Publications – Google Research

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.

Preview abstract Semantic data models express high-level business concepts and metrics, capturing the business logic needed to query a database correctly. Most data modeling solutions are built as layers above SQL query engines, with bespoke query languages or APIs. The layered approach means that semantic models can’t be used directly in SQL queries. This paper focuses on an open problem in this space – can we define semantic models in SQL, and make them naturally queryable in SQL? In parallel, graph query is becoming increasingly popular, including in SQL. SQL/PGQ extends SQL with an embedded subset of the GQL graph query language, adding property graph views and making graph traversal queries easy. We explore a surprising connection: semantic data models are graphs, and defining graphs is a data modeling problem. In both domains, users start by defining a graph model, and need query language support to easily traverse edges in the graph, which means doing joins in the underlying data. We propose some useful SQL extensions that make it easier to use higher-level data model abstractions in queries. Users can define a “semantic data graph” view of their data, encapsulating the complex business logic required to query the underlying tables correctly. Then they can query that semantic graph model easily with SQL. Our SQL extensions are useful independently, simplifying many queries – particularly, queries with joins. We make declared foreign key relationships usable for joins at query time – a feature that seems obvious but is notably missing in standard SQL. In combination, these extensions provide a practical approach to extend SQL incrementally, bringing semantic modeling and graph query together with the relational model and SQL. View details

Preview abstract AI coding assistants are rapidly becoming integral to modern software development. A key challenge in this space is the continual need to migrate and modernize codebases in response to evolving software ecosystems. Traditionally, such migrations have relied on rule-based systems and human intervention. With the advent of powerful large language models (LLMs), AI-driven agentic frameworks offer a promising alternative—but their effectiveness remains underexplored. In this paper, we introduce FreshBrew, a novel benchmark for evaluating AI-based agentic frameworks on project-level Java migrations. We benchmark several such frameworks, powered by state-of-the-art LLMs, and compare their performance against established rule-based tools. Our evaluation of AI agents on this benchmark of 228 repositories shows that the top-performing model, Gemini 2.5 Flash, can successfully migrate 56.5% of projects to JDK 17. Our empirical analysis reveals novel insights into the critical strengths and limitations of current agentic approaches, offering actionable insights into their real-world applicability. By releasing FreshBrew publicly upon acceptance, we aim to facilitate rigorous, reproducible evaluation and catalyze progress in AI-driven codebase modernization. View details

Preview abstract How many T gates are needed to approximate an arbitrary n-qubit quantum state to within a given precision ϵ? Improving prior work of Low, Kliuchnikov and Schaeffer, we show that the optimal asymptotic scaling is Θ(sqrt{2^n log(1/ε)} + log(1/ε)) if we allow an unlimited number of ancilla qubits. We also show that this is the optimal T-count for implementing an arbitrary diagonal n-qubit unitary to within error ϵ. We describe an application to batched synthesis of single-qubit unitaries: we can approximate a tensor product of m = O(log log(1/ϵ)) arbitrary single-qubit unitaries to within error ϵ with the same asymptotic T-count as is required to approximate just one single-qubit unitary. View details

Preview abstract For many practical applications of quantum computing, the slowest and most costly steps involve coherently accessing classical data. We help address this challenge by applying mass production techniques, which can sometimes allow us to perform operations many times in parallel for a cost that is comparable to a single execution[1-3]. We combine existing mass-production results with modern approaches for loading classical data using ``quantum read-only memory.'' We show that quantum mass production techniques offer no benefit when we consider a cost model that focuses purely on the number of non-Clifford gates. However, analyzing the constant factors in a more nuanced cost model, we find that it may be possible to obtain a reduction in cost of an order or magnitude or more for a variety reasonably-sized fault-tolerant quantum algorithms. We present several applications of quantum mass-production techniques beyond naive parallelization, including a strategy for reducing the cost of serial calls to the same data loading step. View details

Preview abstract We present CrossCheck, a system that validates inputs to the Software-Defined Networking (SDN) controller in a Wide Area Network (WAN). By detecting incorrect inputs—often stemming from bugs in the SDN control infrastructure—CrossCheck alerts operators before they trigger network outages. Our analysis at a large-scale WAN operator identifies invalid inputs as a leading cause of major outages, and we show how CrossCheck would have prevented those incidents. We deployed CrossCheck as a shadow validation system for four weeks in a production WAN, during which it accurately detected the single incident of invalid inputs that occurred while sustaining a 0% false positive rate under normal operation, hence imposing little additional burden on operators. In addition, we show through simulation that CrossCheck reliably detects a wide range of invalid inputs (e.g., detecting demand perturbations as small as 5% with 100% accuracy) and maintains a near-zero false positive rate for realistic levels of noisy, missing, or buggy telemetry data (e.g., sustaining zero false positives with up to 30% of corrupted telemetry data). View details

Preview abstract Health is a fundamental pillar of human wellness, and the rapid advancements in large language models (LLMs) have driven the development of a new generation of health agents. However, the solution to fulfill diverse needs from individuals in daily non-clinical settings is underexplored. In this work, we aim to build a comprehensive personal health assistant that is able to reason about multimodal data from everyday consumer devices and personal health records. To understand end users’ needs when interacting with such an assistant, we conducted an in-depth analysis of query data from users, alongside qualitative insights from users and experts gathered through a user-centered design process. Based on these findings, we identified three major categories of consumer health needs, each of which is supported by a specialist subagent: (1) a data science agent that analyzes both personal and population-level time-series wearable and health record data to provide numerical health insights, (2) a health domain expert agent that integrates users’ health and contextual data to generate accurate, personalized insights based on medical and contextual user knowledge, and (3) a health coach agent that synthesizes data insights, drives multi-turn user interactions and interactive goal setting, guiding users using a specified psychological strategy and tracking users’ progress. Furthermore, we propose and develop a multi-agent framework, Personal Health Insight Agent Team (PHIAT), that enables dynamic, personalized interactions to address individual health needs. To evaluate these individual agents and the multi-agent system, we develop a set of N benchmark tasks and conduct both automated and human evaluations, involving 100’s of hours of evaluation from health experts, and 100’s of hours of evaluation from end-users. Our work establishes a strong foundation towards the vision of a personal health assistant accessible to everyone in the future and represents the most comprehensive evaluation of a consumer AI health agent to date. View details

Preview abstract As public policy advances the rights of people with disabilities, and as corporations begin to recognize disabled people as core market segments, industry practice in accessible technology design and development improves day by day. Yet, as with other research domains there are lags and gaps in the implementation of accessibility between the lab and industry practice. In this panel, we describe multiple barriers to collaboration between academia and industry and how these barriers manifest as issues in implementation of research findings in industry and lack of adoption of best practices in academia, and vice versa. We then discuss how these specific cases complicate the imagined divide between academic and industry approaches to accessibility. How can notions of accessibility be expanded in both contexts to include overlooked dimensions like ethics, dark patterns, and cognition? This discussion moves towards more inclusive, impactful, and actionable accessibility practices across industry, academia, and public policy. View details

Preview abstract We study a natural complexity measure of Boolean functions known as the (exact) rational degree. For total functions f, it is conjectured that rdeg (f) is polynomially related to deg (f), where deg (f) is the Fourier degree. Towards this conjecture, we show that symmetric functions have rational degree at least deg (f)/2 and monotone functions have rational degree at least deg(f). We observe that both of these lower bounds are tight. In addition, we show that all read-once depth-d Boolean formulae have rational degree at least Ω( deg (f )^1/d). Furthermore, we show that almost every Boolean function on n variables has rational degree at least n/2 − O(√n). In contrast to total functions, we exhibit partial functions that witness unbounded separations between rational and approximate degree, in both directions. As a consequence, we show that for quantum computers, post-selection and bounded-error are incomparable resources in the black-box model. View details

Preview abstract We introduce the Balls-and-Bins sampling for differentially private (DP) optimization methods such as DP-SGD. While it has been common practice to use some form of shuffling in DP-SGD implementations, privacy accounting algorithms have typically assumed that Poisson subsampling is used instead. Recent work by Chua et al. (2024) however pointed out that shuffling based DP-SGD can have a much larger privacy cost in practical regime of parameters. We show that the Balls-and-Bins sampling achieves the “best-of-both” samplers, namely, the implementation of Balls-and-Bins sampling is similar to that of Shuffling and models trained with Balls-and-Bins based DP-SGD achieve utility comparable to those trained with Shuffle based DP-SGD at the same noise multiplier, and yet, Balls-and-Bins sampling enjoys similar-or-better privacy amplification as compared to Poisson subsampling. View details

Preview abstract Subject-driven text-to-image (T2I) generation aims to produce images that align with a given textual description, while preserving the visual identity from a referenced subject image. Despite its broad downstream applicability - ranging from enhanced personalization in image generation to consistent character representation in video rendering - progress in this field is limited by the lack of reliable automatic evaluation. Existing methods either assess only one aspect of the task (i.e., textual alignment or subject preservation), misalign with human judgments, or rely on costly API-based evaluation. To address this gap, we introduce RefVNLI, a cost-effective metric that evaluates both textual alignment and subject preservation in a single run. Trained on a large-scale dataset derived from video-reasoning benchmarks and image perturbations, RefVNLI outperforms or statistically matches existing baselines across multiple benchmarks and subject categories (e.g., Animal, Object), achieving up to 6.4-point gains in textual alignment and 5.9-point gains in subject preservation. View details

