Offshore
Photo
AK
PLay: Parametrically Conditioned Layout Generation using Latent Diffusion
abs: https://t.co/x5S2KXi6WH https://t.co/gbYNDZnJb5
tweet
PLay: Parametrically Conditioned Layout Generation using Latent Diffusion
abs: https://t.co/x5S2KXi6WH https://t.co/gbYNDZnJb5
tweet
Offshore
GIF
AK
3DShape2VecSet: A 3D Shape Representation for Neural Fields and Generative Diffusion Models
abs: https://t.co/Cg7BxuI6sn
project page: https://t.co/QBQUH9p2VZ https://t.co/NaLv2Cex9N
tweet
3DShape2VecSet: A 3D Shape Representation for Neural Fields and Generative Diffusion Models
abs: https://t.co/Cg7BxuI6sn
project page: https://t.co/QBQUH9p2VZ https://t.co/NaLv2Cex9N
tweet
Offshore
Video
Aran Komatsuzaki
Moûsai: Text-to-Music Generation with Long-Context Latent Diffusion
Develops a cascading latent diffusion approach that can generate multiple minutes of high-quality stereo music at 48kHz from textual descriptions.
abs: https://t.co/UpoywkQI05
repo: https://t.co/XqHCYSyg2l https://t.co/EiPn9luH6f
tweet
Moûsai: Text-to-Music Generation with Long-Context Latent Diffusion
Develops a cascading latent diffusion approach that can generate multiple minutes of high-quality stereo music at 48kHz from textual descriptions.
abs: https://t.co/UpoywkQI05
repo: https://t.co/XqHCYSyg2l https://t.co/EiPn9luH6f
tweet
Offshore
Photo
Aran Komatsuzaki
Leveraging the Third Dimension in Contrastive Learning
Leveraging the additional depth channel dimension of BYOL leads to an increase in downstream classification accuracy from 85.3% to 88.0% on ImageNette and 84.1% to 87.0% on ImageNet-C.
https://t.co/DleM5sZ0lY https://t.co/xPXSMkyOeT
tweet
Leveraging the Third Dimension in Contrastive Learning
Leveraging the additional depth channel dimension of BYOL leads to an increase in downstream classification accuracy from 85.3% to 88.0% on ImageNette and 84.1% to 87.0% on ImageNet-C.
https://t.co/DleM5sZ0lY https://t.co/xPXSMkyOeT
tweet
Offshore
Photo
Aran Komatsuzaki
SWARM Parallelism: Training Large Models Can Be Surprisingly Communication-Efficient
Shows that it is possible to train a billion-scale LM on preemptible servers with low-power GPUs and low network bandwidth while achieving high training throughput.
https://t.co/3Sx9Klc4i2 https://t.co/Yrz9t8o5HK
tweet
SWARM Parallelism: Training Large Models Can Be Surprisingly Communication-Efficient
Shows that it is possible to train a billion-scale LM on preemptible servers with low-power GPUs and low network bandwidth while achieving high training throughput.
https://t.co/3Sx9Klc4i2 https://t.co/Yrz9t8o5HK
tweet
Offshore
Photo
Daily AI Papers
Exploration via Elliptical Episodic Bonuses
https://t.co/l9mz0BQMkx
Exploration via Elliptical Episodic Bonuses (E3B) is a new method which extends count-based episodic bonuses to continuous state spaces. The...
🧵 👇 https://t.co/j1ZKxGGcGm
tweet
Exploration via Elliptical Episodic Bonuses
https://t.co/l9mz0BQMkx
Exploration via Elliptical Episodic Bonuses (E3B) is a new method which extends count-based episodic bonuses to continuous state spaces. The...
🧵 👇 https://t.co/j1ZKxGGcGm
tweet
Offshore
Photo
Stable Diffusion 🎨 AI Art
A girl on the streets of Tokyo #AIArt
#StableDiffusion2 / #StableDiffusion #DreamStudio https://t.co/fFUY9cVqi6
tweet
A girl on the streets of Tokyo #AIArt
#StableDiffusion2 / #StableDiffusion #DreamStudio https://t.co/fFUY9cVqi6
tweet