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.

people standing in front of a screen with images and a chipboard

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
  • Title
  • Title, descending
  • Year
  • Year, descending
1 - 15 of 11927 publications
Preview abstract Large Language Models (LLMs) are increasingly deployed to automate enterprise recruitment pipelines, yet evaluating their decision consistency and underlying logic remains heavily throttled by a lack of credible, real-world benchmark datasets. Free-form floating-point scoring methods frequently used in recruitment AI suffer from severe numerical calibration drift and mask the models' intermediate reasoning. To address these challenges, we introduce Djinni-CB, a novel, two-tiered evaluation benchmark containing 2,484 real candidate-job pairs spanning 21 diverse job categories, bridged entirely from naturally occurring, anonymized candidate résumés and standardized employer postings. Rather than relying on generic scores, we implement a rigorous 29-question structured questionnaire across 9 diagnostic groups to audit pure deterministic reasoning boundaries. We establish comprehensive baseline results at temperature T=0.0 comparing frontier model architectures: Gemini 4.1m, DeepSeek Pro, GPT-5.4, and Gemini 2.5 Pro. Over 235,000 discrete decision points reveal a distinct dichotomy in the LLM ecosystem: a "Literal Gatekeeper" cluster (DeepSeek Pro and GPT-5.4) that prioritizes strict keyword matching, and a "Holistic" cluster (the Gemini family) that leverages contextual domain inference. Our dataset provides a highly structured resource complete with verbatim natural-language reasoning logs, offering an essential framework for future research into interpretability, factuality, and consistency in automated hiring. View details
Preview abstract Modern transformers have been scaled up in size, data, training-time compute, and more recently test-time compute based on Chain-of-Thought (CoT), but the consequences of trading off these parameters are unknown. We construct explicit tradeoffs between two of these aspects: the number of layers in a transformer and the number of CoT tokens. Our first result is general: with mild assumptions and a single-token prompt, for any integer k, a transformer using soft attention and SwiGLU MLP layers with L layers and n CoT tokens can be converted to a transformer with at most ckL layers and ⌈n/k⌉ CoT tokens, and vice versa, although with an increase to the size of other parameters. We then a multi-token prompt path following task, finding that additional preprocessing from the multi-token prompt changes the depth-CoT relationship in this case. We construct an attention-only network with L layers and 2^(L−2) heads that can traverse 2^(L−1) − 1 edges along a path with a single CoT token, demonstrating at least an exponential relationship between increases in depth and reductions in the number of CoT tokens needed to traverse the entire path. This relationship for the path following task holds empirically as well, as demonstrated by training networks with up to 5 layers, with the tradeoff between depth and the maximum number of edge traversals in a single CoT token exactly matching our theoretical constructions. View details
Preview abstract Enterprise digital transformation projects are among the most complex and costly initiatives in modern organizations, yet conventional governance frameworks rely predominantly on reactive, lagging indicators of project health. This paper presents an empirical binary classification study applying machine learning to identify high-risk implementations in enterprise digital transformation portfolios. A publicly available dataset of 4,000 project records encoding 49 governance-relevant features is used; an original four-level risk taxonomy is collapsed into two governance-actionable classes: High-Risk and Not High-Risk Implementation. This study proposes a governance-actionable machine learning framework that converts multi-level project risk into binary escalation decisions and combines cross-validated model selection with direct SHAP attribution for enterprise transformation portfolios. Seven classifiers are benchmarked against a majority-class baseline across six metrics. Model selection incorporates five-fold stratified cross-validation and grid-search hyperparameter optimization for the two strongest classifiers. Five-fold stratified cross-validation confirms generalization stability across all models: AdaBoost achieves the highest cross-validation performance—85.85% ± 1.46% accuracy, 85.56% ± 1.32% balanced accuracy, 84.02% ± 1.36% F1-score, and 0.9380 ± 0.0070 ROC-AUC—and also leads on the held-out test set (90.87% accuracy, 90.69% balanced accuracy, 89.76% F1-score, 0.9766 ROC-AUC). All seven classifiers exhibit standard deviations below the 3.0 percentage-point instability threshold, confirming robust generalization without overfitting. SHAP-based explainability, applied directly to the AdaBoost model via TreeExplainer, identifies Complexity_Score, Org_Process_Maturity, and Technology_Familiarity as the primary drivers of high-risk classification, consistent with digital transformation governance literature. Practical integration scenarios for portfolio dashboards and governance escalation workflows are discussed. View details
