Publications
Our teams aspire to make discoveries that impact everyone, and core to our approach is sharing our research and tools to fuel progress in the field.
Our teams aspire to make discoveries that impact everyone, and core to our approach is sharing our research and tools to fuel progress in the field.
Sort By
1 - 15 of 10458 publications
Preview abstract
We discuss the challenges posed by growing machine learning workloads on
datacenter networks and present how Google’s Jupiter network fabrics effectively support
diverse traffic.
View details
Steering Self-Evaluation: Interpreting LLM’s Reasoning Across Domains and Languages
Praveen Hegde
2025
Preview abstract
Understanding and controlling the reasoning processes of large language models (LLMs) is crucial for their reliable deployment. In this work, we investigate the latent representation of self-evaluation behavior - the ability of a model to assess its own reasoning steps - a vital behavior for robust reasoning. Through targeted steering vector computation, we identify a direction within LLM activations that represents this self-evaluation behavior. Crucially, we demonstrate that this steering vector for self-evaluation exhibits remarkable cross-contextual efficacy, working well across different domains (e.g., math and medicine) and languages (e.g., English and Spanish). This suggests that the identified latent direction captures a fundamental, abstract representation of self-evaluation within the LLM's internal state, offering a promising avenue for interpretable and controllable reasoning across diverse applications.
View details
Preview abstract
Large language models (LLMs), optimized through human feedback, have rapidly emerged as a leading paradigm for developing intelligent conversational assistants. However, despite their strong performance across many benchmarks, LLM-based agents might still lack conversational skills such as disambiguation -- when they are faced with ambiguity, they often overhedge or implicitly guess users' true intents rather than asking clarification questions. Under task-specific settings, high-quality conversation samples are often limited, constituting a bottleneck for LLMs' ability to learn optimal dialogue action policies. We propose Action-Based Contrastive Self-Training (ACT), a quasi-online preference optimization algorithm based on Direct Preference Optimization (DPO), that enables data-efficient dialogue policy learning in multi-turn conversation modeling. We demonstrate ACT's efficacy under data-efficient tuning scenarios, even when there is no action label available, using multiple real-world conversational tasks: tabular-grounded question-answering, machine reading comprehension, and AmbigSQL, a novel task for disambiguating information-seeking requests for complex SQL generation towards data analysis agents. Additionally, we propose evaluating LLMs' ability to function as conversational agents by examining whether they can implicitly recognize and reason about ambiguity in conversation. ACT demonstrates substantial conversation modeling improvements over standard tuning approaches like supervised fine-tuning and DPO.
View details
Towards Conversational AI for Disease Management
Khaled Saab
David Stutz
Kavita Kulkarni
Sara Mahdavi
Joelle Barral
James Manyika
Ryutaro Tanno
Adam Rodman
arXiv (2025)
Preview abstract
While large language models (LLMs) have shown promise in diagnostic dialogue, their capabilities for effective management reasoning - including disease progression, therapeutic response, and safe medication prescription - remain under-explored. We advance the previously demonstrated diagnostic capabilities of the Articulate Medical Intelligence Explorer (AMIE) through a new LLM-based agentic system optimised for clinical management and dialogue, incorporating reasoning over the evolution of disease and multiple patient visit encounters, response to therapy, and professional competence in medication prescription. To ground its reasoning in authoritative clinical knowledge, AMIE leverages Gemini's long-context capabilities, combining in-context retrieval with structured reasoning to align its output with relevant and up-to-date clinical practice guidelines and drug formularies. In a randomized, blinded virtual Objective Structured Clinical Examination (OSCE) study, AMIE was compared to 21 primary care physicians (PCPs) across 100 multi-visit case scenarios designed to reflect UK NICE Guidance and BMJ Best Practice guidelines. AMIE was non-inferior to PCPs in management reasoning as assessed by specialist physicians and scored better in both preciseness of treatments and investigations, and in its alignment with and grounding of management plans in clinical guidelines. To benchmark medication reasoning, we developed RxQA, a multiple-choice question benchmark derived from two national drug formularies (US, UK) and validated by board-certified pharmacists. While AMIE and PCPs both benefited from the ability to access external drug information, AMIE outperformed PCPs on higher difficulty questions. While further research would be needed before real-world translation, AMIE's strong performance across evaluations marks a significant step towards conversational AI as a tool in disease management.
