Anirban Ray
@anirbanray.bsky.social
PhD Student working on bioimaging inverse problems with @florianjug.bsky.social at @humantechnopole.bsky.social + @tudresden.bsky.social | Prev: computer vision Hitachi R&D, Tokyo.
🔗: https://rayanirban.github.io/
Likes 🏸🏋️🏔️🏓 and ✈️
🔗: https://rayanirban.github.io/
Likes 🏸🏋️🏔️🏓 and ✈️
Pinned
Anirban Ray
@anirbanray.bsky.social
· Nov 18
So exciting to see everyone here. A brief self-intro for others:
I am currently a PhD student with @florianjug.bsky.social. My work involves the application of AI in bioimage analysis. I am interested in the application of #GenAI to solve inverse problems with #DiffusionModels #FlowMatching #VAEs.
I am currently a PhD student with @florianjug.bsky.social. My work involves the application of AI in bioimage analysis. I am interested in the application of #GenAI to solve inverse problems with #DiffusionModels #FlowMatching #VAEs.
Reposted by Anirban Ray
💥 DeepInverse is now part of the official PyTorch Landscape💥
We are excited to join an ecosystem of great open-source AI libraries, including @hf.co diffusers, MONAI, einops, etc.
pytorch.org/blog/deepinv...
We are excited to join an ecosystem of great open-source AI libraries, including @hf.co diffusers, MONAI, einops, etc.
pytorch.org/blog/deepinv...
DeepInverse Joins the PyTorch Ecosystem: the library for solving imaging inverse problems with deep learning – PyTorch
pytorch.org
November 5, 2025 at 5:31 PM
💥 DeepInverse is now part of the official PyTorch Landscape💥
We are excited to join an ecosystem of great open-source AI libraries, including @hf.co diffusers, MONAI, einops, etc.
pytorch.org/blog/deepinv...
We are excited to join an ecosystem of great open-source AI libraries, including @hf.co diffusers, MONAI, einops, etc.
pytorch.org/blog/deepinv...
Reposted by Anirban Ray
I’m at #AIS25 today and tomorrow…
Where scientists, industry leaders, investors, and policymakers meet to explore the transformative impact of artificial intelligence on scientific discovery.
I think this is a very important conversation we must have NOW! 👍
Where scientists, industry leaders, investors, and policymakers meet to explore the transformative impact of artificial intelligence on scientific discovery.
I think this is a very important conversation we must have NOW! 👍
November 3, 2025 at 1:27 PM
I’m at #AIS25 today and tomorrow…
Where scientists, industry leaders, investors, and policymakers meet to explore the transformative impact of artificial intelligence on scientific discovery.
I think this is a very important conversation we must have NOW! 👍
Where scientists, industry leaders, investors, and policymakers meet to explore the transformative impact of artificial intelligence on scientific discovery.
I think this is a very important conversation we must have NOW! 👍
Reposted by Anirban Ray
Anirban Ray, Vera Galinova, Florian Jug
ResMatching: Noise-Resilient Computational Super-Resolution via Guided Conditional Flow Matching
https://arxiv.org/abs/2510.26601
ResMatching: Noise-Resilient Computational Super-Resolution via Guided Conditional Flow Matching
https://arxiv.org/abs/2510.26601
October 31, 2025 at 6:41 AM
Anirban Ray, Vera Galinova, Florian Jug
ResMatching: Noise-Resilient Computational Super-Resolution via Guided Conditional Flow Matching
https://arxiv.org/abs/2510.26601
ResMatching: Noise-Resilient Computational Super-Resolution via Guided Conditional Flow Matching
https://arxiv.org/abs/2510.26601
super cool work😎
Generative Point Tracking with Flow Matching
My latest project with Adam W. Harley, @csprofkgd.bsky.social, Derek Nowrouzezahrai, @chrisjpal.bsky.social.
Project page: mtesfaldet.net/genpt_projpa...
Paper: arxiv.org/abs/2510.20951
Code: github.com/tesfaldet/ge...
My latest project with Adam W. Harley, @csprofkgd.bsky.social, Derek Nowrouzezahrai, @chrisjpal.bsky.social.
