Securing private data at scale with differentially private partition selection
We present novel algorithms to preserve user privacy in data releases, improving the state of the art in differentially private…
A scalable framework for evaluating health language models
Evaluation of language models in complex domains (such as health) can be expensive and labor intensive. We present a new…
How Google’s AI can help transform health professions education
We explore the utility of Google’s AI models as helpful tools in medical learning environments. By employing a learner-centered and…
Accelerating scientific discovery with AI-powered empirical software
Our new AI system helps scientists write empirical software, achieving expert-level results on six diverse, challenging problems. In scientific research,…
Smarter nucleic acid design with NucleoBench and AdaBeam
We developed an open-source software benchmark for nucleic acid sequence design, and introduced a novel algorithm, AdaBeam, that outperforms existing…
Speculative cascades — A hybrid approach for smarter, faster LLM inference
We introduce “speculative cascades”, a new approach that improves LLM efficiency and computational costs by combining speculative decoding with standard…
VaultGemma: The world’s most capable differentially private LLM
We introduce VaultGemma, the most capable model trained from scratch with differential privacy. As AI becomes more integrated into our…
Learn Your Way: Reimagining textbooks with generative AI
New research into GenAI in education demonstrates a novel approach to reimagining textbooks that led to improved learning outcomes in…
Making LLMs more accurate by using all of their layers
We introduce SLED, a decoding strategy that enhances the accuracy of LLMs by aligning their output with the model’s intrinsic…
Sensible Agent: A framework for unobtrusive interaction with proactive AR agents
Sensible Agent is a research prototype that enables AR agents to proactively adapt what they suggest and how they interact,…
SensorLM: Learning the language of wearable sensors
We present SensorLM, a new family of sensor–language foundation models trained on 60 million hours of data, connecting multimodal wearable…
Simulating large systems with Regression Language Models
We propose text-to-text regression with language models to solve all numeric prediction problems. Large language models (LLMs) often improve by learning…
MLE-STAR: A state-of-the-art machine learning engineering agent
MLE-STAR is a state-of-the-art machine learning engineering agent capable of automating various machine learning tasks across diverse data modalities while…
Highly accurate genome polishing with DeepPolisher: Enhancing the foundation of genomic research
DeepPolisher, is a new deep learning tool that significantly improves the accuracy of genome assemblies by precisely correcting base-level errors,…
Insulin resistance prediction from wearables and routine blood biomarkers
Leveraging wearable data and routine blood tests, we propose a novel method for effectively predicting insulin resistance, providing a scalable…
Achieving 10,000x training data reduction with high-fidelity labels
A new active learning method for curating high-quality data that reduces training data requirements for fine-tuning LLMs by orders of…
Enabling physician-centered oversight for AMIE
We introduce guardrailed-AMIE (g-AMIE), a diagnostic AI designed for history-taking. g-AMIE operates with a guardrail that prohibits it from giving…
Beyond billion-parameter burdens: Unlocking data synthesis with a conditional generator
We present a novel privacy-preserving synthetic data generation algorithm that enables automatic topic-wise distribution matching, making it accessible even for…
Securing private data at scale with differentially private partition selection
We present novel algorithms to preserve user privacy in data releases, improving the state of the art in differentially private…
From massive models to mobile magic: The tech behind YouTube real-time generative AI effects
We detail how YouTube delivers real-time generative AI effects on mobile devices by using knowledge distillation and on-device optimization with…
A scalable framework for evaluating health language models
Evaluation of language models in complex domains (such as health) can be expensive and labor intensive. We present a new…
How Google’s AI can help transform health professions education
We explore the utility of Google’s AI models as helpful tools in medical learning environments. By employing a learner-centered and…
UAE warns Israel’s West Bank annexation would cross ‘red line’ and end regional integration efforts
Israel’s annexation of any part of the occupied West Bank would be a “red line” that would “end the pursuit of regional…
UK blocks Israeli government from major London arms expo
The British government has barred Israeli officials from attending a major arms fair in London next month, citing Israel’s escalation of its…
