Hilde Kuehne
@hildekuehne.bsky.social
Professor for CS at the Tuebingen AI Center and affiliated Professor at MIT-IBM Watson AI lab - Multimodal learning and video understanding - GC for ICCV 2025 - https://hildekuehne.github.io/
Looking for something better than LoRA?
Check out CaRA - Canonical Rank Adaptation at #icml2025 next week!
What is it? A multidimensional low-rank update for fine-tuning ViT using Canonical Polyadic (CP) factors
Paper: openreview.net/forum?id=vex...
Check out CaRA - Canonical Rank Adaptation at #icml2025 next week!
What is it? A multidimensional low-rank update for fine-tuning ViT using Canonical Polyadic (CP) factors
Paper: openreview.net/forum?id=vex...
July 7, 2025 at 7:25 AM
Looking for something better than LoRA?
Check out CaRA - Canonical Rank Adaptation at #icml2025 next week!
What is it? A multidimensional low-rank update for fine-tuning ViT using Canonical Polyadic (CP) factors
Paper: openreview.net/forum?id=vex...
Check out CaRA - Canonical Rank Adaptation at #icml2025 next week!
What is it? A multidimensional low-rank update for fine-tuning ViT using Canonical Polyadic (CP) factors
Paper: openreview.net/forum?id=vex...
one for thinking, one for fun …
July 5, 2025 at 2:19 PM
one for thinking, one for fun …
Happening now! Check out the great work from Felix and Co. We improve video action grounding by >=10% on V-HICO and DALY(hope we didn't miss anyone)!
Fri 13 Jun 10:30 a.m. CDT — 12:30 p.m. CDT
ExHall D Poster #306
Paper: openaccess.thecvf.com/content/CVPR...
Fri 13 Jun 10:30 a.m. CDT — 12:30 p.m. CDT
ExHall D Poster #306
Paper: openaccess.thecvf.com/content/CVPR...
June 13, 2025 at 3:58 PM
Happening now! Check out the great work from Felix and Co. We improve video action grounding by >=10% on V-HICO and DALY(hope we didn't miss anyone)!
Fri 13 Jun 10:30 a.m. CDT — 12:30 p.m. CDT
ExHall D Poster #306
Paper: openaccess.thecvf.com/content/CVPR...
Fri 13 Jun 10:30 a.m. CDT — 12:30 p.m. CDT
ExHall D Poster #306
Paper: openaccess.thecvf.com/content/CVPR...
#cvpr2025 in Paris satellite event kicking off!
Thanks so much for organizing! @davidpicard.bsky.social @vickykalogeiton.bsky.social and @matthieucord.bsky.social
Thanks so much for organizing! @davidpicard.bsky.social @vickykalogeiton.bsky.social and @matthieucord.bsky.social
June 6, 2025 at 8:05 AM
#cvpr2025 in Paris satellite event kicking off!
Thanks so much for organizing! @davidpicard.bsky.social @vickykalogeiton.bsky.social and @matthieucord.bsky.social
Thanks so much for organizing! @davidpicard.bsky.social @vickykalogeiton.bsky.social and @matthieucord.bsky.social
💡 Our approach:
➡️Use VLMs to generate frame-wise video descriptions
➡️Use LLMs to filter concepts: objects, verbs, activities (see example below)
➡️Analyze three types of bias:
🔹Concept bias (objects only? verbs only? )
🔹Temporal bias (single frame enough?)
🔹Common sense vs dataset bias
➡️Use VLMs to generate frame-wise video descriptions
➡️Use LLMs to filter concepts: objects, verbs, activities (see example below)
➡️Analyze three types of bias:
🔹Concept bias (objects only? verbs only? )
🔹Temporal bias (single frame enough?)
🔹Common sense vs dataset bias
April 29, 2025 at 3:57 PM
💡 Our approach:
➡️Use VLMs to generate frame-wise video descriptions
➡️Use LLMs to filter concepts: objects, verbs, activities (see example below)
➡️Analyze three types of bias:
🔹Concept bias (objects only? verbs only? )
🔹Temporal bias (single frame enough?)
🔹Common sense vs dataset bias
➡️Use VLMs to generate frame-wise video descriptions
➡️Use LLMs to filter concepts: objects, verbs, activities (see example below)
➡️Analyze three types of bias:
🔹Concept bias (objects only? verbs only? )
🔹Temporal bias (single frame enough?)
🔹Common sense vs dataset bias
Without going too deep into the discussion and just bc it's so funny and tragic ... here is wikipedia defining AGI as systems that "surpass human capabilities across most or all economically valuable cognitive work" 🤣 ... we just got tariffs for penguins. Do we really need to argue?
April 5, 2025 at 9:21 AM
Without going too deep into the discussion and just bc it's so funny and tragic ... here is wikipedia defining AGI as systems that "surpass human capabilities across most or all economically valuable cognitive work" 🤣 ... we just got tariffs for penguins. Do we really need to argue?
Hi everyone working towards the #ICCV2025 deadline!
Remember to sign off the checklist before going to 🛌 or going 🏠!
Two more points on personal experience:
- NO! acknowledgements to funding sources or urls which can be considered identifying information.
-Entered the paper ID in latex?
Remember to sign off the checklist before going to 🛌 or going 🏠!
Two more points on personal experience:
- NO! acknowledgements to funding sources or urls which can be considered identifying information.
-Entered the paper ID in latex?
March 7, 2025 at 7:28 PM
Hi everyone working towards the #ICCV2025 deadline!
Remember to sign off the checklist before going to 🛌 or going 🏠!
Two more points on personal experience:
- NO! acknowledgements to funding sources or urls which can be considered identifying information.
-Entered the paper ID in latex?
Remember to sign off the checklist before going to 🛌 or going 🏠!
Two more points on personal experience:
- NO! acknowledgements to funding sources or urls which can be considered identifying information.
-Entered the paper ID in latex?
To keep in mind …
January 21, 2025 at 10:00 PM
To keep in mind …
Celebrating @neuripsconf.bsky.social … 10 years on ☕️
December 10, 2024 at 8:43 AM
Celebrating @neuripsconf.bsky.social … 10 years on ☕️
November 30, 2024 at 5:25 PM
Just to provide some content and get everyone in the mood for NeurIPS ✈️, we have a very nice, unusual CV paper there:
Convolutional Differentiable Logic Gate Networks
It's what it says - a ConvNet builds exclusively from logic gates (the basic stuff).
And Felix did a great video:
Convolutional Differentiable Logic Gate Networks
It's what it says - a ConvNet builds exclusively from logic gates (the basic stuff).
And Felix did a great video:
November 22, 2024 at 2:30 PM
Just to provide some content and get everyone in the mood for NeurIPS ✈️, we have a very nice, unusual CV paper there:
Convolutional Differentiable Logic Gate Networks
It's what it says - a ConvNet builds exclusively from logic gates (the basic stuff).
And Felix did a great video:
Convolutional Differentiable Logic Gate Networks
It's what it says - a ConvNet builds exclusively from logic gates (the basic stuff).
And Felix did a great video: