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 10292 publications
    Preview abstract We study the existence of almost fair and near-optimal solutions to a routing problem as defined in the seminal work of Rosenthal. We focus on the setting where multiple alternative routes are available for each potential request (which corresponds to a potential user of the network). This model captures a collection of diverse applications such as packet routing in communication networks, routing in road networks with multiple alternative routes, and the economics of transportation of goods. Our recommended routes have provable guarantees in terms of both the total cost and fairness concepts such as approximate envy-freeness. We employ and appropriately combine tools from algorithmic game theory and fair division. Our results apply on two distinct models: the splittable case where the request is split among the selected paths (e.g., routing a fleet of trucks) and the unsplittable case where the request is assigned to one of its designated paths (e.g., a single user request). Finally, we conduct an empirical analysis to test the performance of our approach against simpler baselines using the real world road network of New York City. View details
    Scaling Laws for Downstream Task Performance in Machine Translation
    Natalia Ponomareva
    Hussein Hazimeh
    Sanmi Koyejo
    International Conference on Learning Representations (ICLR) (2025) (to appear)
    Preview abstract Scaling laws provide important insights that can guide the design of large language models (LLMs). Existing work has primarily focused on studying scaling laws for pretraining (upstream) loss. However, in transfer learning settings, in which LLMs are pretrained on an unsupervised dataset and then finetuned on a downstream task, we often also care about the downstream performance. In this work, we study the scaling behavior in a transfer learning setting, where LLMs are finetuned for machine translation tasks. Specifically, we investigate how the choice of the \emph{pretraining} data and its size affect downstream performance (translation quality) as judged by: downstream cross-entropy and translation quality metrics such as BLEU and COMET scores. Our experiments indicate that the size of the finetuning dataset and the distribution alignment between the pretraining and downstream data significantly influence the scaling behavior. With sufficient alignment, both downstream cross-entropy and translation quality scores improve monotonically with more pretraining data. In such cases, we show that it is possible to predict the downstream translation quality metrics with good accuracy using a log-law. However, there are cases where moderate misalignment causes the downstream translation scores to fluctuate or get worse with more pretraining, whereas downstream cross-entropy monotonically improves. By analyzing these, we provide new practical insights for choosing appropriate pretraining data. View details
    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
    Security Signals: Making Web Security Posture Measurable At Scale
    David Dworken
    Artur Janc
    Santiago (Sal) Díaz
    Workshop on Measurements, Attacks, and Defenses for the Web (MADWeb)
    Preview abstract The area of security measurability is gaining increased attention, with a wide range of organizations calling for the development of scalable approaches for assessing the security of software systems and infrastructure. In this paper, we present our experience developing Security Signals, a comprehensive system providing security measurability for web services, deployed in a complex application ecosystem of thousands of web services handling traffic from billions of users. The system collects security-relevant information from production HTTP traffic at the reverse proxy layer, utilizing novel concepts such as synthetic signals augmented with additional risk information to provide a holistic view of the security posture of individual services and the broader application ecosystem. This approach to measurability has enabled large-scale security improvements to our services, including prioritized rollouts of security enhancements and the implementation of automated regression monitoring. Furthermore, it has proven valuable for security research and prioritization of defensive work. Security Signals addresses shortcomings of prior web measurability proposals by tracking a comprehensive set of security properties relevant to web applications, and by extracting insights from collected data for use by both security experts and non-experts. We believe the lessons learned from the implementation and use of Security Signals offer valuable insights for practitioners responsible for web service security, potentially inspiring new approaches to web security measurability. View details
    Sufficient Context: A New Lens on Retrieval Augmented Generation Systems
    Hailey Joren
    Jianyi Zhang
    Chun-Sung Ferng
    Ankur Taly
    International Conference on Learning Representations (ICLR) (2025)
    Preview abstract Augmenting LLMs with context leads to improved performance across many applications. Despite much research on Retrieval Augmented Generation (RAG) systems, an open question is whether errors arise because LLMs fail to utilize the context from retrieval or the context itself is insufficient to answer the query. To shed light on this, we develop a new notion of sufficient context, along with a method to classify instances that have enough information to answer the query. We then use sufficient context to analyze several models and datasets. By stratifying errors based on context sufficiency, we find that larger models with higher baseline performance (Gemini 1.5 Pro, GPT 4o, Claude 3.5) excel at answering queries when the context is sufficient, but often output incorrect answers instead of abstaining when the context is not. On the other hand, smaller models with lower baseline performance (Llama 3.1, Mistral 3, Gemma 2) hallucinate or abstain often, even with sufficient context. We further categorize cases when the context is useful, and improves accuracy, even though it does not fully answer the query and the model errs without the context. Building on our findings, we explore ways to reduce hallucinations in RAG systems, including a new selective generation method that leverages sufficient context information for guided abstention. Our method improves the fraction of correct answers among times where the model responds by 2--10% for Gemini, GPT, and Gemma. View details
