Sign in to confirm you’re not a bot
This helps protect our community. Learn more

Flow Matching for Generative Modeling (Paper Explained)

Yannic Kilcher
284K subscribers
<__slot-el>
<__slot-el>
69K views 1 year ago
Flow matching is a more general method than diffusion and serves as the basis for models like Stable Diffusion 3. Paper: https://arxiv.org/abs/2210.02747 Abstract: We introduce a new paradigm for generative modeling built on Continuous Normalizing Flows (CNFs), allowing us to train CNFs at unprecedented scale. Specifically, we present the notion of Flow Matching (FM), a simulation-free approach for training CNFs based on regressing vector fields of fixed conditional probability paths. Flow Matching is compatible with a general family of Gaussian probability paths for transforming between noise and data samples -- which subsumes existing diffusion paths as specific instances. Interestingly, we find that employing FM with diffusion paths results in a more robust and stable alternative for training diffusion models. Furthermore, Flow Matching opens the door to training CNFs with other, non-diffusion probability paths. An instance of particular in …...more
...more
Attention Formula Black Mug
$19.99
Spring
Dark items with the attention formula printed on
Shop
Spring
Channel Logo Light Leggings
$39.99
Spring
Items with light color (white or grey) and the channel logo printed on. Stretch your limits in Teespring's athletic leggings. Made from 88% polyester and 12% lycra, our leggings offer great compression and flexibility. Full Color Custom Printing. Reinforced Waistband. Overlock Stitching.BreathTech.
Shop
Spring
Channel Logo Light Comfort Tee
$24.99
Spring
Items with light color (white or grey) and the channel logo printed on. Custom designed graphic is printed in vivid color and high resolution using state of the art color transfer technology. Shirts are made from super soft 100% combed ringspun cotton. Printed in the USA.
Shop
Spring
Channel Logo Light Unisex Premium Pullover Hoodie
$49.99
Spring
Items with light color (white or grey) and the channel logo printed on. Custom designed graphic is printed in vivid color and high resolution using state of the art color transfer technology. Hoodies & Sweatshirts are made from super soft 50% cotton, 50% polyester. Printed in the USA. Guaranteed to last through hundreds of washes.
Shop
Spring
Channel Logo Light Comfort Tee
$24.99
Spring
Items with light color (white or grey) and the channel logo printed on. Custom designed graphic is printed in vivid color and high resolution using state of the art color transfer technology. Shirts are made from super soft 100% combed ringspun cotton. Printed in the USA.
Shop
Spring
Channel Logo Dark Comfort Tee
$24.99
Spring
Dark items with the channel logo printed on. Custom designed graphic is printed in vivid color and high resolution using state of the art color transfer technology. Shirts are made from super soft 100% combed ringspun cotton. Printed in the USA.
Shop
Spring
Yannic Kilcher tagged products below. Learn more
Flow Matching for Generative Modeling (Paper Explained)
1.9KLikes
69,845Views
2024Apr 8
Flow matching is a more general method than diffusion and serves as the basis for models like Stable Diffusion 3. Paper: https://arxiv.org/abs/2210.02747 Abstract: We introduce a new paradigm for generative modeling built on Continuous Normalizing Flows (CNFs), allowing us to train CNFs at unprecedented scale. Specifically, we present the notion of Flow Matching (FM), a simulation-free approach for training CNFs based on regressing vector fields of fixed conditional probability paths. Flow Matching is compatible with a general family of Gaussian probability paths for transforming between noise and data samples -- which subsumes existing diffusion paths as specific instances. Interestingly, we find that employing FM with diffusion paths results in a more robust and stable alternative for training diffusion models. Furthermore, Flow Matching opens the door to training CNFs with other, non-diffusion probability paths. An instance of particular interest is using Optimal Transport (OT) displacement interpolation to define the conditional probability paths. These paths are more efficient than diffusion paths, provide faster training and sampling, and result in better generalization. Training CNFs using Flow Matching on ImageNet leads to consistently better performance than alternative diffusion-based methods in terms of both likelihood and sample quality, and allows fast and reliable sample generation using off-the-shelf numerical ODE solvers. Authors: Yaron Lipman, Ricky T. Q. Chen, Heli Ben-Hamu, Maximilian Nickel, Matt Le Links: Homepage: https://ykilcher.com Merch: https://ykilcher.com/merch YouTube:    / yannickilcher   Twitter:   / ykilcher   Discord: https://ykilcher.com/discord LinkedIn:   / ykilcher   If you want to support me, the best thing to do is to share out the content :) If you want to support me financially (completely optional and voluntary, but a lot of people have asked for this): SubscribeStar: https://www.subscribestar.com/yannick... Patreon:   / yannickilcher   Bitcoin (BTC): bc1q49lsw3q325tr58ygf8sudx2dqfguclvngvy2cq Ethereum (ETH): 0x7ad3513E3B8f66799f507Aa7874b1B0eBC7F85e2 Litecoin (LTC): LQW2TRyKYetVC8WjFkhpPhtpbDM4Vw7r9m Monero (XMR): 4ACL8AGrEo5hAir8A9CeVrW8pEauWvnp1WnSDZxW7tziCDLhZAGsgzhRQABDnFy8yuM9fWJDviJPHKRjV4FWt19CJZN9D4n

Follow along using the transcript.

Yannic Kilcher

284K subscribers