Project page: mtesfaldet.net/genpt_projpa...
Paper: arxiv.org/abs/2510.20951
Code: github.com/tesfaldet/ge...
October 27, 2025 at 7:13 PM
super cool work😎
“We may not win every battle, but we will win the war.” --- Such an appropriate characterization for posterior samplers. Each posterior sample fights its own battle against noise and degradation; some win, some lose. But the MMSE estimate wins the war 😉.
#iykuk #ImageRestoration
#iykuk #ImageRestoration
October 19, 2025 at 9:33 AM
“We may not win every battle, but we will win the war.” --- Such an appropriate characterization for posterior samplers. Each posterior sample fights its own battle against noise and degradation; some win, some lose. But the MMSE estimate wins the war 😉.
#iykuk #ImageRestoration
#iykuk #ImageRestoration
Reposted by Anirban Ray
Diffusion Transformers with Representation Autoencoders by Boyang Zheng, et al (arxiv.org/abs/2510.116...)
Unexpected result: swapping the SD-VAE for a pretrained visual encoder improves FID, challenging the idea that encoders' information compression is not suited for generative modeling!
Unexpected result: swapping the SD-VAE for a pretrained visual encoder improves FID, challenging the idea that encoders' information compression is not suited for generative modeling!
October 14, 2025 at 7:08 PM
Diffusion Transformers with Representation Autoencoders by Boyang Zheng, et al (arxiv.org/abs/2510.116...)
Unexpected result: swapping the SD-VAE for a pretrained visual encoder improves FID, challenging the idea that encoders' information compression is not suited for generative modeling!
Unexpected result: swapping the SD-VAE for a pretrained visual encoder improves FID, challenging the idea that encoders' information compression is not suited for generative modeling!
👏👏👏
Happy to share that ShapeEmbed has been accepted at @neuripsconf.bsky.social 🎉 SE is self-supervised framework to encode 2D contours from microscopy & natural images into a latent representation invariant to translation, scaling, rotation, reflection & point indexing
📄 arxiv.org/pdf/2507.01009 (1/N)
📄 arxiv.org/pdf/2507.01009 (1/N)
September 23, 2025 at 8:39 AM
👏👏👏
Reposted by Anirban Ray
We had an awesome #OMIBS2025
Thanks to all the lecturers, staff members, vendor faculty, sponsors, and participants for making this an amazing course year!
Thanks to all the lecturers, staff members, vendor faculty, sponsors, and participants for making this an amazing course year!
August 26, 2025 at 6:05 PM
We had an awesome #OMIBS2025
Thanks to all the lecturers, staff members, vendor faculty, sponsors, and participants for making this an amazing course year!
Thanks to all the lecturers, staff members, vendor faculty, sponsors, and participants for making this an amazing course year!
Reposted by Anirban Ray
Introducing Latent-X — our all-atom frontier AI model for protein binder design.
State-of-the-art lab performance, widely accessible via the Latent Labs Platform.
Free tier: platform.latentlabs.com
Blog: latentlabs.com/latent-x/
Technical report: tinyurl.com/latent-X
State-of-the-art lab performance, widely accessible via the Latent Labs Platform.
Free tier: platform.latentlabs.com
Blog: latentlabs.com/latent-x/
Technical report: tinyurl.com/latent-X
July 22, 2025 at 6:21 AM
Introducing Latent-X — our all-atom frontier AI model for protein binder design.
State-of-the-art lab performance, widely accessible via the Latent Labs Platform.
Free tier: platform.latentlabs.com
Blog: latentlabs.com/latent-x/
Technical report: tinyurl.com/latent-X
State-of-the-art lab performance, widely accessible via the Latent Labs Platform.
Free tier: platform.latentlabs.com
Blog: latentlabs.com/latent-x/
Technical report: tinyurl.com/latent-X
Reposted by Anirban Ray
New episode in this line of work from @giannisdaras.bsky.social et al. on training diffusion models with mostly bad/low-quality/corrupted data (+few high-quality samples). This time for proteins!
📄 Ambient diffusion Omni: arxiv.org/pdf/2506.10038
📄 Ambient Proteins: www.biorxiv.org/content/10.1...
