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 11352 publications
Preview abstract
We consider a setting where we have a ground set ℳ together with real-valued set functions f₁, … , f_n, and the goal is to partition ℳ into two sets S₁,S₂ such that |f_i(S₁) - f_i(S₂)| is small for every i. Many results in discrepancy theory can be stated in this form with the functions f_i being additive. In this work, we initiate the study of the unstructured case where f_i is not assumed to be additive. We show that even without the additivity assumption, the upper bound remains at most O(√{n log n}).
Our result has implications on the fair allocation of indivisible goods. In particular, we show that a consensus halving up to O(√{n log n}) goods always exists for n agents with monotone utilities. Previously, only an O(n) bound was known for this setting.
View details
VISTA: A Test-Time Self-Improving Video Generation Agent
Xuan Long Do
Hootan Nakhost
The IEEE/CVF Conference on Computer Vision and Pattern Recognition (to appear) (2026)
Preview abstract
Despite rapid advances in text-to-video (T2V) synthesis, generated video quality remains critically dependent on precise user prompts. Existing test-time optimization methods, successful in other domains, struggle with the multi-faceted nature of video. To address this, we introduce VISTA, a novel multi-agent system that autonomously refines prompts to improve video generation. VISTA operates in an iterative loop, first decomposing a user's idea into a structured temporal plan. After generation, the best video is identified through a robust pairwise tournament. This winning video is then critiqued by a trio of specialized agents focusing on visual, audio, and contextual fidelity. Finally, a reasoning agent synthesizes this feedback to introspectively rewrite and enhance the prompt for the next generation cycle. To rigorously evaluate our proposed approach, we introduce MovieGen-Bench, a new benchmark of diverse single- and multi-scene video generation tasks. Experiments show that while prior methods yield inconsistent gains, VISTA consistently improves video quality, achieving up to 60% pairwise win rate against state-of-the-art baselines. Human evaluators concur, preferring VISTA's outputs in 68% of comparisons.
View details
ToolGrad: Efficient Tool-use Dataset Generation with Textual "Gradients"
Kohei Uehara
Haoyu Zhang
Jingtao Zhou
Lin Gu
Zheng Xu
Tatsuya Harada
ACL 2026 (2026)
Preview abstract
Prior work synthesizes tool-use LLM datasets by first generating a user query, followed by complex tool-use annotations like depth-first search (DFS). This leads to inevitable annotation failures and low efficiency in data generation. We introduce ToolGrad, an agentic framework that inverts this paradigm. ToolGrad first constructs valid tool-use chains through an iterative process guided by textual "gradients", and then synthesizes corresponding user queries. This "answer-first" approach led to ToolGrad-500, a dataset generated with more complex tool use, lower cost, and almost 100% pass rate. Experiments show that ToolGrad models outperform those trained on expensive baseline datasets and proprietary LLMs.
View details
See2Refine: Vision-Language Feedback Improves LLM-Based eHMI Action Designers
Ding Xia
Xinyue Gui
Mark Colley
Fan Gao
Dongyuan Li
Renhe Jiang
Takeo Igarashi
ACL 26 (2026)
Preview abstract
Automated vehicles lack natural communication channels with other road users, making external Human-Machine Interfaces (eHMIs) essential for conveying intent and maintaining trust in shared environments. However, most eHMI studies rely on developer-crafted message-action pairs, which are difficult to adapt to diverse and dynamic traffic contexts. A promising alternative is to use Large Language Models (LLMs) as action designers that generate context-conditioned eHMI actions, yet such designers lack perceptual verification and typically depend on fixed prompts or costly human-annotated feedback for improvement. We present See2Refine, a human-free, closed-loop framework that uses vision-language models (VLMs) for perceptual evaluation as automated visual feedback to improve an LLM-based eHMI action designer. Given a driving context and a candidate eHMI action, the VLM evaluates the perceived appropriateness of the action, and this feedback is used to iteratively revise the designer's outputs, enabling systematic refinement without human supervision. We evaluate our framework across three eHMI modalities (lightbar, eyes, and arm) and multiple LLM model sizes. Across settings, our framework consistently outperforms prompt-only LLM designers and manually specified baselines in both VLM-based metrics and human-subject evaluations. Results further indicate that the improvements generalize across modalities and that VLM evaluations are well aligned with human preferences, supporting the robustness and effectiveness of \systemName for scalable action design.