    Enhancing Remote Sensing Representations through Mixed-Modality Masked Autoencoding
    Ori Linial
    George Leifman
    Yochai Blau
    Nadav Sherman
    Yotam Gigi
    Wojciech Sirko
    Proceedings of the Winter Conference on Applications of Computer Vision (WACV) Workshops (2025), pp. 507-516
    Preview abstract This paper presents an innovative approach to pre-training models for remote sensing by integrating optical and radar data from Sentinel-2 and Sentinel-1 satellites. Using a novel variation on the masked autoencoder (MAE) framework, our model incorporates a dual-task setup: reconstructing masked Sentinel-2 images and predicting corresponding Sentinel-1 images. This multi-task design enables the encoder to capture both spectral and structural features across diverse environmental conditions. Additionally, we introduce a "mixing" strategy in the pretraining phase, combining patches from both image sources, which mitigates spatial misalignment errors and enhances model robustness. Evaluation on segmentation and classification tasks, including Sen1Floods11 and BigEarthNet, demonstrates significant improvements in adaptability and generalizability across varied downstream remote sensing applications. Our findings highlight the advantages of leveraging complementary modalities for more resilient and versatile land cover analysis. 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 Despite the surge in popularity of virtual reality (VR), mobile phones remain the primary medium for accessing digital content, offering both privacy and portability. This short paper presents Beyond the Phone, a novel framework that enhances mobile phones in VR with context-aware controls and spatial augmentation. We first establish a comprehensive design space through brainstorming and iterative discussions with VR experts. We then develop a proof-of-concept system that analyzes UI layouts to offer context-aware controls and spatial augmentation, targeting six key application areas within our design space. Finally, we demonstrate that our system can effectively adapt to a broad spectrum of applications at runtime, and discuss future directions with reviews with seven experts. View details
    Zero-Shot Image Moderation in Google Ads with LLM-Assisted Textual Descriptions and Cross-modal Co-embeddings
    Jimin Li
    Eric Xiao
    Katie Warren
    Enming Luo
    Krishna Viswanathan
    Ariel Fuxman
    Bill Li
    Yintao Liu
    (2025)
    Preview abstract We present a scalable and agile approach for ads image content moderation at Google, addressing the challenges of moderating massive volumes of ads with diverse content and evolving policies. The proposed method utilizes human-curated textual descriptions and cross-modal text-image co-embeddings to enable zero-shot classification of policy violating ads images, bypassing the need for extensive supervised training data and human labeling. By leveraging large language models (LLMs) and user expertise, the system generates and refines a comprehensive set of textual descriptions representing policy guidelines. During inference, co-embedding similarity between incoming images and the textual descriptions serves as a reliable signal for policy violation detection, enabling efficient and adaptable ads content moderation. Evaluation results demonstrate the efficacy of this framework in significantly boosting the detection of policy violating content. View details
    PreFix: Optimizing the Performance of Heap-Intensive Applications
    Chaitanya Mamatha Ananda
    Rajiv Gupta
    Han Shen
    CGO 2025: International Symposium on Code Generation and Optimization, Las Vegas, NV, USA (to appear)
    Preview abstract Analyses of heap-intensive applications show that a small fraction of heap objects account for the majority of heap accesses and data cache misses. Prior works like HDS and HALO have shown that allocating hot objects in separate memory regions can improve spatial locality leading to better application performance. However, these techniques are constrained in two primary ways, limiting their gains. First, these techniques have Imperfect Separation, polluting the hot memory region with several cold objects. Second, reordering of objects across allocations is not possible as the original object allocation order is preserved. This paper presents a novel technique that achieves near perfect separation of hot objects via a new context mechanism that efficiently identifies hot objects with high precision. This technique, named PreFix, is based upon Preallocating memory for a Fixed small number of hot objects. The program, guided by profiles, is instrumented to compute context information derived from dynamic object identifiers, that precisely identifies hot object allocations that are then placed at predetermined locations in the preallocated memory. The preallocated memory region for hot objects provides the flexibility to reorder objects across allocations and allows colocation of objects that are part of a hot data stream (HDS), improving spatial locality. The runtime overhead of identifying hot objects is not significant as this optimization is only focused on a small number of static hot allocation sites and dynamic hot objects. While there is an increase in the program’s memory foot-print, it is manageable and can be controlled by limiting the size of the preallocated memory. In addition, PreFix incorporates an object recycling optimization that reuses the same preallocated space to store different objects whose lifetimes are not expected to overlap. Our experiments with 13 heap-intensive applications yields reductions in execution times ranging from 2.77% to 74%. On average PreFix reduces execution time by 21.7% compared to 7.3% by HDS and 14% by HALO. This is due to PreFix’s precision in hot object identification, hot object colocation, and low runtime overhead. View details
    Enhancing Remote Sensing Representations through Mixed-Modality Masked Autoencoding
    Ori Linial
    Yochai Blau
    Nadav Sherman
    Yotam Gigi
    Wojciech Sirko
    Proceedings of the Winter Conference on Applications of Computer Vision (WACV) Workshops (2025), pp. 507-516