Who’s Doing the Thinking Now? Reflexivity and Oversight in GenAI-Mediated Decision-Making
Oliver Herrbach
Siddharth Gaurav Majhi
Under review with Information Systems Journal (2027) (to appear)
Preview abstract Generative AI (GenAI) is increasingly embedded in organizational decision-making, reshaping how judgment, oversight, and productivity are enacted in everyday work. While prior research documents performance gains from AI assistance, less is known about how sustained GenAI use affects human reflexivity, i.e., the capacity for critical evaluation, deliberation, and decision override in AI-mediated tasks. This study advances Algorithmic Reflexivity Reallocation Theory (ALERT), which theorizes how repeated algorithmic mediation reallocates cognitive effort away from human judgment through cognitive offloading and automation bias, producing a trade-off between productivity and reflexive oversight. We test ALERT using a three-stage mixed-methods design comprising qualitative grounding interviews, a controlled behavioral experiment, and a large-scale survey-based structural equation model. Behavioral results show that GenAI assistance increases task efficiency but reduces override behavior, particularly when recommendations are plausible yet incorrect. Structural modeling further demonstrates that human–AI trust, task familiarity, and algorithmic structuration predict cognitive offloading, which in turn suppresses override behavior and weakens individual reflexivity while increasing perceived productivity. Transparency emerges as a key design condition that mitigates cognitive offloading and preserves evaluative engagement. By theorizing reflexivity as a reallocatable cognitive and organizational resource, ALERT extends structuration and human–AI interaction research by explaining how productivity gains can coexist with the erosion of reflective judgment under sustained GenAI use. The study contributes a process theory of AI-mediated decision-making with implications for the design and governance of AI-supported work systems. View details
The Decision-Accountability Gap: Symbolic Oversight Theory and Pivot to Human-in-the-Accountability-Loop (HIAL)
Sreekumar Hariharan
Amrita George
Shaliet Rose Sebastian
Sent for review (2027) (to appear)
Preview abstract As organizations increasingly deploy agentic artificial intelligence (AI) systems, Human-in-the-Loop (HITL) oversight structures are widely mandated to ensure compliance, ethics, and control. However, the substantive efficacy of these structures remains poorly understood under conditions of AI scale and high operational complexity. This study introduces Symbolic Oversight Theory (SOT), a novel AI governance framework rooted in neo-institutional theory and bounded rationality. We theorize that when AI systems exhibit high epistemic authority under intense operational velocity and regulatory pressures, traditional decision-centric HITL mechanisms decouple from operational realities. This creates a state of "symbolic oversight"—a ritualistic formality where formal review checkpoints persist for institutional legitimacy, but substantive human cognition, reflexivity, and intervention capacity erode, structurally shifting accountability loads onto individuals without true decisional agency. To validate and apply SOT, we employ a rigorous mixed-method research design across three studies: (1) a controlled behavioral experiment tracking heuristic processing and error-detection among finance professionals, (2) a field survey validating the structural relations and psychometric properties of our constructs, and (3) an empirical study integrating anonymized interaction telemetry logs from internal enterprise agentic AI systems with semi-structured, anonymized internal employee interviews. Our findings demonstrate an empirical divergence between nominal and enacted oversight, showing that symbolic HITL suppresses human reflexivity through cognitive offloading and proceduralization, ultimately yielding a systemic decision-accountability gap. To resolve this governance design failure, we present a design-science architecture shifting governance from decision-centric loops to a consequence-centric Human-in-the-Accountability-Loop (HIAL) framework, establishing new metrics for consequence traceability, accountability latency, and auditability-controllability symmetry. View details