Time capsule sealed by Princess Diana unearthed at London hospital
Atlanta — A time capsule sealed by Princess Diana in 1991 has been dug out of a children’s hospital in London…
Trump administration plans four-year limit on foreign students studying in the US
A Harvard University graduate with a message in support of international students on her mortarboard sits with fellow students at…
Furniture, architecture, fashion: 3 designers explain how AI is transforming everything
The work of (from left to right) architectural designer Tim Fu, fashion designer Norma Kamali and industrial designer Philippe Starck.…
1 killed, 3 injured after small planes collide midair at Colorado airport
One person was killed and three others injured after two small planes collided in midair by the Fort Morgan Municipal…
Common heart attack drug doesn’t work and may raise risk of death for some women, new studies say
A class of drugs called beta-blockers — used for decades as a first-line treatment after a heart attack— doesn’t benefit…
These are the conditions that make you eligible for an updated Covid-19 vaccine
This year’s updated Covid-19 vaccines have been approved by the US Food and Drug Administration for adults 65 and older…
Judge blocks removal of Guatemalan children in US custody, some of whom were already on planes
A federal judge on Sunday afternoon temporarily blocked the removals of unaccompanied Guatemalan minors in US custody as the government…
Discover powerful AI marketing strategies for
Understanding the AI Marketing Strategy Landscape AI marketing strategies are revolutionizing how businesses approach business growth. By automating complex tasks, personalizing customer experiences, and…
Wagner Mutiny Puts Russia’s Military Bloggers on a Razor’s Edge
Telegram “war correspondents” have promoted the Kremlin’s invasion of Ukraine, but many have also supported mercenaries who launched a failed…
Imagen Editor and EditBench: Advancing and evaluating text-guided image inpainting
Imagen Editor Imagen Editor is a diffusion-based model fine-tuned on Imagen for editing. It targets improved representations of linguistic inputs, fine-grained control and high-fidelity…
Enabling delightful user experiences via predictive models of human attention
Attention-guided image editing Human attention models usually take an image as input (e.g., a natural image or a screenshot of…
Unifying image-caption and image-classification datasets with prefix conditioning
High-level idea We note that classification datasets tend to be biased in at least two ways: (1) the images mostly…
On-device diffusion plugins for conditioned text-to-image generation
Background With diffusion models, image generation is modeled as an iterative denoising process. Starting from a noise image, at each…
Advances in document understanding
Benchmark requirements First, we compared state-of-the-art model accuracy (e.g., with FormNet and LayoutLMv2) on real-world use cases to academic benchmarks (e.g., FUNSD, CORD, SROIE). We observed…
STUDY: Socially aware temporally causal decoder recommender systems
Data Learning Ally has a large digital library of curated audiobooks targeted at students, making it well-suited for building a social…
RO-ViT: Region-aware pre-training for open-vocabulary object detection with vision transformers
Region-aware image-text pre-training Existing VLMs are trained to match an image as a whole to a text description. However, we…
Modeling and improving text stability in live captions
Metric Inspired by previous work, we propose a flicker-based metric to quantify text stability and objectively evaluate the performance of live…
TSMixer: An all-MLP architecture for time series forecasting
TSMixer architecture A key difference between linear models and Transformers is how they capture temporal patterns. On one hand, linear…
MediaPipe FaceStylizer: On-device real-time few-shot face stylization
Few-shot on-device face stylization An end-to-end pipeline Our goal is to build a pipeline to support users to adapt the…
DynIBaR: Space-time view synthesis from videos of dynamic scenes
A mobile phone’s camera is a powerful tool for capturing everyday moments. However, capturing a dynamic scene using a single…
PI-ARS: Accelerating Evolution-Learned Visual-Locomotion with Predictive Information Representations
Typical deep learning models for computer vision, like convolutional neural networks (CNNs) and vision transformers (ViT), process signals assuming planar (flat) spaces. For example,…
Open sourcing Project Guideline: A platform for computer vision accessibility technology
System design The primary use-case is an Android application, however we wanted to be able to run, test, and debug…
Improving simulations of clouds and their effects on climate
Large-eddy simulations on TPUs In this work, we focus on stratocumulus clouds, which cover ~20% of the tropical oceans and…
Unsupervised speech-to-speech translation from monolingual data
