isabela_alb's profile picture. Research Scientist @DeepMind. She/her. 🇧🇷

Isabela Albuquerque

@isabela_alb

Research Scientist @DeepMind. She/her. 🇧🇷

Isabela Albuquerque reposted

What do text-to-image models know about numbers? Find out in our new paper "Evaluating Numerical Reasoning in text-to-image Models" to be presented at #NeurIPS2024 (Wed 4:30-7:30 PM, #5304). Dataset: github.com/google-deepmin… (1386 prompts, 52,721 images, 479,570 annotations)


Isabela Albuquerque reposted

Presenting RepLiQA at #NeurIPS2024! 🎯 RepLiQA enables testing LLMs on unseen fictional contexts so there's no memory confounding, & their ability to say "I don't know" when appropriate. 🚨 Fresh data split dropping during the conference! 👀🎉 Paper: nips.cc/virtual/2024/p…

We released the first split of RepLiQA. The data is given by context-question-answer triplets, and contexts are documents concerning made-up things/people/places. As such, RepLiQA can be used for reliably testing the ability of LLMs to seek informantion on provided context.



Isabela Albuquerque reposted

By generating synthetic image samples specific to underrepresented groups, #DiffusionModels help medical #ImageClassifiers to achieve greater fairness metrics across a variety of medical disciplines and demographic attributes. @s0f1ra @isabela_alb @sgowal nature.com/articles/s4159…


Isabela Albuquerque reposted

We released the first split of RepLiQA. The data is given by context-question-answer triplets, and contexts are documents concerning made-up things/people/places. As such, RepLiQA can be used for reliably testing the ability of LLMs to seek informantion on provided context.

📢 New paper out! We release RepLiQA: a Q&A dataset designed to evaluate language models in a truly unseen context. RepLiQA covers made-up entities so that knowledge obtained during training cannot be re-used to answer its questions. #LLM #Evaluation @ServiceNow @ServiceNowRSRCH

ServiceNowRSRCH's tweet image. 📢 New paper out! We release RepLiQA: a Q&A dataset designed to evaluate language models in a truly unseen context. RepLiQA covers made-up entities so that knowledge obtained during training cannot be re-used to answer its questions. #LLM #Evaluation @ServiceNow @ServiceNowRSRCH


Isabela Albuquerque reposted

📢 New paper out! We release RepLiQA: a Q&A dataset designed to evaluate language models in a truly unseen context. RepLiQA covers made-up entities so that knowledge obtained during training cannot be re-used to answer its questions. #LLM #Evaluation @ServiceNow @ServiceNowRSRCH

ServiceNowRSRCH's tweet image. 📢 New paper out! We release RepLiQA: a Q&A dataset designed to evaluate language models in a truly unseen context. RepLiQA covers made-up entities so that knowledge obtained during training cannot be re-used to answer its questions. #LLM #Evaluation @ServiceNow @ServiceNowRSRCH

Isabela Albuquerque reposted

Check out Gecko 🦎: @GoogleDeepMind's latest work looking at how to evaluate text-to-image technology with: 📊 a new benchmark 🕵️ 100K+ human ratings of state-of-the-art T2I models 🤖 a better human-correlated auto-eval metric arxiv.org/abs/2404.16820

ebugliarello's tweet image. Check out Gecko 🦎: @GoogleDeepMind's latest work looking at how to evaluate text-to-image technology with:
📊 a new benchmark
🕵️ 100K+ human ratings of state-of-the-art T2I models
🤖 a better human-correlated auto-eval metric

arxiv.org/abs/2404.16820

Isabela Albuquerque reposted

Excited to demonstrate in our work published today at @NatureMedicine that we can leverage powerful diffusion models to “augment” our medical imaging datasets in a steerable fashion to oversample underrepresented conditions and attributes. @GoogleDeepMind @GoogleResearch

By generating synthetic image samples specific to underrepresented groups, diffusion models help medical image classifiers to achieve greater #fairness metrics across a variety of medical disciplines and demographic attributes. @OliviaW47557022, @s0f1ra nature.com/articles/s4159…

NatureMedicine's tweet image. By generating synthetic image samples specific to underrepresented groups, diffusion models help medical image classifiers to achieve greater #fairness metrics across a variety of medical disciplines and demographic attributes. @OliviaW47557022, @s0f1ra nature.com/articles/s4159…


Isabela Albuquerque reposted

Google presents Revisiting Text-to-Image Evaluation with Gecko On Metrics, Prompts, and Human Ratings While text-to-image (T2I) generative models have become ubiquitous, they do not necessarily generate images that align with a given prompt. While previous work has

_akhaliq's tweet image. Google presents Revisiting Text-to-Image Evaluation with Gecko

On Metrics, Prompts, and Human Ratings

While text-to-image (T2I) generative models have become ubiquitous, they do not necessarily generate images that align with a given prompt. While previous work has

Isabela Albuquerque reposted

XC-Cache Cross-Attending to Cached Context for Efficient LLM Inference In-context learning (ICL) approaches typically leverage prompting to condition decoder-only language model generation on reference information. Just-in-time processing of a context is inefficient

