart #notdalle2 #Diffusion #clip
"The magnificent portal of mother Gaia"
"Framing reality"
"From undirected to directed networks of dynamical agents"
Today's SFI Seminar from @robinus88 (@UCSantaBarbara), streaming now:
(Follow this 🧵 for highlights and select slides)
Today's SFI Seminar from @robinus88 (@UCSantaBarbara), streaming now:
(Follow this 🧵 for highlights and select slides)
"[This is] the main question when we talk about power grids...it could be water, it could be gas, it could be opinions transmitted over social media:"
- @robinus88 (@UCSantaBarbara), streaming now:
- @robinus88 (@UCSantaBarbara), streaming now:
"We want to keep the right-hand side of this equation as close to zero as possible. What happens if you produce too much, the frequency increases, which we don't want for a variety of reasons."
- @robinus88 (@UCSantaBarbara), streaming now:
#electricity
- @robinus88 (@UCSantaBarbara), streaming now:
#electricity
Analysis: #NASDAQ $DFFN
Case 524 #Diffusion Pharmaceuticals Inc.
Requested by: @statseo
DISCLAIMER: The analysis is strictly for educational purposes and should not be construed as an invitation to trade.
DFFN 1/3
Case 524 #Diffusion Pharmaceuticals Inc.
Requested by: @statseo
DISCLAIMER: The analysis is strictly for educational purposes and should not be construed as an invitation to trade.
DFFN 1/3
Daily Chart: Most of the Dec. 2020 gains have been lost and #support was found at the Mar. 2020 #trendline. A short term signal has appeared, but price needs to close above the medium term synthetic then above 2.0731 control. Above .....
DFFN 2/3
DFFN 2/3
..... 2.4590 target #resistance at 3.0185- 3.2753. Control is 0.7399. We're planning to enter this rather quickly.
The #Strategy is printed on the image.
DFFN 3/3
The #Strategy is printed on the image.
DFFN 3/3
Partage de simulations de modèles de diffusion qui illustrent bien problème & solution du #COVIDー19 .
1/5 : #diffusion libre - profil #exponentiel - avec #population en circulation libre & contact aléatoire non contrôlé, comme en #France avant jeudi 12/3.
Src @washingtonpost
1/5 : #diffusion libre - profil #exponentiel - avec #population en circulation libre & contact aléatoire non contrôlé, comme en #France avant jeudi 12/3.
Src @washingtonpost
2/5 : modèle de diffusion avec création de zone de #confinement et tentative de #quarantaine forcée, comme en #Chine (Hubei).
En pratique, impossible d'isoler (à part sur un bateau) la population #COVIDー19 de la #population saine. La zone finit par devenir poreuse.
Src Wshgtpost
En pratique, impossible d'isoler (à part sur un bateau) la population #COVIDー19 de la #population saine. La zone finit par devenir poreuse.
Src Wshgtpost
3/5 (#solution) : modèle de diffusion avec #SocialDistancing ou #DistanciationSociale pour 3/4 de la population et circulation libre d'un quart de la population.
Deux simulations dans la vidéo suivante qui montrent des charges de contamination plutôt contrôlées.
#COVIDー19
Deux simulations dans la vidéo suivante qui montrent des charges de contamination plutôt contrôlées.
#COVIDー19
SFI President David Krakauer introduces our annual board of trustees symposium on New #Complexity #Economics - follow this thread for live coverage of talks by @EricBeinhocker @AKStanger @ole_b_peters @JacksonmMatt @cmoncap & W. Brian Arthur today...
"I think we have to be aware that we're not just discussing financial systems. We are discussing the future of the stability of the planet."
- David Krakauer
on #Complexity #Economics
- David Krakauer
on #Complexity #Economics
"What is the #economy? I would argue that it doesn't exist in the physical world. It is a product of our #imagination. It is made out of #ideas. It does have an effect on the physical world. But the imagined order that we have is not succeeding."
- @EricBeinhocker
- @EricBeinhocker