📄 Ambient diffusion Omni: arxiv.org/pdf/2506.10038
📄 Ambient Proteins: www.biorxiv.org/content/10.1...
July 7, 2025 at 7:43 PM
New episode in this line of work from @giannisdaras.bsky.social et al. on training diffusion models with mostly bad/low-quality/corrupted data (+few high-quality samples). This time for proteins!
📄 Ambient diffusion Omni: arxiv.org/pdf/2506.10038
📄 Ambient Proteins: www.biorxiv.org/content/10.1...
📄 Ambient diffusion Omni: arxiv.org/pdf/2506.10038
📄 Ambient Proteins: www.biorxiv.org/content/10.1...
these are so beautiful 🤩.
June 24, 2025 at 8:56 AM
these are so beautiful 🤩.
Reposted by Anirban Ray
New paper on the generalization of Flow Matching www.arxiv.org/abs/2506.03719
🤯 Why does flow matching generalize? Did you know that the flow matching target you're trying to learn *can only generate training points*?
w @quentinbertrand.bsky.social @annegnx.bsky.social @remiemonet.bsky.social 👇👇👇
🤯 Why does flow matching generalize? Did you know that the flow matching target you're trying to learn *can only generate training points*?
w @quentinbertrand.bsky.social @annegnx.bsky.social @remiemonet.bsky.social 👇👇👇
June 18, 2025 at 8:08 AM
New paper on the generalization of Flow Matching www.arxiv.org/abs/2506.03719
🤯 Why does flow matching generalize? Did you know that the flow matching target you're trying to learn *can only generate training points*?
w @quentinbertrand.bsky.social @annegnx.bsky.social @remiemonet.bsky.social 👇👇👇
🤯 Why does flow matching generalize? Did you know that the flow matching target you're trying to learn *can only generate training points*?
w @quentinbertrand.bsky.social @annegnx.bsky.social @remiemonet.bsky.social 👇👇👇
Reposted by Anirban Ray
Spotiflow, our deep learning based spot detection method for microscopy, is now published in @natmethods.nature.com!
Since the pre-print, we have added many features, notably native 3D detection!
@maweigert.bsky.social @gioelelamanno.bsky.social @epfl-brainmind.bsky.social
Paper: rdcu.be/epIB7
(1/N)
Since the pre-print, we have added many features, notably native 3D detection!
@maweigert.bsky.social @gioelelamanno.bsky.social @epfl-brainmind.bsky.social
Paper: rdcu.be/epIB7
(1/N)
Spotiflow: accurate and efficient spot detection for fluorescence microscopy with deep stereographic flow regression
Nature Methods - Spotiflow uses deep learning for subpixel-accurate spot detection in diverse 2D and 3D images. The improved accuracy offered by Spotiflow enables improved biological insights in...
rdcu.be
June 6, 2025 at 6:58 PM
Spotiflow, our deep learning based spot detection method for microscopy, is now published in @natmethods.nature.com!
Since the pre-print, we have added many features, notably native 3D detection!
@maweigert.bsky.social @gioelelamanno.bsky.social @epfl-brainmind.bsky.social
Paper: rdcu.be/epIB7
(1/N)
Since the pre-print, we have added many features, notably native 3D detection!
@maweigert.bsky.social @gioelelamanno.bsky.social @epfl-brainmind.bsky.social
Paper: rdcu.be/epIB7
(1/N)
Really cool and clean idea 😇👏
George Stoica, Vivek Ramanujan, Xiang Fan, Ali Farhadi, Ranjay Krishna, Judy Hoffman
Contrastive Flow Matching
https://arxiv.org/abs/2506.05350
Contrastive Flow Matching
https://arxiv.org/abs/2506.05350
June 6, 2025 at 4:29 PM
Really cool and clean idea 😇👏
Reposted by Anirban Ray
June 6, 2025 at 3:14 PM
Reposted by Anirban Ray
Kullback–Leibler (KL) divergence is a cornerstone of machine learning.
We use it everywhere, from training classifiers and distilling knowledge from models, to learning generative models and aligning LLMs.
BUT, what does it mean, and how do we (actually) compute it?
