@ItamarZimerman Profile picture

Itamar Zimerman

@ItamarZimerman

PhD candidate @ Tel Aviv University. AI Research scientist @ IBM Research. Interested in deep learning and algorithms.

Similar User
Ofir Lindenbaum photo

@Ofirlin

Yoad Tewel photo

@YoadTewel

Shelly Sheynin photo

@ShellySheynin

Tal Shaharabany photo

@TShrbny

Sagie Benaim photo

@BenaimSagie

Itai Lang photo

@ItaiLang

Ohad Rubin photo

@OhadRubin

Nir Zabari‎ photo

@nirzabari

Idan Schwartz photo

@idansc

Shir Gur 🎗️ photo

@shir_gur

Dr. Haim Ben Yakov photo

@haim_ben_yakov

Mechanical Engineering Department,Thapar Institute photo

@HmedThapar

Pinned

New!🚨📰 Mamba is a cool, efficient, and effective DL architecture, but what do we know about Mamba? How does it capture interactions between tokens? Can it be the attention-killer? In our work, "The Hidden Attention of Mamba Models" we provide answers to these questions! [1/4]

Tweet Image 1

Itamar Zimerman Reposted

Excited to announce LTX-Video! Our new text-to-video model generates stunning, high-quality videos faster than real-time—5 seconds of 24fps video at 768x512 in just 4 seconds on an Nvidia H100! ⚡ We’re open-sourcing the code & weights. Check out the results 🎥👇


Itamar Zimerman Reposted

🚀 Excited to release the code and demo for ConsiStory, our #SIGGRAPH2024 paper! No fine-tuning needed — just fast, subject-consistent image generation! Check it out here 👇 Code: github.com/NVlabs/consist… Demo: build.nvidia.com/nvidia/consist…

Nvidia presents ConsiStory Training-Free Consistent Text-to-Image Generation paper page: huggingface.co/papers/2402.03… enable Stable Diffusion XL (SDXL) to generate consistent subjects across a series of images, without additional training.



Itamar Zimerman Reposted

🎙️ Proud to share our new preprint: Continuous Speech Synthesis using per-token Latent Diffusion Check it out: huggingface.co/papers/2410.16… @ArnonTu, @NimrodShabtay #IBMResearch #SpeechSynthesis

Tweet Image 1

Itamar Zimerman Reposted

Introducing LiveXiv, a new, challenging and maintainable scientific multi-modal live dataset Paper: arxiv.org/abs/2410.10783 Github: github.com/NimrodShabtay/… Dataset: huggingface.co/datasets/LiveX…


Itamar Zimerman Reposted

I'm going to present ConsiStory📖 at #SIGGRAPH2024 this Monday @ 2pm! If you're around this week, DM me if you want to chat! Details below⬇️🧵

Nvidia presents ConsiStory Training-Free Consistent Text-to-Image Generation paper page: huggingface.co/papers/2402.03… enable Stable Diffusion XL (SDXL) to generate consistent subjects across a series of images, without additional training.



Given a polynomial P and an input x, the magic of homomorphic encryption enables the computation of Dec(P(Enc(x)), yielding P(x) without exposing information about x or P. By converting transformers into polynomials, we provide a new framework for confidential computing with LMs!

Tweet Image 1

1/5 Worried about your sensitive data while using LLMs? Come and join me tomorrow at @icmlconf at 11:30, presenting our work: "Converting Transformers to Polynomial Form for Secure Inference over Homomorphic Encryption" (#2115). #IBMResearch #HELayers >>

Tweet Image 1


Itamar Zimerman Reposted

Want to train short and evaluate long? 📚🤖 DeciMamba code is out! Github 👉 github.com/assafbk/DeciMa… Paper 👉 arxiv.org/abs/2406.14528

New Work! 🐍 What prevents Mamba from extrapolating to sequences that are significantly longer than those it was trained on? Furthermore, can Mamba solve long-range NLP tasks using short-range training only? 🧵🧵🧵

Tweet Image 1


NEW! 📰📢 What are Mamba's length generalization capabilities? What limits them? How can we unlock their potential for real-world long NLP tasks? In a very fun work led by Assaf @abk_tau, we dive deep into these questions! arxiv.org/abs/2406.14528

New Work! 🐍 What prevents Mamba from extrapolating to sequences that are significantly longer than those it was trained on? Furthermore, can Mamba solve long-range NLP tasks using short-range training only? 🧵🧵🧵

Tweet Image 1


Itamar Zimerman Reposted

Why do state-space models work so well? With @orvieto_antonio, we study their learning dynamics and find that diagonal recurrence is key: 1. It helps in better conditioning the loss landscape 2. It facilitates Adam's job by making the Hessian diagonal 📝 arxiv.org/abs/2405.21064

Tweet Image 1

Itamar Zimerman Reposted

🧵 Exciting news from our latest research on Interpretability of LMs and fairness! Our paper "Natural Language Counterfactuals through Representation Surgery" introduces a novel approach to interpret representational level interventions and mitigate biases in language models.👇


Itamar Zimerman Reposted

Introducing Menteebot: Groundbreaking Humanoid Robot. We're proud to unveil Menteebot, the culmination of a two-year journey by our brilliant team! Menteebot is a groundbreaking humanoid robot designed for versatility. Visit our website menteebot.com for more demos.


Loading...

Something went wrong.


Something went wrong.