Shaked Brody

Applied Scientist @AWSCloud

[Amit Bracha @amit_bracha]

🎉 Exciting news! Our paper has been accepted to #ECCV2024 @eccvconf! Check out our project page for more details, including our results on the DTU dataset!

Special thanks to @janusch_patas and @zhenjun_zhao for sharing our work! 🙏

[Noam Rotstein @NoamRot]

Our Paint-by-Inpaint demo is now live! 🥳

Big thanks to @_akhaliq and @Gradio for sharing our paper, and to @huggingface for the GPU grant!

🤗 Check it out here:
Space: https://t.co/7ljEumQrck
Project Page: https://t.co/0ydwgO2BYL

[Noam Rotstein @NoamRot]

🥁 🥁 🥁
Announcing our new work-
Paint by Inpaint: Learning to Add Image Objects by Removing Them First

Together with Navve Wasserman, @roy_ganz, and Ron Kimmel, we've developed a framework designed for adding objects to images!

Paper Page: https://t.co/0ydwgO2BYL

1/7

[Roy Ganz @roy_ganz]

I am thrilled to announce that our work was accepted as SPOTLIGHT to @CVPR!
The official code is available at https://t.co/qdk012iO2h (currently, code and checkpoints for inference. Training will be available soon).
@AmazonScience

[Rishabh (in SF) @rishabh16_]

My two contenders:

1) Why LayerNorm is so helpful in Attention and what it’s actually doing under the hood (https://t.co/UyxO4FnnPJ)

2) Explaining the oversquashing phenomenon in GNNs (https://t.co/dNDaoQAyMt)

Really cool papers with superrr approachable math/theory 🙌🏻

[Noam Rotstein @NoamRot]

Thanks @_akhaliq! I am happy to share that our FuseCap paper has been accepted to @wacv_official!

Project page: https://t.co/9Ai5V8Dg26

[Amir Barkol @BarkolAmir]

I call upon all Harry Potter fans around the world to share this picture in memory of the lovely Noya, a 13-year-old girl on the autistic spectrum who adored Harry Potter and whose tracks disappeared during the murderous terror attack by Hamas. Unfortunately, no whisper helped, and her story did not have a happy ending. Last night, her body was found. Hamas killed her too. Share in memory of Noya! #HamasisISIS

[Shaked Brody @shakedbr]

@yosit קיבלנו את המטענים הניידים. תודה!!

[Aviv Slobodkin @NeurIPS @lovodkin93]

Ever skimmed an article, pinpointing key info, and wished for a tailor-made summary without crafting it yourself?🤔
Introducing SummHelper: your go-to for personalized summarization. 📜✏️

w/ Niv Nachum @pyshmulik @obspp18 Ido Dagan

1/n

[Shaked Brody @shakedbr]

We're presenting our paper today at #RepL4NLP Workshop @aclmeeting , 11:00-12:00, poster session. Come and see me!

[Shaked Brody @shakedbr]

I'm thrilled to announce that our paper "On the Expressivity Role of LayerNorm in Transformers' Attention" has been accpeted to Findings of ACL 2023 #ACL2023.

1/5

[Eran Yahav @yahave]

@shakedbr finally gets to attend a non-virtual conference, presenting https://t.co/2FwkbhZsp5 at #ACL2023
joint work with @urialon1

https://t.co/XyVnxg772O for more details

[AK @_akhaliq]

FuseCap: Leveraging Large Language Models to Fuse Visual Data into Enriched Image Captions

propose FuseCap - a novel method for enriching captions with additional visual information, obtained from vision experts, such as object detectors, attribute recognizers, and Optical Character Recognizers (OCR). Our approach fuses the outputs of such vision experts with the original caption using a large language model (LLM), yielding enriched captions that present a comprehensive image description. We validate the effectiveness of the proposed caption enrichment method through both quantitative and qualitative analysis. Our method is then used to curate the training set of a captioning model based BLIP which surpasses current state-of-the-art approaches in generating accurate and detailed captions while using significantly fewer parameters and training data. As additional contributions, we provide a dataset comprising of 12M image-enriched caption pairs and show that the proposed method largely improves image-text retrieval.

paper page: https://t.co/iCst1Q1dgf
demo: https://t.co/J5MTDUlEFe

[Noam Rotstein @NoamRot]

Introducing FuseCap!

A framework designed to generate semantically rich image captions.

Project page: https://t.co/9Ai5V8Dg26
Paper: https://t.co/9XQ6dyUyfC
Demo: https://t.co/gWN83z4o83

w/ David Bensaid, @shakedbr, @roy_ganz, and Ron Kimmel.

🧵[1/6]

[galvanize (gail weiss) 💔🎗️ @gail_w]

Distractors hate him: this one weird trick* helps language models solve reasoning tasks even in the presence of irrelevant information!

https://t.co/YBsP0E0beH

*memorising knowledge instead of holding it in context

w/ @ZemingChen5, @ericmitchellai, @real_asli, and @ABosselut

[AI Safety Papers @safe_paper]

On the Expressivity Role of LayerNorm in Transformers' Attention
Shaked Brody (@shakedbr), @urialon1, Eran Yahav (@yahave)

Notes: A cool short paper on the role of LayerNorm in transformers. The authors break down this operator into two things: projection and scaling.

[Shaked Brody @shakedbr]

Code: https://t.co/cfIEOT8fh3
Paper: https://t.co/mOpTGg3oAS

5/5

[Shaked Brody @shakedbr]

I'm thrilled to announce that our paper "On the Expressivity Role of LayerNorm in Transformers' Attention" has been accpeted to Findings of ACL 2023 #ACL2023.

1/5

[Shaked Brody @shakedbr]

(a) projection allows the attention to create an attention query that attends to all keys equally, offloading the need to learn this operation by the attention.
(b) scaling prevents keys vector from being "unselectable".

4/5