A Data-Driven Biophotonic Skin Model for Wearable Sensing
Kaiyuan (Kai) Yao
Hamed Vavadi
Biomedical Optics Express (2027) (to appear)
Preview abstract Accurate modeling of light-matter interactions in biological tissue is fundamental to the advancement of biophotonic sensing, in-vivo diagnostics, and the development of next-generation wearable health monitors. While computational methods such as Monte Carlo simulations are powerful tools for virtual prototyping, the reliability of their predictions is strictly limited by the fidelity of the underlying skin model. Current industry-standard models primarily rely on optical parameters derived from ex-vivo tissue samples or dataset with limited size and variabilities in measurement methods, often failing to capture the complex spectral responses and inter-subject variability found in living populations. To address this limitation, we present a data-driven biophotonic skin model rigorously calibrated and validated against extensive in-vivo dataset. Our study utilizes smartwatch data acquired across a broad spectral range (UV to SWIR) and multiple sensor geometries, with subjects covering a diverse range of skin tones. By grounding the standard multilayer skin architecture in experimental data, this work provides a robust, statistically validated foundation for the design, simulation, and optimization of future wearable optical sensors. View details
A 3D Scene Graphs Survey: Open Challenges and Future Directions
Dennis Rotondi
Francesco Argenziano
Sebastian Koch
Nathan Hughes
Martin Büchner
Johanna Wald
Lukas Schmid
Daniele Nardi
Abhinav Valada
Liam Paul
Luca Carlone
Kai Arras
Annual Review of Control, Robotics, and Autonomous Systems (ARCRAS), 10 (2027) (to appear)
Preview abstract 3D Scene Graphs (3DSGs) have emerged as a powerful representation for spatial AI by combining geometric grounding with semantic and relational abstractions of the environment. Their expressiveness has made them relevant to a broad range of problems in robotics and computer vision, including mapping, task and motion planning, scene understanding, and many others. However, the field remains fragmented: different communities adopt distinct formulations, construction pipelines, and evaluation protocols, making it difficult to compare methods, identify common assumptions, and assess remaining challenges for robust real- world deployment. This survey provides a unified and critical review of 3DSGs, with particular emphasis on open challenges and future directions. We first formalize 3DSGs under a common definition and analyze the principal modeling choices that characterize existing formulations, including node and edge attributes, hierarchical structure, dynamic scene representations, and affordance-aware extensions. We then review how 3DSGs are constructed from raw sensory observations, covering both learning-oriented and construction-oriented systems. Finally, we examine downstream applications and evaluation strategies, from intrinsic graph quality to task-level performance. To support the community, we also provide a dedicated website that organizes and extends the surveyed works. View details
Preview abstract Recent reports have highlighted how mobile apps share user location data with third parties, risking user privacy and platform trust. Although location data is highly sensitive, when users grant apps location access, they may not know the full extent to which it is used. We study how requiring Android apps to show a reason for location access could impact developers, users, and the platform. We surveyed 323 Android app developers and found most supported such a requirement. The majority said it would have a positive impact on user privacy, trust for apps, and trust for Android, where impact on user trust for Android correlated most strongly with support. Many developers also said the intervention would increase the number of users granting location access. Yet their open-ended comments also revealed consistent concerns, such as apps providing dishonest reasons and platform verification. To study the impact on user behavior, we conducted a randomized controlled experiment with 2579 US Android users. We tested how users' decisions to grant location access were impacted by app type, whether reasons were included in the requests, and the content of the reasons, including monetization. We did not find the reasons impacted users' decisions; decisions were instead driven by app type and demographics. Yet we did find the reasons could have a positive impact on user perception for the platform when the reasons did not include using data for ads. Our findings provide insights into developers' willingness to implement privacy-enhancing changes, and expose limits to improving user privacy by simply adding information to user interfaces. View details