Translatotron 3 Translatotron 3 addresses the problem of unsupervised S2ST, which can eliminate the requirement for bilingual speech datasets. To…
Summary report optimization in the Privacy Sandbox Attribution Reporting API
ARA summary reports We use the following example to illustrate our notation. Imagine a fictional gift shop called Du & Penc that…
A new quantum algorithm for classical mechanics with an exponential speedup
Simulating coupled oscillators The systems we consider consist of classical harmonic oscillators. An example of a single harmonic oscillator is…
VALID: A perceptually validated virtual avatar library for inclusion and diversity
Creation and validation of the library Our initial selection of races and ethnicities for the diverse avatar library follows the…
Advancements in machine learning for machine learning
ML compilers ML compilers are software routines that convert user-written programs (here, mathematical instructions provided by libraries such as TensorFlow)…
Simulations illuminate the path to post-event traffic flow
But of course, exiting the arena is only the first step. Next, people must navigate the traffic that builds up…
VideoPoet: A large language model for zero-shot video generation
The diagram below illustrates VideoPoet’s capabilities. Input images can be animated to produce motion, and (optionally cropped or masked) video…
A year of groundbreaking advances in AI and computing
Advances in products & technologies This was the year generative AI captured the world’s attention, creating imagery, music, stories, and…
Cappy: Outperforming and boosting large multi-task language models with a small scorer
We present Cappy, a small pre-trained scorer model that enhances and surpasses the performance of large multi-task language models. We…
MELON: Reconstructing 3D objects from images with unknown poses
We discuss MELON, a technique that can determine object-centric camera poses entirely from scratch while reconstructing the object in 3D.…
Solving the minimum cut problem for undirected graphs
We discuss a recent (best-paper award) publication at ACM-SIAM Symposium on Discrete Algorithms (SODA24) which gives a near-linear running time…
SOAR: New algorithms for even faster vector search with ScaNN
SOAR is an algorithmic improvement to vector search that introduces effective and low-overhead redundancy to ScaNN, Google’s vector search library,…
Preparing and stabilizing quantum states through engineered dissipation
In our latest work, published in Science, we explore a counterintuitive effect of environmental dissipation which is often viewed as…
AutoBNN: Probabilistic time series forecasting with compositional bayesian neural networks
AutoBNN combines the interpretability of traditional probabilistic approaches with the scalability and flexibility of neural networks for building sophisticated time…
Computer-aided diagnosis for lung cancer screening
Introducing a generalizable user-centric interface to help radiologists leverage machine learning models for lung cancer screening. The system takes computed…
Using AI to expand global access to reliable flood forecasts
Large-scale global flood forecasting has been out of reach for a long time. In our Nature paper published today we…
ScreenAI: A visual language model for UI and visually-situated language understanding
We introduce ScreenAI, a vision-language model for user interfaces and infographics that achieves state-of-the-art results on UI and infographics-based tasks.…
SCIN: A new resource for representative dermatology images
The Skin Condition Image Network (SCIN) dataset offers a diverse and representative collection of skin condition images, bridging important gaps…
Safely repairing broken builds with ML
Automatically repairing non-building code increases productivity as measured by overall task completion and appears to introduce no detectable negative impact…
Improving Gboard language models via private federated analytics
To boost Google’s keyboard performance while keeping user data private, we have: worked with language experts to refine its dictionaries,…
Robust speech recognition in AR through infinite virtual rooms with acoustic modeling
Acoustic room simulations allow the training of robust sound separation models for speech recognition on AR Glasses with minimal amounts…
Robots That Write Their Own Code
What if when given instructions from people, robots could autonomously write their own code to interact with the world? It…
Patchscopes: A unifying framework for inspecting hidden representations of language models
Patchscopes is a new framework that aims to unify a variety of previous methods for interpreting the inner workings of…
Solving the minimum cut problem for undirected graphs
We discuss a recent (best-paper award) publication at ACM-SIAM Symposium on Discrete Algorithms (SODA24) which gives a near-linear running time…
Robust speech recognition in AR through infinite virtual rooms with acoustic modeling