View details
XProf: An Open, Scalable and Extensible Profiling System for the Modern ML Stack
Naveen Kumar
Jose Baiocchi Paredes
Scott Goodson
Kelvin Le
Yin Zhang
Kan Cai
Jiten Thakkar
Sai Ganesh Bandiatmakuri
Yogesh SY
Ani Udipi
Vikas Aggarwal
Ninth Conference on Machine Learning and Systems (2026)
Preview abstract
Optimizing Large Models across thousands of accelerators requires deep system expertise. To address modern machine learning (ML) optimization needs, we present XProf, the ML profiler for the OpenXLA ecosystem. XProf delivers actionable optimization suggestions and in-depth performance analysis, empowering ML researchers and framework users to improve efficiency without specialized systems knowledge. XProf provides a unified, full-stack view of both host (CPU) and device (accelerator - TPUs/GPUs) performance, leveraging tools like the Roofline Model for comprehensive analysis. XProf’s distributed architecture is designed to monitor thousands of chips with minimal workload overhead (<1%). This architecture is made pluggable through the open-source PJRT C API extension, which has facilitated its adoption by third-party accelerator vendors. XProf has been instrumental in achieving significant efficiency gains at Google and winning MLPerf submissions. This paper presents the design and architecture of XProf, showcases its differentiating tools and capabilities, and highlights its impact within Google and across the industry as a state of the art ML profiler. XProf is available as part of the OpenXLA project at https://github.com/openxla/xprof.
View details
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
Browser fingerprinting is the practice of tracking users across the Web by collecting attributes from their devices and combining them to create unique identifiers. This practice poses major privacy risks to users, and more than a decade of research has quantified fingerprinting risks due to various attributes, leading browser developers to implement many privacy-enhancing changes. Early work used Shannon entropy to quantify risks. However, Shannon entropy can grow with dataset size, limiting the ability to compare datasets and results. Researchers then introduced normalized entropy as a measure for comparing browser fingerprinting datasets of different sizes and numerous works followed using normalized entropy for this purpose.
We identify and address a resulting problem in the fingerprinting literature. We show normalized entropy is ill-suited to compare datasets of different sizes — it decreases as dataset size increases. We show this both analytically and empirically, leveraging a recently published dataset of browser attributes commonly used for fingerprinting.
Given the unmet need for a better fingerprinting risk measure, we define a minimal set of desired properties for such a measure: scale-invariance, monotonicity and estimability. We then propose to use Tsallis entropy as a more interpretable fingerprinting risk measure. We evaluate Shannon, normalized, and Tsallis entropy with respect to the properties, and prove that only Tsallis entropy satisfies all of them.
View details
Preview abstract
Generative AI is reshaping software development, yet its psychological impact remains under-researched. During May and August 2025 we conducted reflexive thematic analysis of interviews with 12 senior engineers (≥5 years experience) recruited from Western technology hubs to explore shifts in professional identity. We identify a central transition from "coder to conductor," where AI acts as a cognitive partner. Key findings include: (1) a re-architecting of focus from implementation to strategy; (2) a shift in productivity metrics from output to impact; and (3) a dual-impact on agency, where AI empowers autonomy but threatens competence through de-skilling anxieties. These findings suggest that as implementation becomes commoditised, organisational training and career progression must prioritise architectural mastery and metacognitive oversight to ensure sustained developer motivation and system integrity.
View details
Diffusion Controller: Framework, Algorithms and Parameterization
Tong Yang
Moonkyung Ryu
Guy Tennenholtz
Yuejie Chi
Proceedings of the 43rd International Conference on Machine Learning (ICML-26), Seoul, South Korea (2026)
Preview abstract
Controllable generation with diffusion models is often treated as a collection of heuristics rather than a unified optimization problem. We propose a principled control formulation by viewing the diffusion reverse process as an instance of a (generalized) linearly-solvable Markov decision process (LS-MDP). This perspective turns controllable generation into regularized optimal control around a pretrained diffusion policy, yielding tractable objectives and algorithmic updates.
Under this framework, we study two practical finetuning regimes. When paired target data are available, we obtain a supervised finetuning (SFT) objective. When only a terminal reward model is available, we derive reinforcement-learning finetuning (RLFT) methods from the LS-MDP solution structure, including (i) a reward-weighted regression loss and (ii) a policy-gradient approach (with standard extensions such as PPO).
Crucially, the LS-MDP optimality conditions imply an explicit relationship between the optimal and pretrained score functions. We leverage this to derive a new score-function parameterization that isolates the control signal and enables “gray-box” finetuning with substantially fewer trainable parameters. Experiments across SFT and RLFT show this parameterization improves over existing finetuning baselines while achieving stronger sample/parameter efficiency.