_akhaliq's tweet image. XC-Cache

Cross-Attending to Cached Context for Efficient LLM Inference

In-context learning (ICL) approaches typically leverage prompting to condition decoder-only language model generation on reference information. Just-in-time processing of a context is inefficient

Isabela Albuquerque reposted

📢 Super proud of our new work on adding MoEs, and in particular Soft MoEs, to value-based deep RL agents; results in more parameter-scalable models! Great collaboration with amazing colleagues!! 🥳🎉 🧵Check it out below for how we did it 👇

📢Mixtures of Experts unlock parameter scaling for deep RL! Adding MoEs, and in particular Soft MoEs, to value-based deep RL agents results in more parameter-scalable models. Performance keeps increasing as we increase number of experts (green line below)! 1/9

pcastr's tweet image. 📢Mixtures of Experts unlock parameter scaling for deep RL!

Adding MoEs, and in particular Soft MoEs, to value-based deep RL agents results in more parameter-scalable models.

Performance keeps increasing as we increase number of experts (green line below)!
1/9


Isabela Albuquerque reposted

Excited to share our latest work on chest X-ray report generation in a clinician-AI collaborative setting! Our thorough human evaluation carried out in 🇮🇳 + 🇺🇸 showed >=1 radiologist preferred the AI report in > 80% of inpatient 🇮🇳 and >66% of ICU cases 🇺🇸.


Isabela Albuquerque reposted

Student researcher position applications are open at Google Deepmind! I'm hosting a SR in the intersection of bias and generative models. If you're an interested PhD student please reach out! google.com/about/careers/

google.com

Google Careers

Find your next job at Google — Careers at Google.


Isabela Albuquerque reposted

Excited to attend #NeurIPS2023 next week! We'll present two posters on robustness, both resulting from fruitful collaborations between @ServiceNowRSRCH and @Mila_Quebec through the hard work of the amazing @tsirigoc and @Cguille_escuret who I was so lucky to collaborate with!


Isabela Albuquerque reposted

Eu estou aceitando alunos de mestrado e de doutorado em aprendizado por reforço. Eu adoraria receber algumas aplicações do Brasil. I’m accepting MSc and PhD students to work on reinforcement learning. I would love to receive some applications from Brazil.

It's that time of the year once again! With @cottascience and 20+ other amazing mentors, we will have set up a simple "mentorship" program to help you get your Ph.D. application in shape. brazilians-in-ds-ml.github.io

manoelribeiro's tweet image. It's that time of the year once again!

With @cottascience and 20+ other amazing mentors, we will have set up a simple "mentorship" program to help you get your Ph.D. application in shape.

brazilians-in-ds-ml.github.io


Isabela Albuquerque reposted

It's that time of the year once again! With @cottascience and 20+ other amazing mentors, we will have set up a simple "mentorship" program to help you get your Ph.D. application in shape. brazilians-in-ds-ml.github.io

manoelribeiro's tweet image. It's that time of the year once again!

With @cottascience and 20+ other amazing mentors, we will have set up a simple "mentorship" program to help you get your Ph.D. application in shape.

brazilians-in-ds-ml.github.io

Isabela Albuquerque reposted

📢Are you: Interested/working in AI? Looking to meet other LatAm AI researchers/enthusiasts? Wanting to attend an awesome conference in Quito next February? Living in LatAm and need a travel grant to attend? Then pre-register for @riiaa_org now!! riiaa.org/riiaa6

pcastr's tweet image. 📢Are you:
Interested/working in AI?
Looking to meet other LatAm AI researchers/enthusiasts?
Wanting to attend an awesome conference in Quito next February?
Living in LatAm and need a travel grant to attend?

Then pre-register for @riiaa_org now!!
riiaa.org/riiaa6

Isabela Albuquerque reposted

🌟 Calling all bright minds from South America 🌎✨ With just 3% of applications coming from South America, we’re looking for more Scholars Program applicants in South America! If you’re in search of an opportunity to develop your research skills, your journey starts here.

Cohere_Labs's tweet image. 🌟 Calling all bright minds from South America 🌎✨

With just 3% of applications coming from South America, we’re looking for more Scholars Program applicants in South America! If you’re in search of an opportunity to develop your research skills, your journey starts here.

Isabela Albuquerque reposted

Scientists whose first language is not English spend much longer to read and write papers in English and prepare for international conferences. @tatsuya_amano and colleagues worked to measure this invisible struggle. I wrote about their work for @Nature nature.com/articles/d4158…


Isabela Albuquerque reposted

Thank you for the coffee and the conversation @isabela_alb 🥳 As BR que se conhecem fora do país e continuam se encontrando fora! #NeurIPS22

mirianfrsilva's tweet image. Thank you for the coffee and the conversation @isabela_alb 🥳

As BR que se conhecem fora do país e continuam se encontrando fora! #NeurIPS22

Loading...

Something went wrong.


Something went wrong.