Acoustic room simulations allow the training of robust sound separation models for speech recognition on AR Glasses with minimal amounts…
Improving Gboard language models via private federated analytics
To boost Google’s keyboard performance while keeping user data private, we have: worked with language experts to refine its dictionaries,…
Safely repairing broken builds with ML
Automatically repairing non-building code increases productivity as measured by overall task completion and appears to introduce no detectable negative impact…
Scaling hierarchical agglomerative clustering to trillion-edge graphs
We describe a series of our recent works on building more scalable graph clustering, culminating in our “TeraHAC: Hierarchical Agglomerative…
Ten years of neuroscience at Google yields maps of human brain
Marking ten years of connectomics research at Google, we are releasing a publication in Science about a reconstruction at the…
Model Explorer: Graph visualization for large model development
Model Explorer is a powerful graph visualization tool that helps one understand, debug, and optimize ML models. It specializes in…
Advancing medical AI with Med-Gemini
An introduction to Med-Gemini, a family of Gemini models fine-tuned for multimodal medical domain applications. For AI models to perform…
LANISTR: Multimodal learning from structured and unstructured data
LANISTR is a new framework that enables multimodal learning by ingesting unstructured (image, text) and structured (time series, tabular) data,…
Effective large language model adaptation for improved grounding
We introduce AGREE, a learning-based framework that enables LLMs to provide accurate citations in their responses, making them more reliable…
USER-LLM: Efficient LLM contextualization with user embeddings
USER-LLM is a framework that enhances LLMs with a deep understanding of users by distilling diverse user interactions into user…
Beyond billion-parameter burdens: Unlocking data synthesis with a conditional generator
We present a novel privacy-preserving synthetic data generation algorithm that enables automatic topic-wise distribution matching, making it accessible even for…
From diagnosis to treatment: Advancing AMIE for longitudinal disease management
We advance AMIE’s capabilities beyond diagnosis towards treating and managing disease over time. In our randomized study, AMIE matched or…
Validating random circuit sampling as a benchmark for measuring quantum progress
We examine random circuit sampling as a method for evaluating quantum computer performance in the presence of noise, specifically their…
Tx-LLM: Supporting therapeutic development with large language models
Introducing Tx-LLM, a language model fine-tuned to predict properties of biological entities across the therapeutic development pipeline, from early-stage target…
Predicting fetal well-being from cardiotocography signals using AI
We present our work on developing and evaluating a machine learning model for cardiotocography, to predict fetal well-being, and to…
Scaling up linear programming with PDLP
This post describes the award winning product called PDLP, a new first-order method based solver for large-scale linear programming. Classic linear…
Open Buildings 2.5D Temporal dataset tracks building changes across the Global South
Advances in building detection from public satellite imagery lead to a pioneering open dataset of building changes across the Global…
Speculative RAG: Enhancing retrieval augmented generation through drafting
Speculative RAG is a novel Retrieval Augmented Generation framework that uses a smaller specialist LM to generate draft texts that…
Transformers in music recommendation
We present a music recommendation ranking system that uses Transformer models to better understand the sequential nature of user actions…
HALVA: Hallucination Attenuated Language and Vision Assistant
Presenting a new a contrastive tuning strategy to mitigate hallucinations while retaining general performance in multimodal LLMs. Recent advancements in…
Smoothly editing material properties of objects with text-to-image models and synthetic data
We present a method that augments an image generation model with parametric editing of the material properties, such as color,…
A step towards making heart health screening accessible for billions with PPG signals
We describe an approach for using photoplethysmograph (PPG) data for potential use in early detection of cardiovascular disease risk and…
Fast, accurate climate modeling with NeuralGCM
Today we report on NeuralGCM, a model that can rapidly, efficiently, and accurately simulate Earth’s atmosphere. While we know that our…
Harnessing hidden genetic information in clinical data with REGLE
We present a novel method for genetic discovery that can harness hidden information in high-dimensional clinical data. REpresentation learning for…
Accelerating code migrations with AI