Preview abstract Machine learning (ML) fairness evaluation in real-world, industry settings presents unique challenges due to business-driven constraints that influence decision-making processes. While prior research has proposed fairness frameworks and evaluation methodologies, these approaches often focus on idealized conditions and may lack consideration for the practical realities faced by industry practitioners. To understand these practical realities, we conducted a semi-structured interview study with 21 experts from academia and industry specializing in ML fairness. Through this study, we explore three constraints of ML fairness evaluation in industry— balancing competing interests, lacking power/access, and getting buy-in—and how these constraints lead to satisficing, seeking satisfactory rather than ideal outcomes. We define the path from these constraints to satisficing as pragmatic fairness. Using recommender systems as a case study, we explore how practitioners navigate these constraints and highlight actionable strategies to improve fairness evaluations within these business-minded boundaries. This paper provides practical insights to guide fairness evaluations in industry while also showcasing how the FAccT community can better align research goals with the operational realities of practitioners. View details

Preview abstract A growing body of research has demonstrated that the behavior of large language models can be effectively controlled at inference time by directly modifying their internal states, either through vector additions to their activations or through updates to their weight matrices. These techniques, while powerful, are often guided by empirical heuristics, such as deriving steering vectors from the average activations of contrastive prompts. This work provides a theoretical foundation for these interventions, explaining how they emerge from the fundamental computations of the transformer architecture. Building on the recent finding that a prompt's influence can be mathematically mapped to implicit weight updates (Dherin et al., 2025), we generalize this theory to deep, multi-block transformers. We show how the information contained in any chunk of a user prompt is represented and composed internally through weight vectors and weight matrices. We then derive a principled method for condensing this information into token-independent thought vectors and thought matrices. These constructs provide a theoretical explanation for existing vector- and matrix-based model editing techniques and offer a direct, computationally-grounded method for transmuting textual input into reusable weight updates. View details

Preview abstract Cost-sensitive loss functions are crucial in many real-world prediction problems, where different types of errors are penalized differently; for example, in medical diagnosis, a false negative prediction can lead to severe consequences. However, traditional PAC learning theory has mostly focused on the symmetric 0-1 loss, leaving cost-sensitive losses largely unaddressed. In this work, we extend the celebrated theory of boosting to incorporate both cost-sensitive and multi-objective losses. Cost-sensitive losses assign costs to the entries of a confusion matrix, and are used to control the sum of prediction errors accounting for the cost of each error type. Multi-objective losses, on the other hand, simultaneously track multiple cost-sensitive losses, and are useful when the goal is to satisfy several criteria at once (e.g., minimizing false positives while keeping false negatives below a critical threshold). We develop a comprehensive theory of cost-sensitive and multi-objective boosting, providing a taxonomy of weak learning guarantees that distinguishes which guarantees are trivial (i.e., they can always be achieved), which are boostable (i.e., they imply strong learning), and which are intermediate, implying non-trivial yet not arbitrarily accurate learning. For binary classification, we establish a dichotomy: a weak learning guarantee is either trivial or boostable. In the multiclass setting, we describe a more intricate landscape of intermediate weak learning guarantees. Our characterization relies on a geometric interpretation of boosting, revealing a useful duality between cost-sensitive and multi-objective losses. View details

Preview abstract Phone automation agents aim to autonomously perform a given natural-language user request, such as scheduling appointments or booking a hotel. While much research effort has been devoted to screen understanding and action planning, complex tasks often necessitate user interaction for successful completion. Aligning the agent with the user's expectations is crucial for building trust and enabling personalized experiences. This requires the agent to proactively engage the user when necessary, avoiding actions that violate their preferences while refraining from unnecessary questions where a default action is expected. We argue that such subtle agent-initiated interaction with the user deserves focused research attention. To promote such research, this paper introduces a task formulation for detecting the need for user interaction and generating appropriate messages. We thoroughly define the task, including aspects like interaction timing and the scope of the agent's autonomy. Using this definition, we derived annotation guidelines and created a diverse dataset for the task, leveraging an existing UI automation dataset. We tested several text-based and multimodal baseline models for the task, finding that it is very challenging for current LLMs. We suggest that our task formulation, dataset, baseline models and analysis will be valuable for future UI automation research, specifically in addressing this crucial yet often overlooked aspect of agent-initiated interaction. This work provides a needed foundation to allow personalized agents to properly engage the user when needed, within the context of phone UI automation. View details

Preview abstract We study the problem of learning the optimal item pricing for a unit-demand buyer with independent item values, and we have query access to the underlying value distributions. We consider two common query model in the literature: the “sample complexity” model where we can obtain a sample of each item value, and the “pricing query complexity” model where we can set a price to each item and obtain a binary signal on whether the sampled value of the item is greater than our proposed price. In this work we give nearly tight sample complexity and pricing query complexity of the problem. View details