Preview abstract Traditional change detection identifies where changes occur, but does not explain what changed in natural language. Existing remote sensing change captioning datasets typically describe overall image-level differences, leaving fine-grained localized semantic reasoning largely unexplored. To close this gap, we present RSRCC, a new benchmark for remote sensing change question-answering containing 126k questions, split into 87k training, 17.1k validation, and 22k test instances. Unlike prior datasets, RSRCC is built around localized, change-specific questions that require reasoning about a particular semantic change. To the best of our knowledge, this is the first remote sensing change question-answering benchmark designed explicitly for such fine-grained reasoning-based supervision. To construct RSRCC, we introduce a hierarchical semi-supervised curation pipeline that uses Best-of-N ranking as a critical final ambiguity-resolution stage. First, candidate change regions are extracted from semantic segmentation masks, then initially screened using an image-text embedding model, and finally validated through retrieval-augmented vision-language curation with Best-of-N ranking. This process enables scalable filtering of noisy and ambiguous candidates while preserving semantically meaningful changes. The dataset is available at \url{https://huggingface.co/datasets/google/RSRCC}. View details
DeduBB: Binary Code Size Reduction via Post-Link Basic Block De-duplication
Chaitanya Mamatha Ananda
Rajiv Gupta
Mahbod Afarin
Han Shen
LCTES (Languages, Compilers, Tools and Theory of Embedded Systems) (2026) (to appear)
Preview abstract Binary sizes of newer versions of software applications tend to be larger, primarily due to feature bloat. This poses various challenges, particularly for mobile applications. It affects upgrade rates directly impacting revenues, increases maintenance costs of supporting multiple versions, and prevents some users from getting critical security fixes. Code bloat also poses a problem for large warehouse-scale applications. Such applications experience performance degradation when their code size exceeds what smaller and more efficient code models can handle. In this paper, we introduce a post-link optimization tech nique called DeduBB, which deduplicates basic blocks of an application across procedure boundaries. While prior tech- niques used function outlining to de-duplicate redundant code sequences, it missed out on many opportunities as it cannot handle code that manipulates the program stack. In addition, previous techniques were either limited to the scope of a module or lacked scalable implementations required to handle large warehouse-scale applications. Our technique, DeduBB, handles all types of code duplication as we use a novel save-and-jump code pattern to execute de-duplicated code blocks. In addition, DeduBB has been designed to work on scalable post-link optimizers and can even be applied to large warehouse-scale datacenter applications. Finally, DeduBB is profile-guided and can be applied selectively to infrequently executed cold basic blocks to not affect application performance. In fact, in several cases, the performance of the smaller application binary improves due to reductions in its hot working set size. We have implemented our technique on the state-of-the-art post link optimizers, BOLT and Propeller. Experiments show that we can significantly reduce the code size of several benchmarks by 1.55% to 18.63%, on both Arm and x86 platforms, and on binaries that have already been heavily optimized for size using existing code size reduction features. Furthermore, aided by profiles, our technique can retain more than 80% of the maximal code size savings without affecting performance. View details
Theoretical Perspectives on Data Quality and Synergistic Effects in Pre- and Post-Training Reasoning Models
Adel Javanmard
Baharan Mirzasoleiman
arXiv preprint arXiv:2603.01293 (to appear in ICML 2026) (2026)
Preview abstract We will study a theoretical model (one layer linear self attention with in-context linear task) to rigorously explain several observations (already reported in previous empirical work in the literature) about the characteristics of pretraining and post-training data that are relevant for good performance. Our analysis explain the importance of having a balanced diverse data in pretraining, and hard small set of examples for SFT, and large volume data fro RL. The experiments are on synthetic toy settings and on GPT-2 with synthetic linear regression data. View details