Generative AI-powered workflows allow Google to migrate code faster and maintain its codebase more effectively In the past decades, source…
FireBench: Using high-performance computing to advance machine learning and wildfire research
FireBench is a simulation dataset designed to advance wildfire research by simulating controlled fire spread scenarios, which are crucial for…
Assessing ASR performance with meaning preservation
We report progress on using large language models (LLMs) to assess meaning preservation of ASR transcripts, proposing it as an…
Efficient data generation for source-grounded information-seeking dialogs: A use case for meeting transcripts
We release the MISeD (Meeting Information Seeking Dialogs) dataset of information-seeking dialogs focused on meeting transcripts, with corresponding baseline models.…
Generating synthetic data with differentially private LLM inference
We describe an inference-only approach to generating differentially private synthetic data via prompting off-the-shelf large language models with many examples…
Evaluating progress of LLMs on scientific problem-solving
We introduce CURIE, a scientific long-Context Understanding, Reasoning and Information Extraction benchmark to measure the potential of large language models…
Load balancing with random job arrivals
We examine classical scheduling problems, common in computational cluster management, and present improved upper and lower bounds for load balancing…
Google Research at Google I/O 2024
Google I/O exhibits some of Google’s most exciting innovations and cutting-edge technologies. Here we present some of the projects presented…
Rich human feedback for text-to-image generation
We propose rich human feedback for text-to-image (T2I) generation, and show various ways to improve T2I models with our model…
Dynamics of magnetization at infinite temperature in a Heisenberg spin chain
Using a chain of 46 superconducting qubits, we provide evidence against the outstanding conjecture that the 1D Heisenberg spin chain…
Pre-translation vs. direct inference in multilingual LLM applications
A comprehensive evaluation comparing pre-translation with direct inference of PaLM2 on multilingual tasks, demonstrating its improved performance using direct inference…
Human I/O: Detecting situational impairments with large language models
Human I/O is a unified approach that uses egocentric vision, multimodal sensing, and LLM reasoning to detect situational impairments and…
Smart Paste for context-aware adjustments to pasted code
We present Smart Paste, an internal tool that streamlines the code authoring workflow by automating adjustments to pasted code. We…
Advancing personal health and wellness insights with AI
Our research introduces a novel large language model that aims to understand and reason about personal health questions and data.…
AI in software engineering at Google: Progress and the path ahead
Progress of AI-based assistance for software engineering in Google’s internal tooling and our projections for the future. In 2019, a…
Using generative AI to investigate medical imagery models and datasets
We present a framework for understanding AI models in medical imaging, leveraging generative AI and interdisciplinary expert review to identify…
Heuristics on the high seas: Mathematical optimization for cargo ships
We present a previously unknown solution to the Liner Shipping Network Design and Scheduling Problem, which is part of our…
Adversarial Nibbler Challenge: Continuous open red-teaming with diverse communities
The Adversarial Nibbler Challenge is a joint effort between several academic and industrial partners offering a red-teaming methodology for crowdsourcing…
CodecLM: Aligning language models with tailored synthetic data
We propose CodecLM, an end-to-end data synthesis framework that tailors high-quality data to align LLMs for different downstream tasks without…
Few-shot tool-use doesn’t really work (yet)
Instructing language models to use tools based on few demonstrations, while a popular approach, is not as effective as initially…
ChatDirector: Enhancing video conferencing with space-aware scene rendering and speech-driven layout transition
ChatDirector is a research prototype that transforms traditional video conferences into using 3D video avatars, shared 3D scenes, and automatic…
USER-LLM: Efficient LLM contextualization with user embeddings
USER-LLM is a framework that enhances LLMs with a deep understanding of users by distilling diverse user interactions into user…
Effective large language model adaptation for improved grounding
We introduce AGREE, a learning-based framework that enables LLMs to provide accurate citations in their responses, making them more reliable…
Augmented object intelligence with XR-Objects
XR-Objects is an innovative augmented reality research prototype system that transforms physical objects into interactive digital portals using real-time object…
Copyright © All rights reserved Designed and Developed by Digital Flavers