Video: youtu.be/tXE23653JrU
We use it everywhere, from training classifiers and distilling knowledge from models, to learning generative models and aligning LLMs.
BUT, what does it mean, and how do we (actually) compute it?
Video: youtu.be/tXE23653JrU
June 4, 2025 at 2:58 PM
Kullback–Leibler (KL) divergence is a cornerstone of machine learning.
We use it everywhere, from training classifiers and distilling knowledge from models, to learning generative models and aligning LLMs.
BUT, what does it mean, and how do we (actually) compute it?
Video: youtu.be/tXE23653JrU
We use it everywhere, from training classifiers and distilling knowledge from models, to learning generative models and aligning LLMs.
BUT, what does it mean, and how do we (actually) compute it?
Video: youtu.be/tXE23653JrU
Always found the connection between rectified flows and optimal transport a bit confusing. Glad to see some clarity. TL;DR: Rectified flows ≠ optimal transport. arxiv.org/abs/2505.19712
On the Relation between Rectified Flows and Optimal Transport
This paper investigates the connections between rectified flows, flow matching, and optimal transport. Flow matching is a recent approach to learning generative models by estimating velocity fields th...
arxiv.org
May 27, 2025 at 1:48 PM
Always found the connection between rectified flows and optimal transport a bit confusing. Glad to see some clarity. TL;DR: Rectified flows ≠ optimal transport. arxiv.org/abs/2505.19712
Reposted by Anirban Ray
"Energy Matching: Unifying Flow Matching and
Energy-Based Models for Generative Modeling" by Michal Balcerak et al. arxiv.org/abs/2504.10612
I'm not sure EBM will beat flow-matching/diffusion models, but this article is very refreshing.
Energy-Based Models for Generative Modeling" by Michal Balcerak et al. arxiv.org/abs/2504.10612
I'm not sure EBM will beat flow-matching/diffusion models, but this article is very refreshing.
May 21, 2025 at 12:37 PM
"Energy Matching: Unifying Flow Matching and
Energy-Based Models for Generative Modeling" by Michal Balcerak et al. arxiv.org/abs/2504.10612
I'm not sure EBM will beat flow-matching/diffusion models, but this article is very refreshing.
Energy-Based Models for Generative Modeling" by Michal Balcerak et al. arxiv.org/abs/2504.10612
I'm not sure EBM will beat flow-matching/diffusion models, but this article is very refreshing.
Reposted by Anirban Ray
New Feature in DeepInverse (deepinv.github.io):
🚀 Custom Diffusion Solver Design
DeepInverse now simplifies the process with:
✔ Standard SDEs (VP, VE, etc.)
✔ Pretrained denoisers for multiple noise levels
✔ ODE/SDE solvers (Euler, Heun)
✔ Noisy data fidelity terms for guidance
🚀 Custom Diffusion Solver Design
DeepInverse now simplifies the process with:
✔ Standard SDEs (VP, VE, etc.)
✔ Pretrained denoisers for multiple noise levels
✔ ODE/SDE solvers (Euler, Heun)
✔ Noisy data fidelity terms for guidance
Redirecting to https://deepinv.github.io/deepinv/
deepinv.github.io
May 19, 2025 at 2:48 PM
New Feature in DeepInverse (deepinv.github.io):
🚀 Custom Diffusion Solver Design
DeepInverse now simplifies the process with:
✔ Standard SDEs (VP, VE, etc.)
✔ Pretrained denoisers for multiple noise levels
✔ ODE/SDE solvers (Euler, Heun)
✔ Noisy data fidelity terms for guidance
🚀 Custom Diffusion Solver Design
DeepInverse now simplifies the process with:
✔ Standard SDEs (VP, VE, etc.)
✔ Pretrained denoisers for multiple noise levels
✔ ODE/SDE solvers (Euler, Heun)
✔ Noisy data fidelity terms for guidance
Reposted by Anirban Ray
We will be kicking off the Neurogenomics Conference with two sessions on Neurodevelopment with talks from Wieland Huttner, @bassemh.bsky.social, @mareikealbert.bsky.social, @naelnadifkasri-lab.bsky.social, Yukiko Gotoh, @boyanbonev.bsky.social and others! #neurogen25
Monday is the big day! Very much looking forward to welcoming all participants and speakers of the Neurogenomics Conference to @humantechnopole.bsky.social in Milan. It promises to be an exciting few days filled with amazing science.
May 19, 2025 at 6:24 AM
We will be kicking off the Neurogenomics Conference with two sessions on Neurodevelopment with talks from Wieland Huttner, @bassemh.bsky.social, @mareikealbert.bsky.social, @naelnadifkasri-lab.bsky.social, Yukiko Gotoh, @boyanbonev.bsky.social and others! #neurogen25
Really interesting paper on per-frequency control in diffusion models: arxiv.org/abs/2505.112.... They tackle the frequency degradation rate imbalance of the forward process by enforcing Equal SNR across Fourier components. Finally, someone’s taking @sedielem.bsky.social blogs to heart 👀🔥
A Fourier Space Perspective on Diffusion Models
Diffusion models are state-of-the-art generative models on data modalities such as images, audio, proteins and materials. These modalities share the property of exponentially decaying variance and mag...
arxiv.org
May 19, 2025 at 6:49 AM
Really interesting paper on per-frequency control in diffusion models: arxiv.org/abs/2505.112.... They tackle the frequency degradation rate imbalance of the forward process by enforcing Equal SNR across Fourier components. Finally, someone’s taking @sedielem.bsky.social blogs to heart 👀🔥
Reposted by Anirban Ray
Here's the third and final part of Slater Stich's "History of diffusion" interview series!
The other two interviewees' research played a pivotal role in the rise of diffusion models, whereas I just like to yap about them 😬 this was a wonderful opportunity to do exactly that!
The other two interviewees' research played a pivotal role in the rise of diffusion models, whereas I just like to yap about them 😬 this was a wonderful opportunity to do exactly that!
History of Diffusion - Sander Dieleman
YouTube video by Bain Capital Ventures
www.youtube.com
May 14, 2025 at 4:11 PM
Here's the third and final part of Slater Stich's "History of diffusion" interview series!
The other two interviewees' research played a pivotal role in the rise of diffusion models, whereas I just like to yap about them 😬 this was a wonderful opportunity to do exactly that!
The other two interviewees' research played a pivotal role in the rise of diffusion models, whereas I just like to yap about them 😬 this was a wonderful opportunity to do exactly that!
Reposted by Anirban Ray
We are proudly hosting
𝐃𝐫. 𝐓𝐚𝐥𝐥𝐞𝐲 𝐋𝐚𝐦𝐛𝐞𝐫𝐭 @talley.codes (CITE, Harvard Medical School):
𝐒𝐢𝐦𝐮𝐥𝐚𝐭𝐢𝐧𝐠 𝐫𝐞𝐚𝐥𝐢𝐬𝐭𝐢𝐜 𝐥𝐢𝐠𝐡𝐭 𝐦𝐢𝐜𝐫𝐨𝐬𝐜𝐨𝐩𝐲 𝐢𝐦𝐚𝐠𝐞𝐬 𝐮𝐬𝐢𝐧𝐠 𝐦𝐢𝐜𝐫𝐨𝐬𝐢𝐦
When: 𝐌𝐚𝐲 𝟐𝟗𝐭𝐡 @ 𝟏𝟏 𝐚𝐦 𝐄𝐃𝐓 (Boston time)
Join us on Zoom! harvard.zoom.us/j/9762343464...
𝐃𝐫. 𝐓𝐚𝐥𝐥𝐞𝐲 𝐋𝐚𝐦𝐛𝐞𝐫𝐭 @talley.codes (CITE, Harvard Medical School):
𝐒𝐢𝐦𝐮𝐥𝐚𝐭𝐢𝐧𝐠 𝐫𝐞𝐚𝐥𝐢𝐬𝐭𝐢𝐜 𝐥𝐢𝐠𝐡𝐭 𝐦𝐢𝐜𝐫𝐨𝐬𝐜𝐨𝐩𝐲 𝐢𝐦𝐚𝐠𝐞𝐬 𝐮𝐬𝐢𝐧𝐠 𝐦𝐢𝐜𝐫𝐨𝐬𝐢𝐦
When: 𝐌𝐚𝐲 𝟐𝟗𝐭𝐡 @ 𝟏𝟏 𝐚𝐦 𝐄𝐃𝐓 (Boston time)
Join us on Zoom! harvard.zoom.us/j/9762343464...