View details
Preview abstract
Google has a long tradition of open-source software, which encompasses the field of operations research with OR-Tools. In development since 2008, it offers several solvers useful to many OR practitioners:
- PDLP, a revolutionary first-order linear solver that is reshaping the landscape of linear optimisation;
- CP-SAT, an award-winning constraint-programming solver;
- Glop, an accurate linear solver;
- Routing, a vehicle routing solver underpinning Google Maps Platform Route Optimization.
OR-Tools has long had its features accessible from other languages: the core algorithms are implemented in C++ for performance, but users can tap into them in Python, Java, C#, or Go.
It is recently available in Julia too, with a current focus on the linear and constraint solvers, either locally or remotely.
We provide a wrapper for our solvers that brings them to JuMP.jl through MathOptInterface.jl.
This tutorial will walk you through the features of OR-Tools and its solvers, then show examples of using OR-Tools from within Julia, either through JuMP or a lower-level interface.
We will also share our experience of C++-Julia interop.
View details
User-Centered Delivery of AI-Powered Health Care Technologies in Clinical Settings: Mixed Methods Case Study
Randall Brandt
Hien Brown
Christine Silva
JMIR Human Factors (2025)
Preview abstract
Background:
Providers spend a large percentage of their day using electronic health record (EHR) technology and frequently report frustration when EHR tasks are time-consuming and effortful. To solve these challenges, artificial intelligence (AI)–based enhancements to EHR technology are increasingly being deployed. However, AI-based implementations for EHR features often lack user-centered evaluation.
Objective:
This study evaluates, using a user-centered approach, the implementation of an AI-powered search and clinical discovery tool within an EHR system.
Methods:
We conducted a mixed methods study consisting of interviews, observations, and surveys for 5 months.
Results:
High adoption rates for the AI-based features (163/176, 93% users after 3 months) and significant increases across key metrics, including user satisfaction (U=49; P<.001) and perception of time saved (U=49; P<.001), demonstrated that the AI-based features were not only successfully integrated into various clinical workflows but also improved the user experience for clinicians.
Conclusions:
Our results underscore the feasibility and effectiveness of using a user-centered approach for the deployment of clinical AI tools. High adoption rates and positive user experiences were driven by our user-centered research program, which emphasized close collaboration with users, rapid incorporation of feedback, and tailored user training. This study program can be used as a starting framework for the design and integration of human-centered research methods for AI tool deployment in clinical settings.
View details
"It is important to consult" a linguist: Verb-Argument Constructions in ChatGPT and human experts' medical and financial advice
Chris Stewart
Alistair Windsor
J. Elliott Casal
PLOS One (2025)
Preview abstract
This paper adopts a Usage-Based Construction Grammar perspective to compare human- and AI-generated language, focusing on Verb-Argument Constructions (VACs) as a lens for analysis. Specifically, we examine solicited advice texts in two domains—Finance and Medicine—produced by humans and ChatGPT across different GPT models (3.5, 4, and 4o) and interfaces (3.5 Web vs. 3.5 API). Our findings reveal broad consistency in the frequency and distribution of the most common VACs across human- and AI-generated texts, though ChatGPT exhibits a slightly higher reliance on the most frequent constructions. A closer examination of the verbs occupying these constructions uncovers significant differences in the meanings conveyed, with a notable growth away from human-like language production in macro level perspectives (e.g., length) and towards humanlike verb-VAC patterns with newer models. These results underscore the potential of VACs as a powerful tool for analyzing AI-generated language and tracking its evolution over time.