    Preview abstract This paper presents an innovative approach to pre-training models for remote sensing by integrating optical and radar data from Sentinel-2 and Sentinel-1 satellites. Using a novel variation on the masked autoencoder (MAE) framework, our model incorporates a dual-task setup: reconstructing masked Sentinel-2 images and predicting corresponding Sentinel-1 images. This multi-task design enables the encoder to capture both spectral and structural features across diverse environmental conditions. Additionally, we introduce a "mixing" strategy in the pretraining phase, combining patches from both image sources, which mitigates spatial misalignment errors and enhances model robustness. Evaluation on segmentation and classification tasks, including Sen1Floods11 and BigEarthNet, demonstrates significant improvements in adaptability and generalizability across varied downstream remote sensing applications. Our findings highlight the advantages of leveraging complementary modalities for more resilient and versatile land cover analysis. View details
    Preview abstract Recent work suggested utilizing inference compute, showing that scaling of number of samples consistently improves the fractions of problems solved by any attempt, namely the coverage. In this work, we suggest that inference scaling gains should be compared with proper baselines, as some datasets become degenerate when allowing a large number of attempts. We focus on two domains - mathematical reasoning and factual knowledge, showing that for the MATH and Entity Questions datasets, informed answer enumeration obtains similar or even better results than repeated model sampling, with a much lower sample budget. While we believe that inference scaling is a promising approach for unlocking the potential of language models, we recommend carefully selecting models and datasets when applying this method. Otherwise, the results of inference scaling should be interpreted with caution. View details
    Preview abstract In today's rapidly evolving business landscape, Governance, Risk, and Compliance (GRC) leaders in large, complex organizations face unprecedented challenges. The cloud has revolutionized how businesses operate, offering unprecedented scalability, flexibility, cost-efficiency, additional security and resilience. However, this transformation also presents new challenges for GRC professionals. In a cloud-native world, where applications are built and deployed in dynamic, distributed environments, traditional GRC on-prem approaches, manual processes and spreadsheets struggle to keep pace. The key to success lies in embracing a data-driven GRC strategy that leverages the power of the cloud to enhance agility, visibility, and resilience. View details
    Mufu: Multilingual Fused Learning for Low- Resource Translation with LLM
    Zheng Lim
    Honglin Yu
    Trevor Cohn
    International Conference on Learning Representations (ICLR) 2025
    Preview abstract Multilingual large language models (LLMs) are great translators, but this is largely limited to high-resource languages. For many LLMs, translating in and out of low-resource languages remains a challenging task. To maximize data efficiency in this low-resource setting, we introduce Mufu, which includes a selection of automatically generated multilingual candidates and an instruction to correct inaccurate translations in the prompt. Mufu prompts turn a translation task into a postediting one, and seek to harness the LLM's reasoning capability with auxiliary translation candidates, from which the model is required to assess the input quality, align the semantics cross-lingually, copy from relevant inputs and override instances that are incorrect. Our experiments on En-XX translations over the Flores-200 dataset show LLMs finetuned against Mufu-style prompts are robust to poor quality auxiliary translation candidates, achieving performance superior to NLLB 1.3B distilled model in 64% of low- and very-low-resource language pairs. We then distill these models to reduce inference cost, while maintaining on average 3.1 chrF improvement over finetune-only baseline in low-resource translations. View details
    Gemini & Physical World: Large Language Models Can Estimate the Intensity of Earthquake Shaking from Multi-Modal Social Media Posts
    Marc Stogaitis
    Tajinder Gadh
    Richard Allen
    Alexei Barski
    Robert Bosch
    Patrick Robertson
    Youngmin Cho
    Nivetha Thiruverahan
    Aman Raj
    Geophysical Journal International (2025), ggae436
    Preview abstract This paper presents a novel approach for estimating the ground shaking intensity using real-time social media data and CCTV footage. Employing the Gemini 1.5 Pro’s (Reid et al. 2024) model, a multi-modal language model, we demonstrate the ability to extract relevant information from unstructured data utilizing generative AI and natural language processing. The model’s output, in the form of Modified Mercalli Intensity (MMI) values, align well with independent observational data. Furthermore, our results suggest that beyond its advanced visual and auditory understanding abilities, Gemini appears to utilize additional sources of knowledge, including a simplified understanding of the general relationship between earthquake magnitude, distance, and MMI intensity, which it presumably acquired during its training, in its reasoning and decision-making processes. These findings raise intriguing questions about the extent of Gemini's general understanding of the physical world and its phenomena. Gemini’s ability to generate results consistent with established scientific knowledge highlights the potential of LLMs like Gemini in augmenting our understanding of complex physical phenomena such as earthquakes. More specifically, the results of this study highlight the potential of LLMs like Gemini to revolutionize citizen seismology by enabling rapid, effective, and flexible analysis of crowdsourced data from eyewitness accounts for assessing earthquake impact and providing crisis situational awareness. This approach holds a great promise for improving early warning systems, disaster response, and overall resilience in earthquake-prone regions. This study provides a significant step toward harnessing the power of social media and AI for earthquake disaster mitigation. View details