Raymond A. Yeh

@jbhuang0604 Change import numpy to import torch😎. Jokes aside, I would probably introduce DL methods by removing details of mid-level representation. E.g., teach edge detection but not the variants, and introduce HED instead. Assuming some DL background of students…

Teaching undergrad computer vision next semester! BUT, I am a bit confused about what to teach, given that so many earlier attempts are outdated and replaced by learning-based methods in practice. 🤔

Worried that open-weight models may be distilled from unauthorized sources? We present Knowledge Distillation Detection for Open-Weights Models at #NeurIPS2025, introducing a framework for determining whether an open-weights model was distilled from another model.

[1/3] 🚀 Introducing ShapeGen4D: video → high-quality 4D mesh sequences. A native, end-to-end video-to-4D model that turns monocular videos into high-quality mesh sequences (no per-frame optimization). details 👉 shapegen4d.github.io

The Lillian Gilbreth Postdoctoral Fellowship at Purdue is Accepting Applications! Deadline: Oct. 10, 2025 Together with @purduesong , we have a project on AI and 3D Imaging for Real-Time Anomaly Detection. - To learn more: engineering.purdue.edu/Engr/Research/…

For more details, please see Project page: amberyzheng.com/immu_cond_num/ Code: github.com/amberyzheng/mo…

Open‐weight models can be misused after release. With a few fine‐tuning steps, one can enable jailbreaks or introduce undesirable content to the model. We ask whether it is possible to make models harder to fine-tune (“immunized”) on these attacks but not on other tasks.

Tomorrow, we are presenting “Model Immunization from a Condition Number Perspective” at ICML: 📢Oral: Jul 17, 1:45–2:00 p.m. EDT @ West Exhib. Hall C 📌Poster: 2:00–4:30 p.m. EDT @ East Exhib. Hall A-B (E-1604) Come talk to Cedar and learn more about reducing model misuse!

🚨 We are presenting our ICML 2025 oral next week: Model Immunization from a Condition Number Perspective 📄 amberyzheng.com/immu_cond_num/ (1/n)

🔥Time to Upgrade Your Classifier-Free Guidance🔥 🌠CFG-Zero*🌠 offers consistently better *visual quality* and *text alignment* on text-to-image/video - Project: weichenfan.github.io/webpage-cfg-ze… - Code: github.com/WeichenFan/CFG… - Demo @Gradio: huggingface.co/spaces/weepies… . Thanks @_akhaliq!

We are excited to share our #NeurIPS2024 work "Multi-Object 3D Grounding with Dynamic Modules and Language Informed Spatial Attention" (D-LISA). Please drop by our poster session on Dec 11th 11 a.m. to 2 p.m. (Poster #1501) and learn from @cajoeyang on Multi-Object 3D Grounding!

Excited to present our work on CoDA-NO at #NeurIPS2024 We develop a novel neural operator architecture designed to solve coupled partial differential equations (PDEs) in multiphysics systems. Unlike existing approaches that are limited to fixed sets of variables, CoDA-NO can handle varying combinations of physical variables by tokenizing functions along their codomain space. It extends transformer architecture concepts like positional encoding, self-attention, and normalization to function spaces, allowing it to learn representations of different PDE systems with a single model. We demonstrate CoDA-NO's effectiveness through experiments on complex multiphysics problems like fluid-structure interactions and Rayleigh-Bénard convection. Using a two-stage approach of self-supervised pretraining followed by few-shot supervised finetuning, CoDA-NO outperforms existing methods by over 36% on these challenging tasks. The architecture shows strong generalization capabilities, being able to adapt to new physical variables and geometries not seen during training, while maintaining discretization convergence properties that make it resolution-agnostic. Come view our poster on Wednesday the 11th from 11 am to 2 pm PST. Links: Paper: arxiv.org/abs/2403.125532 Poster: neurips.cc/virtual/2024/p… Codebase: github.com/neuraloperator… Authors: @Ashiq_Rahman_s @Robertljg, Mogab Elleithy, @DanielLeibovici, @ZongyiLiCaltech, Boris Bonev, @crwhite_ml, @julberner, @RaymondYeh, @JeanKossaifi, @Azizzadenesheli

Excited to share that I'll be joining University of California at Irvine as a CS faculty in '25!🌟 Faculty apps: @_krishna_murthy, @liuzhuang1234 & I share our tips: unnat.github.io/notes/Hidden_C… PhD apps: I'm looking for students in vision, robot learning, & AI4Science. Details👇

@unnatjain2010 @_krishna_murthy @liuzhuang1234 Congrats!

@sirbayes Agree! Similar reasoning from the CVPR template, "It is important for readers to be able to refer to any particular equation. Just because you did not refer to it in the text does not mean some future reader might not need to refer to it."

This looks like an excerpt from my book probml.github.io/pml-book/book2… (p253). I intentionally assigned numbers to every equation so people can refer to them (eg student: "what is the const term in 5.170?" or github issue: "eqn 5.171 is wrong").

@eccvconf @maharajamihir To confirm. Suppose two authors (and one is a student) share two papers, then two full onsite registrations are required.

@maharajamihir Each paper must be attached to at least one full, onsite registration.

@eccvconf, does only one author need to be registered under a full registration type, and can the co-authors have a Student registration?

We will be presenting "Alpha Invariance: On Inverse Scaling Between Distance and Volume Density in Neural Radiance Fields" in the morning session poster #83. Come and learn more about how activation functions and initialization affect NeRFs! #CVPR2024

We will be presenting "Making Vision Transformers Truly Shift-Equivariant" in the afternoon session poster #70. Come and learn more from @renanrojasg! #CVPR2024

I will be giving an oral presentation on "Towards Safer AI Content Creation by Immunizing Text-to-Image Models" at #AI4CC @CVPR 2024 on June 17th at 13:40. Our poster session is the same day from 17:15 to 18:00. Excited to share our work! @RaymondYeh

@stanley_h_chan Congrats!

Happy to announce that I am promoted to full professor. Big shout out to my former and current students without whom this would never be possible. purdue.edu/newsroom/purdu…

@sirbayes Any pointers to why these models are considered semi-parametric? Thanks!

Sora from OpenAI and Gemini 1.5 from Google are both super impressive. Both rely on long context transformers. I think the lesson is that semi parametric models are great - condition on all past data, whether real or self generated, just like a good Bayesian :)