May 7, 2025 at 3:59 PM
We are proudly hosting
𝐃𝐫. 𝐓𝐚𝐥𝐥𝐞𝐲 𝐋𝐚𝐦𝐛𝐞𝐫𝐭 @talley.codes (CITE, Harvard Medical School):
𝐒𝐢𝐦𝐮𝐥𝐚𝐭𝐢𝐧𝐠 𝐫𝐞𝐚𝐥𝐢𝐬𝐭𝐢𝐜 𝐥𝐢𝐠𝐡𝐭 𝐦𝐢𝐜𝐫𝐨𝐬𝐜𝐨𝐩𝐲 𝐢𝐦𝐚𝐠𝐞𝐬 𝐮𝐬𝐢𝐧𝐠 𝐦𝐢𝐜𝐫𝐨𝐬𝐢𝐦
When: 𝐌𝐚𝐲 𝟐𝟗𝐭𝐡 @ 𝟏𝟏 𝐚𝐦 𝐄𝐃𝐓 (Boston time)
Join us on Zoom! harvard.zoom.us/j/9762343464...
𝐃𝐫. 𝐓𝐚𝐥𝐥𝐞𝐲 𝐋𝐚𝐦𝐛𝐞𝐫𝐭 @talley.codes (CITE, Harvard Medical School):
𝐒𝐢𝐦𝐮𝐥𝐚𝐭𝐢𝐧𝐠 𝐫𝐞𝐚𝐥𝐢𝐬𝐭𝐢𝐜 𝐥𝐢𝐠𝐡𝐭 𝐦𝐢𝐜𝐫𝐨𝐬𝐜𝐨𝐩𝐲 𝐢𝐦𝐚𝐠𝐞𝐬 𝐮𝐬𝐢𝐧𝐠 𝐦𝐢𝐜𝐫𝐨𝐬𝐢𝐦
When: 𝐌𝐚𝐲 𝟐𝟗𝐭𝐡 @ 𝟏𝟏 𝐚𝐦 𝐄𝐃𝐓 (Boston time)
Join us on Zoom! harvard.zoom.us/j/9762343464...
Reposted by Anirban Ray
Very cool article from Panagiotis Theodoropoulos et al: https://arxiv.org/abs/2410.14055
Feedback Schrödinger Bridge Matching introduces a new method to improve transfer between two data distributions using only a small number of paired samples!
Feedback Schrödinger Bridge Matching introduces a new method to improve transfer between two data distributions using only a small number of paired samples!
April 25, 2025 at 5:03 PM
Very cool article from Panagiotis Theodoropoulos et al: https://arxiv.org/abs/2410.14055
Feedback Schrödinger Bridge Matching introduces a new method to improve transfer between two data distributions using only a small number of paired samples!
Feedback Schrödinger Bridge Matching introduces a new method to improve transfer between two data distributions using only a small number of paired samples!
Reposted by Anirban Ray
One weird trick for better diffusion models: concatenate some DINOv2 features to your latent channels!
Combining latents with PCA components extracted from DINOv2 features yields faster training and better samples. Also enables a new guidance strategy. Simple and effective!
Combining latents with PCA components extracted from DINOv2 features yields faster training and better samples. Also enables a new guidance strategy. Simple and effective!
1/n Introducing ReDi (Representation Diffusion): a new generative approach that leverages a diffusion model to jointly capture
– Low-level image details (via VAE latents)
– High-level semantic features (via DINOv2)🧵
– Low-level image details (via VAE latents)
– High-level semantic features (via DINOv2)🧵
April 25, 2025 at 1:03 PM
One weird trick for better diffusion models: concatenate some DINOv2 features to your latent channels!
Combining latents with PCA components extracted from DINOv2 features yields faster training and better samples. Also enables a new guidance strategy. Simple and effective!
Combining latents with PCA components extracted from DINOv2 features yields faster training and better samples. Also enables a new guidance strategy. Simple and effective!