View details
Binamix -- A Python Library for Generating Binaural Audio Datasets
Dan Barry
Davoud Shariat Panah
Alessandro Ragano
Andrew Hines
AES 158th Audio Engineering Society Convention (2025)
Preview abstract
The increasing demand for spatial audio in applications such as virtual reality, immersive media, and spatial audio research necessitates robust solutions to generate binaural audio data sets for use in testing and validation. Binamix is an open-source Python library designed to facilitate programmatic binaural mixing using the extensive SADIE II Database, which provides Head Related Impulse Response (HRIR) and Binaural Room Impulse Response (BRIR) data for 20 subjects. The Binamix library provides a flexible and repeatable framework for creating large-scale spatial audio datasets, making it an invaluable resource for codec evaluation, audio quality metric development, and machine learning model training. A range of pre-built example scripts, utility functions, and visualization plots further streamline the process of custom pipeline creation. This paper presents an overview of the library’s capabilities, including binaural rendering, impulse response interpolation, and multi-track mixing for various speaker layouts. The tools utilize a modified Delaunay triangulation technique to achieve accurate HRIR/BRIR interpolation where desired angles are not present in the data. By supporting a wide range of parameters such as azimuth, elevation, subject Impulse Responses (IRs), speaker layouts, mixing controls, and more, the library enables researchers to create large binaural datasets for any downstream purpose. Binamix empowers researchers and developers to advance spatial audio applications with reproducible methodologies by offering an open-source solution for
binaural rendering and dataset generation. We release the library under the Apache 2.0 License at https://github.com/QxLabIreland/Binamix/
View details
Preview abstract
Self-consistency decoding enhances LLMs’ performance on reasoning tasks by sampling diverse reasoning paths and selecting the most frequent answer. However, it is computationally expensive, as sampling many of these (lengthy) paths is required to increase the chances that the correct answer emerges as the most frequent one. To address this, we introduce Confidence-Informed Self-Consistency (CISC). CISC performs a weighted majority vote based on confidence scores obtained directly from the model. By prioritizing high-confidence paths, it can identify the correct answer with a significantly smaller sample size. When tested on nine models and four datasets, CISC outperforms self-consistency in nearly all configurations, reducing the required number of reasoning paths by over 40% on average. In addition, we introduce the notion of within-question confidence evaluation, after showing that standard evaluation methods are poor predictors of success in distinguishing correct and incorrect answers to the same question. In fact, the most calibrated confidence method proved to be the least effective for CISC. Lastly, beyond these practical implications, our results and analyses show that LLMs can effectively judge the correctness of their own outputs, contributing to the ongoing debate on this topic.
View details
The Cost of Consistency: Submodular Maximization with Constant Recourse
Paul Duetting
Federico Fusco
Ashkan Norouzi Fard
Ola Svensson
Proceedings of the 57th Annual ACM Symposium on Theory of Computing (2025), 1406–1417
Preview abstract
In this work, we study online submodular maximization and how the requirement of maintaining a stable solution impacts the approximation. In particular, we seek bounds on the best-possible
approximation ratio that is attainable when the algorithm is allowed to make, at most, a constant number of updates per step. We show a tight information-theoretic bound of $2/3$ for general monotone submodular functions and an improved (also tight) bound of $3/4$ for coverage functions. Since both these bounds are attained by non poly-time algorithms, we also give a poly-time randomized algorithm that achieves a $0.51$-approximation. Combined with an
information-theoretic hardness of $1/2$ for deterministic algorithms from prior work, our work thus shows a separation between deterministic and randomized algorithms, both information theoretically and for poly-time algorithms.
View details
SSDTrain: Faster Large Language Model Training Using SSD-Based Activation Offloading
Kun Wu
Jeongmin Brian Park
Mert Hidayetoğlu
Vikram Sharma Mailthody
Sitao Huang
Steven Lumetta
Wen-mei Hwu
Design Automation Conference (DAC) (2025)
Preview abstract
The scaling up of Large Language Models (LLMs) demands more memory than current GPUs can provide, hindering the training process. To address this challenge, we propose SSDTrain to efficiently offload activations, the intermediate tensors produced during LLM training, to SSDs. This approach reduces GPU memory usage without impacting performance by adaptively overlapping data transfers with computation. SSDTrain is compatible with popular deep learning frameworks like PyTorch, Megatron, and DeepSpeed, and it employs techniques such as tensor deduplication, forwarding, and adaptive offloading to further enhance efficiency. We conduct extensive experiments on Llama, BERT, and T5. Results demonstrate that SSDTrain effectively reduces 45% of the activation peak memory usage. It can perfectly overlap the IO with the computation without introducing performance penalty. SSDTrain can achieve a performance boost of up to 31% compared to the conventional training strategy using the same GPU systems.
