OmnAI Lab

42 posts

OmnAI Lab banner
OmnAI Lab

OmnAI Lab

@OmnAI_Lab

OmnAI lab, @SapienzaRoma, Computer Science Department (DI) PI: @_iAc

Rome Katılım Ekim 2025
85 Takip Edilen67 Takipçiler
Sabitlenmiş Tweet
OmnAI Lab
OmnAI Lab@OmnAI_Lab·
We are at #ICLR2026 🇧🇷 presenting 5 papers spread across the main conference, 23-24-25 April. Stop by if you are interested in trustworthy and safe AI, generative models, robustness, and model inversion with @Hussain68018934 @BrigliaRosaria @adrianrminut Dario, and Hazem
OmnAI Lab tweet media
English
0
6
13
341
OmnAI Lab retweetledi
Iacopo Masi
Iacopo Masi@_iAc·
🧠 Excited to announce 𝕌&𝕄𝔼 𝟚𝟘𝟚𝟞— 𝘁𝗵𝗲 𝟯𝗿𝗱 𝗪𝗼𝗿𝗸𝘀𝗵𝗼𝗽 & 𝗖𝗵𝗮𝗹𝗹𝗲𝗻𝗴𝗲 𝗼𝗻 𝗨𝗻𝗹𝗲𝗮𝗿𝗻𝗶𝗻𝗴 𝗮𝗻𝗱 𝗠𝗼𝗱𝗲𝗹 𝗘𝗱𝗶𝘁𝗶𝗻𝗴 @eccvconf Organized with @Hussain68018934 @THassner @thakralkartik78 @BrigliaRosaria @MayankVatsa3 @dgolano @sijialiu17 📅 September 8–9, 2026 📍 Malmö, Sweden How do we fix, update & align large generative models — without retraining from scratch? 🎤 Confirmed Speakers: Yezhou Yang — Arizona State University Fabio Galasso — Sapienza University of Rome William Shen — University of Cambridge More speakers coming soon! Stay tuned for submission details 👇 sites.google.com/view/u-and-me-… #ECCV2026 #MachineLearning #ModelEditing #MachineUnlearning #GenerativeAI #DeepLearning
Iacopo Masi tweet media
English
0
7
11
1.2K
OmnAI Lab retweetledi
Richard Sutton
Richard Sutton@RichardSSutton·
The bitter lesson in 26 words: Don’t be distracted by human knowledge, as AI has been historically. Instead focus on methods for creating knowledge that scale with computation, like search and learning.
English
134
964
7.3K
536.1K
OmnAI Lab retweetledi
Giannis Daras
Giannis Daras@giannis_daras·
Join us for a full day of talks related to diffusion and flow based methods in the ReALM-GEN workshop, happening today at ICLR ✨🇧🇷 📍Room 201 A/B
English
1
5
23
1.8K
OmnAI Lab retweetledi
Emanuele Rodolà
Emanuele Rodolà@EmanueleRodola·
presenting this loud one at #ICLR2026 in Rio as well there's gonna be a surprise for pringle lovers. even though that's not actually a pringle. it's an italian crik-crok
GLADIA Research Lab@GladiaLab

LLMs are injective and invertible. In our new paper, we show that different prompts always map to different embeddings, and this property can be used to recover input tokens from individual embeddings in latent space. (1/6)

English
2
9
125
11.8K
OmnAI Lab retweetledi
Iacopo Masi
Iacopo Masi@_iAc·
It just takes a single gradient step on the input using an #EBM loss to boost the #robustness of robust LVLMs like CLIP/LLaVa. We also prove that the accuracy improves if the gradient norm of the gt class is the highest. thanks Odelia!
Hussain Mujtaba@Hussain68018934

Part 1/3 Excited to share that our paper “A Provable Energy-Guided Test-Time Defense: Boosting Adversarial Robustness of Large Vision-Language Models” has been accepted at CVPR 2026 (Main Conference) 🎉

English
1
3
5
459
OmnAI Lab retweetledi
Hussain Mujtaba
Hussain Mujtaba@Hussain68018934·
Part 1/3 Excited to share that our paper “A Provable Energy-Guided Test-Time Defense: Boosting Adversarial Robustness of Large Vision-Language Models” has been accepted at CVPR 2026 (Main Conference) 🎉
Hussain Mujtaba tweet media
English
3
3
4
709
OmnAI Lab retweetledi
Alessandro Salvatore
Alessandro Salvatore@AleSalvatore00·
Why can't we solve adversarial examples? After a decade of work, neural nets still get fooled by imperceptible noise. We think we finally know the geometric reason why — and it connects to AI alignment. 🧵
Alessandro Salvatore tweet media
English
17
91
804
71.2K
OmnAI Lab retweetledi
GLADIA Research Lab
GLADIA Research Lab@GladiaLab·
Language Models are Injective and Hence Invertible (ICLR 2026), aka “pringle paper", is now a public graph on @paradigmainc’s Flywheel In the paper, we show that LMs can be inverted and, contrary to common belief, do not discard information about their inputs at inference time.
GLADIA Research Lab tweet media
GLADIA Research Lab@GladiaLab

LLMs are injective and invertible. In our new paper, we show that different prompts always map to different embeddings, and this property can be used to recover input tokens from individual embeddings in latent space. (1/6)

English
2
18
102
22.1K
OmnAI Lab retweetledi
Paradigma
Paradigma@paradigmainc·
introducing Flywheel: the infrastructure for autonomous research.
English
27
73
551
117.3K
OmnAI Lab retweetledi
Adrian R. Minut
Adrian R. Minut@adrianrminut·
@OmnAI_Lab What we also find very interesting is: - instruct variants show a higher spillage to hallucination correlation; post-training artifacts? - the math behind does not apply to just language modeling, any sequence-to-sequence task abides by the same rules!
English
0
2
3
222
OmnAI Lab
OmnAI Lab@OmnAI_Lab·
6/ We evaluated LLaMA, Mistral, Gemma, and Qwen3 across nine Q/A and reasoning benchmarks, plus a synthetic algebraic stress test. The method effectively localizes the exact answer token and tests for hallucinations.
English
1
0
14
649
OmnAI Lab
OmnAI Lab@OmnAI_Lab·
1/ Large Language Models leak energy when they hallucinate. We built a training-free method to catch the spill and keep them *grounded*. Our #ICLR2026 paper introduces Spilled Energy for SOTA zero-shot detection. TLDR: Hallucinations violate the probability chain rule.
OmnAI Lab tweet mediaOmnAI Lab tweet media
English
7
31
193
11.8K

Keşfet

@eccvconf @Hussain68018934 @THassner @thakralkartik78 @BrigliaRosaria @MayankVatsa3 @dgolano @sijialiu17