Loading video...

Video Failed to Load

There was a problem loading this video. This could be due to a temporary network issue or the video might be unavailable.

Diffusion models leverage a variety of samplers. Deterministic methods like DDIM produce orderly paths. In contrast, stochastic samplers like DDPM produce chaotic trajectories. Despite their differences, both methods draw valid samples from the underlying distribution.

Alec Helbling

11,962 subscribers

103,077 views • 1 year ago •via X (Twitter)

Health & Wellness Education Science & Technology

Anya Rossi• Live Now

Private livecam show

0 Comments

No comments available

Comments from the original post will appear here

Related Videos

Stochastic and deterministic sampling strategies for diffusion models produce strikingly different trajectories, but both ultimately achieve the same aim. I had a great time presenting our work, Diffusion Explorer, this week at IEEE VIS in Vienna.

Stochastic and deterministic sampling strategies for diffusion models produce strikingly different trajectories, but both ultimately achieve the same aim. I had a great time presenting our work, Diffusion Explorer, this week at IEEE VIS in Vienna.

Alec Helbling

279,448 views • 11 months ago

Flow matching produces smooth, deterministic trajectories. In contrast, the sampling process of a diffusion model is chaotic, resembling the random motion of gas particles.

Flow matching produces smooth, deterministic trajectories. In contrast, the sampling process of a diffusion model is chaotic, resembling the random motion of gas particles.

Alec Helbling

437,319 views • 1 year ago

Langevin Monte Carlo allows you to draw samples from a probability distribution using its log gradient ∇ log p(x). By performing a sort of gradient ascent with noise you can navigate around the distribution. Langevin MC is heavily related to modern diffusion models.

Langevin Monte Carlo allows you to draw samples from a probability distribution using its log gradient ∇ log p(x). By performing a sort of gradient ascent with noise you can navigate around the distribution. Langevin MC is heavily related to modern diffusion models.

Alec Helbling

97,001 views • 1 year ago

Israel employs inhumane and degrading methods in the distribution of humanitarian aid to Gaza's residents.

Israel employs inhumane and degrading methods in the distribution of humanitarian aid to Gaza's residents.

TIMES OF GAZA

18,106 views • 1 year ago

There are only a few ways to image the brain without surgery: electricity, magnets, light, sound. All existing methods either produce blurry images or require a room-sized machine. Our image is captured with ultrasound transmitted through the skull, with a contrast agent.

There are only a few ways to image the brain without surgery: electricity, magnets, light, sound. All existing methods either produce blurry images or require a room-sized machine. Our image is captured with ultrasound transmitted through the skull, with a contrast agent.

Aleph

64,823 views • 4 months ago

Veg NI are a cooperative of growers in Northern Ireland promoting their produce in the face of fierce import competition. The brand centres around the idea of ‘Parful Produce’ (‘Powerful Produce’ if you’re not from NI). Find out more on our instagram @jackrenwickstudio

Jack Renwick Studio

26,198 views • 3 years ago

What $30 of fresh produce looks like in El Salvador 🇸🇻

What $30 of fresh produce looks like in El Salvador 🇸🇻

Salvadoran Pride

41,950 views • 2 years ago

I made a tool called Diffusion Explorer that lets you to train and visualize simple 2D diffusion and flow models live in the browser. You can draw your own distributions and observe how the generated samples converge during training. Try it live 👇

I made a tool called Diffusion Explorer that lets you to train and visualize simple 2D diffusion and flow models live in the browser. You can draw your own distributions and observe how the generated samples converge during training. Try it live 👇

Alec Helbling

73,188 views • 1 year ago

What does the endoplasmic reticulum look like in a live cell, thanks to super-resolution microscopy? Nature Methods

What does the endoplasmic reticulum look like in a live cell, thanks to super-resolution microscopy? Nature Methods

Eric Topol

24,062 views • 6 months ago

The Raspy Cricket can produce silk-not from its body like a spider, but from its mouth, to form shelters

The Raspy Cricket can produce silk-not from its body like a spider, but from its mouth, to form shelters

Science girl

147,439 views • 25 days ago

The latent space of earlier generative models like GANS can linearly encode concepts of the data. What if the data was model weights? We present weights2weights, a subspace in diffusion weights that behaves as an interpretable latent space over customized diffusion models.

The latent space of earlier generative models like GANS can linearly encode concepts of the data. What if the data was model weights? We present weights2weights, a subspace in diffusion weights that behaves as an interpretable latent space over customized diffusion models.

Amil Dravid

94,296 views • 2 years ago

Keeping it fresh for 2026 and beyond. Krown Produce and Aston Martin Aramco are excited to announce a multi-year deal, with the Canadian leader in premium produce and pioneer in importing for retail and wholesale distribution.

Keeping it fresh for 2026 and beyond. Krown Produce and Aston Martin Aramco are excited to announce a multi-year deal, with the Canadian leader in premium produce and pioneer in importing for retail and wholesale distribution.

Aston Martin Aramco F1 Team

32,265 views • 1 year ago

Training large models with RL typically runs on a fixed homogeneous cluster of the latest accelerators for numerical stability and faster throughput. This throws away inference optimization work done for other vendors and generations, limits the capacity you can utilize, and is unnecessarily expensive. So we trained a model on agentic coding tasks using fully asynchronous RL and mixed-precision, mixed-accelerator samplers. With off-policy correction, training stayed stable through convergence, on par with homogeneous samplers, at a lower cost.

Training large models with RL typically runs on a fixed homogeneous cluster of the latest accelerators for numerical stability and faster throughput. This throws away inference optimization work done for other vendors and generations, limits the capacity you can utilize, and is unnecessarily expensive. So we trained a model on agentic coding tasks using fully asynchronous RL and mixed-precision, mixed-accelerator samplers. With off-policy correction, training stayed stable through convergence, on par with homogeneous samplers, at a lower cost.

Modal

163,527 views • 11 days ago

If ITV decided to produce an All-Stars series featuring 16 housemates from their four series, what would your cast look like? #BBUK

If ITV decided to produce an All-Stars series featuring 16 housemates from their four series, what would your cast look like? #BBUK

Gabriel 𑁍

134,570 views • 26 days ago

Google is making progress on their diffusion models... It's now as good as a Gemini 2.0 Flash Lite. The writing is on the wall, a majority of language AI use in the future will be diffusion models.

Google is making progress on their diffusion models... It's now as good as a Gemini 2.0 Flash Lite. The writing is on the wall, a majority of language AI use in the future will be diffusion models.

Carlos E. Perez

193,952 views • 9 months ago

Capturing smoke from burning plastic in a bag demonstrates the significant amount of pollution a single fire can produce

Capturing smoke from burning plastic in a bag demonstrates the significant amount of pollution a single fire can produce

Interesting AF

355,793 views • 1 year ago

Block Diffusion: Interpolating Between Autoregressive and Diffusion Language Models "Block diffusion sequentially generates blocks of tokens by performing diffusion within each block and conditioning on previous blocks. By combining strength from autoregressive and diffusion models, block diffusion overcomes the limitations of both approaches by supporting variable-length, higher-quality generation and improving inference efficiency with KV caching and parallel sampling."

Block Diffusion: Interpolating Between Autoregressive and Diffusion Language Models "Block diffusion sequentially generates blocks of tokens by performing diffusion within each block and conditioning on previous blocks. By combining strength from autoregressive and diffusion models, block diffusion overcomes the limitations of both approaches by supporting variable-length, higher-quality generation and improving inference efficiency with KV caching and parallel sampling."

Tanishq Mathew Abraham, Ph.D.

21,813 views • 1 year ago