Preview abstract High-volume enterprise service organizations face a persistent challenge in transitioning from reactive support models to proactive, preventative ones. This paper introduces the Agentic Trend-to-Knowledge (ATK) methodology, a novel, autonomous framework designed to address this gap. The ATK methodology employs an AI agent that operates in a recurring, closed loop. It first uses a two-stage process for the autonomous thematic analysis of recent support cases to identify the most significant recurring issue. It then leverages Retrieval-Augmented Generation (RAG) to source relevant institutional knowledge. A key innovation is the agent's adaptive, bimodal response: if relevant knowledge is found, it drafts a proactive communication for human review; if a knowledge gap is detected, it autonomously creates a content creation task for the appropriate team. This transforms the agent from an automation tool into a proactive process owner that creates a virtuous cycle of continuous improvement for both case deflection and knowledge base quality. By automating the entire workflow from insight to action, the ATK framework provides a concrete methodology for shifting from a "human-in-the-loop" to a more strategic "human-on-the-loop" operational paradigm. View details
Managing and Securing Google's Fleet of Multi-Node Servers
Richard Hanley
Havard Skinnemoen
Andrés Lagar-Cavilla
Michael Wong
Jon McCune
Jeff Andersen
Kishan Prasad
Patrick Leis
Shiva Rao
Chris Koch
Jad Baydoun
Anna Sapek
Communications of the ACM, 69:3 (2026), pp. 82 - 92
Preview abstract Server hardware and software co-design for a secure, efficient cloud. View details
Preview abstract Carbon emission at the urban neighborhood scale is significantly influenced by the built environment (BE) factor. This impact is studied from multiple perspectives, including energy demand, renewable energy generation and urban heat island (UHI) effect. Although these domains have been widely investigated, their integrated assessment remains limited, mainly due to the computational demand of urban-scale simulation models and the lack of efficient workflow for sharing data among models. The present study proposes a generative bottom-up framework for multi-domain simulation that integrates physical simulation models in energy demand, photovoltaic (PV) electricity generation and UHI domains in a unified workflow. In the next step, the study generated and evaluated 442 neighborhood massing-scenarios based on eight building typologies. The result shows that connected typologies, through reducing operational energy demand, have a better overall performance than connected typologies, while offering lower photovoltaic generation potential. Volume to surface ratio and Height to Width factor emerged as the prevalent BE factor in many performance metrics. The findings support block-scale urban design strategies favoring larger parcels, higher site coverage ratio, a building height of six to seven stories and connected building typologies to reduce neighborhood-scale carbon emissions. View details
Preview abstract Optimizing large-language model (LLM) training and serving on large-sacle distributed systems with hundreds and thousands of accelerators is always a challenging task due to the fast evloving LLMs, strong domain expertise required, and various optimization goals from different worklaods. Existing methods rely on either handcrafted optimization performed by human experts, which is tedious and time-consuming or resource-intensive black-box searches, which lack the extensibility to keep pace with evolving models and hardware. To address this, we introduce PROMPTS, a novel multi-agent framework that complements traditional search methods with expert-informed reasoning. It automates the diagnosis of performance bottlenecks by synthesizing profiler data and leverages a knowledge base to propose optimized sharding configurations with detailed justifications. Across eight real-world production workloads, PROMPTS demonstrated remarkable efficiency and accuracy, delivering performance improvements of up to 434%. These workloads spanned diverse model architectures, hardware platforms, computational scales, and various stages of the machine learning lifecycle (pre-training, serving, and post-training). In every case, the configuration adopted by human engineers was identified within the agent's top three proposals from a single invocation. Furthermore, the agent's top-ranked recommendation was the one ultimately adopted in 87.5% of cases, showcasing its ability to not only find optimized solutions, but also to correctly prioritize them. Our work establishes PROMPTS as a scalable, extensible, and explainable methodology for AI-assisted performance engineering in large-scale ML systems. View details
×