View details
Preview abstract
Large language models (LLMs) are increasingly deployed as economic agents in marketplaces, auctions, and bidding settings. Anticipating their behavior in any specific deployment is hard. Existing strategic-reasoning benchmarks evaluate models on fixed canonical games. These benchmarks may saturate as the frontier improves, and they do not allow evaluators to generalize with confidence from benchmark performance to the varied and messy strategic environments that actual deployments involve. We introduce GENSTRAT, which uses procedurally generated strategic environments to address these challenges. Concretely, we generate a distribution of two-player zero-sum imperfect-information card games. The generator can draw fresh games on demand, allowing for evergreen evaluation and resistance to contamination. We pair the game distribution with a capability-profile methodology that decomposes model competence across six axes (state space, temporal depth, information sensitivity, opponent modeling, risk, and brittleness). We also introduce a jaggedness measure of within-distribution smoothness that detects when a model’s advantage jumps unpredictably between strategically similar games. We sample 50 benchmark games from a 2,000-game generated pool and evaluate nine frontier and open-weight LLMs in a head-to-head tournament with over 36,000 matches. Newer frontier-tier models score higher on average. Beyond that average, models with near-identical overall strength show qualitatively different capability profiles, and two of the top three leaderboard models (gpt-5 and claude) are noticeably more locally volatile than the third (gemini-3.1-pro), despite being close in overall strength. Together, the capability profile and the jaggedness measure give a deployment-relevant diagnostic that the overall ranking alone cannot provide.
View details
Preview abstract
In today's AI landscape, success depends not just on prompting large language models but on orchestrating them into intelligent systems that are scalable, compliant, and cost-effective. GenAI on Google Cloud is your hands-on guide to bridging that gap. Whether you're an ML engineer or an enterprise leader, this book offers a practical game plan for taking agentic systems from prototype to production.
Written by practitioners with deep experience in AgentOps, data engineering, and GenAI infrastructure, this guide takes you through real-world workflows from data prep and deployment to orchestration and integration. With concrete examples, field-tested frameworks, and honest insights, you'll learn how to build agentic systems that deliver measurable business value.
> Bridge the production gap that stalls 90% of vertical AI initiatives using systematic deployment frameworks
> Navigate AgentOps complexities through practical guidance on orchestration, evaluation, and responsible AI practices
> Build robust multimodal systems for text, images, and video using proven agent architectures
> Optimize for scale with strategies for cost management, performance tuning, and production monitoring
View details
Preview abstract
Here’s a thought experiment. Say I wave a magic wand across a codebase and an entire class of technical debt, poof, goes away and immediately evaporates if introduced in the future. For example, maybe I make it so that dead feature flags are simply no longer a problem: they just delete themselves as soon as the engineer wills it. Or maybe large-scale migrations just migrate themselves. Maybe we magically have 100% test coverage, without an engineer lifting a finger.
What will happen to developer productivity?
Surely, developer productivity increases overall.
But will the productivity metrics that we all use as a proxy for “developer productivity” move up and to the right. Let’s explore this idea.
View details
Preview abstract
Large language models have achieved remarkable capabilities across domains, yet mechanisms underlying sophisticated reasoning continue to be explored1,2. Recent reasoning-reinforced models, including OpenAI’s o-series and DeepSeek-r1, outperform other merely instruction-tuned models on complex cognitive tasks3,4, attributed to extended test-time computation through longer chains of thought5. Here we show that enhanced reasoning emerges not from extended computation alone, but from the systematic simulation of complex, multi-agent interactions—a society of thought—which enables the deliberate diversification and debate among internal cognitive perspectives characterized by distinct personality traits and domain expertise. Through quantitative analysis using classified outputs and mechanistic interpretability methods applied to reasoning traces6–8, we find that reasoning models like DeepSeek-r1 exhibit much greater perspective diversity than baseline models, activating broader and more conflict between heterogeneous personality- and expertise-related features during reasoning. This multi-agent structure manifests in conversational behaviors including question-answering sequences, perspective shifts, and reconciliation of conflicting views, as well as in socio-emotional roles that characterize back-and-forth conversation, which together account for over 60% of the accuracy advantage in reasoning tasks through both direct and indirect facilitation of cognitive strategies9,10. Controlled reinforcement learning experiments further reveal that priming models with conversational scaffolding—even when dialogues lead to incorrect solutions—substantially accelerates reasoning improvement compared to answer-only training. These findings indicate that the social organization of thought, rather than correctness alone, enables effective exploration of solution spaces. We suggest that reasoning models establish a computational parallel to collective intelligence in human groups11–13, where diversity enables superior problem-solving when systematically structured and suggest new opportunities for agent organization to harness the wisdom of crowds.
View details
Preview abstract
In a prior column, we wrote about how measuring productivity can be viewed as a form of modeling and that all models are wrong, but some are useful. That discussion centered on the idea of ensuring that a productivity model was inclusive of multiple metrics and that those metrics covered the various facets of productivity and covered each facet reasonably well. In that article, we set aside the question of what makes a good individual productivity metric that can be combined with others into a (hopefully) useful model of productivity. In this article, we’ll share some things we consider when building an individual metric, including an example of a novel metric we built in the aftermath of the COVID pandemic.
View details