View details
Preview abstract
Creativity in software development is frequently overlooked, specifically
in the design of developer tools which often focus on productivity. This is likely
because creativity is not always seen as a goal in software engineering; in part,
this can be explained by the unique way in which software engineers relate to
creativity as centered around reusability rather than novelty. However, creativity is
a critical aspect of software engineering, and importantly, there is a clear
possibility for AI to impact creativity in both positive or negative ways. In this
article, we explore the differences in goals for designing AI tools for productivity
compared to creativity and propose strategies to elevate creativity in the software
engineering workflow. Specifically, we apply seamful design to AI powered
software development to consider the role of seamfulness in software
development workflows as a way to support creativity.
View details
Preview abstract
This tutorial examines the progress and scaling limitations of IM-DD based optical technologies and explores how datacenter use cases optimized coherent technology, including a newly proposed polarization-folding, time-diversity approach and a novel single-sideband coherent detection technology—can address some of these challenges
View details
Preview abstract
Browser fingerprinting is an online tracking technique that is being increasingly adopted for profiling and ad targeting purposes. While prior work has analyzed the prevalence and impact of browser fingerprinting on the Web, they have traditionally relied on large-scale automated crawls. Naturally, these cannot replicate real-human interactions, e.g., solve CAPTCHAs, evade bot detectors, or operate behind login pages and paywalls. This prompts the question as to whether or not the fingerprinting ecosystem is appreciably different in real-world browsing sessions. In this paper, we begin to address this question by designing and conducting a user study aimed at collecting actual telemetry data from real browsing sessions of 30 users.
We find that almost half of the fingerprinting websites identified from real user browsing sessions are missed by equivalent automated crawls. This is mainly due to the inability of automated crawls to identify and visit authentication pages, being blocked by bot detectors, and/or failing to perform user interactions that specifically trigger browser fingerprinting scripts. We also find new fingerprinting vectors that are consistently present in fingerprinting scripts captured by real user browsing sessions yet missing from automated crawls. Finally, we assess the feasibility of collecting fingerprinting training data in a privacy-preserving way. We conclude that private models built on real user browsing sessions can detect browser fingerprinting more effectively than models trained on automated crawls alone, while simultaneously providing strong privacy guarantees to users.
View details
Circadian rhythm of heart rate and activity: a cross-sectional study
Maryam Khalid
Logan Schneider
Aravind Natarajan
Conor Heneghan
Karla Gleichauf
Chronobiology International (2025)
Preview abstract
ABSTRACT
Background: Circadian rhythms are commonly observed in a number of physiological processes. Consumer wearable devices have made it possible to obtain continuous time series data from a large number of individuals. We study circadian rhythms from measurements of heart rate, movement, and sleep, from a cohort of nearly 20,000 participants over the course of 30 days.
Methods: Participation was restricted to Fitbit users of age 21 years or older residing in the United States or Canada. Participants were enrolled through a recruitment banner shown on the Fitbit App. The advertisement was shown to 531,359 Fitbit users, and 23,239 enrolled in the program. Of these, we obtained heart rate data from 19,350 participants. We obtain the underlying circadian rhythm from time series heart rate by modeling the circadian rhythm as a sum over the first two Fourier harmonics. The first Fourier harmonic accounts for the 24-hour rhythmicity, while the second harmonic accounts for non-sinusoidal perturbations.
Findings: We observe a circadian rhythm in both heart rate and acceleration. From the diurnal modulation, we obtain the following circadian parameters: (i) amplitude of modulation, (ii) bathyphase, (iii) acrophase, (iv) non-sinusoidal fraction, and (v) fraction of day when the heart rate is greater than the mean. The amplitude, bathyphase, and acrophase depend on sex, and decrease with age. The waketime on average, follows the bathyphase by 2.4 hours. In most individuals, the circadian rhythm of heart rate lags the circadian rhythm of activity.
Interpretation: Circadian metrics for heart rate and activity can be reliably obtained from commercially available wearable devices. Distributions of circadian metrics can be valuable tools for individual-level interpretation.
View details
