arXiv subjects

Tony Wu

Publications and source records attributed to Tony Wu.

15 recordsLinked to original sources

arXiv2026

NeoMME: A Single-Tower Multimodal-Native Multilingual Foundation Encoder for Efficient Fine-Tuning and Inference

Multimodal models often build on architectures designed for generative vision-language modeling, typically combining separately pretrained vision encoders with causal language models. Visual document retrievers such as ColPali repurpose these models as encoders, carrying over the parameter and compute overhead of a VLM for a non-generative task. We introduce NeoMME, a family of 260M and 800M-parameter Multimodal and Multilingual bidirectional Encoders that process multilingual text and raw image patches in a single bidirectional Transformer encoder. Both models are pretrained from scratch with a masked discrete-diffusion text objective, conditioned on visible image patches for multimodal examples. Both support a 16,384-token context, enough to encode up to two standard 4K UHD images. To demonstrate its downstream capabilities, we fine-tune NeoMME with jointly trained dense and late-interaction heads. On the ViDoRe v3 benchmark, the resulting NeoMME-Retriever 260M outperforms all evaluated models strictly below 800M parameters with 0.523 nDCG@10, while NeoMME-Retriever 800M reaches 0.556. At a matched 2048x2048 image input size on an NVIDIA L40S, NeoMME-260M encodes pages with about 2x the throughput of ColModernVBERT. Hierarchical token pooling and asymmetric quantization compress late-interaction multimodal document embeddings by 255x while preserving over 95% of baseline nDCG@10. We contribute NeoMME to Hugging Face Transformers and release the pretrained backbone and retrieval-compatible checkpoints under Apache 2.0 at https://hf.co/collections/Hcompany/neomme.

cs.IR↗

arXiv2026

Learning Gaussian Graphical Models from a Glauber Trajectory Without Mixing

We study the task of learning the structure of a $d$-sparse Gaussian graphical model on $n$ variables from a single trajectory of Glauber dynamics. Beyond algorithmic considerations, many applications present temporally correlated observations rather than i.i.d.\ samples. In the classical i.i.d.\ setting, under comparably general sparsity and minimum edge-strength assumptions, sublinear-in-$n$ sample guarantees are known, but achieving them in polynomial-time remains open. Motivated in part by this gap, we give a polynomial-time algorithm that recovers the conditional-independence graph from a single Glauber trajectory, with a trajectory-length guarantee that does not depend on the mixing time. Technically, our algorithm has three components. First, we estimate the conditional variances and rescale the trajectory to reduce to the unit-diagonal case, without changing the underlying graph. Second, we design a local edge test that extracts adjacency information from short update windows by isolating pairwise influence. Third, we aggregate these local statistics using a robust median-based estimator, and prove accuracy despite temporal dependence arising from a single trajectory.

cs.LG↗

arXiv2025

LiteAttention: A Temporal Sparse Attention for Diffusion Transformers

Diffusion Transformers, particularly for video generation, achieve remarkable quality but suffer from quadratic attention complexity, leading to prohibitive latency. Existing acceleration methods face a fundamental trade-off: dynamically estimating sparse attention patterns at each denoising step incurs high computational overhead and estimation errors, while static sparsity patterns remain fixed and often suboptimal throughout denoising. We identify a key structural property of diffusion attention, namely, its sparsity patterns exhibit strong temporal coherence across denoising steps. Tiles deemed non-essential at step $t$ typically remain so at step $t+\delta$. Leveraging this observation, we introduce LiteAttention, a method that exploits temporal coherence to enable evolutionary computation skips across the denoising sequence. By marking non-essential tiles early and propagating skip decisions forward, LiteAttention eliminates redundant attention computations without repeated profiling overheads, combining the adaptivity of dynamic methods with the efficiency of static ones. We implement a highly optimized LiteAttention kernel on top of FlashAttention and demonstrate substantial speedups on production video diffusion models, with no degradation in quality. The code and implementation details will be publicly released.

cs.CV↗

arXiv2025

Surfer 2: The Next Generation of Cross-Platform Computer Use Agents

Building agents that generalize across web, desktop, and mobile environments remains an open challenge, as prior systems rely on environment-specific interfaces that limit cross-platform deployment. We introduce Surfer 2, a unified architecture operating purely from visual observations that achieves state-of-the-art performance across all three environments. Surfer 2 integrates hierarchical context management, decoupled planning and execution, and self-verification with adaptive recovery, enabling reliable operation over long task horizons. Our system achieves 97.1% accuracy on WebVoyager, 69.6% on WebArena, 60.1% on OSWorld, and 87.1% on AndroidWorld, outperforming all prior systems without task-specific fine-tuning. With multiple attempts, Surfer 2 exceeds human performance on all benchmarks. These results demonstrate that systematic orchestration amplifies foundation model capabilities and enables general-purpose computer control through visual interaction alone, while calling for a next-generation vision language model to achieve Pareto-optimal cost-efficiency.

cs.AI↗

arXiv2025

Surfer-H Meets Holo1: Cost-Efficient Web Agent Powered by Open Weights

We present Surfer-H, a cost-efficient web agent that integrates Vision-Language Models (VLM) to perform user-defined tasks on the web. We pair it with Holo1, a new open-weight collection of VLMs specialized in web navigation and information extraction. Holo1 was trained on carefully curated data sources, including open-access web content, synthetic examples, and self-produced agentic data. Holo1 tops generalist User Interface (UI) benchmarks as well as our new web UI localization benchmark, WebClick. When powered by Holo1, Surfer-H achieves a 92.2% state-of-the-art performance on WebVoyager, striking a Pareto-optimal balance between accuracy and cost-efficiency. To accelerate research advancement in agentic systems, we are open-sourcing both our WebClick evaluation dataset and the Holo1 model weights.

cs.AI↗

arXiv2025

Quantifying and Modeling Driving Styles in Trajectory Forecasting

Trajectory forecasting has become a popular deep learning task due to its relevance for scenario simulation for autonomous driving. Specifically, trajectory forecasting predicts the trajectory of a short-horizon future for specific human drivers in a particular traffic scenario. Robust and accurate future predictions can enable autonomous driving planners to optimize for low-risk and predictable outcomes for human drivers around them. Although some work has been done to model driving style in planning and personalized autonomous polices, a gap exists in explicitly modeling human driving styles for trajectory forecasting of human behavior. Human driving style is most certainly a correlating factor to decision making, especially in edge-case scenarios where risk is nontrivial, as justified by the large amount of traffic psychology literature on risky driving. So far, the current real-world datasets for trajectory forecasting lack insight on the variety of represented driving styles. While the datasets may represent real-world distributions of driving styles, we posit that fringe driving style types may also be correlated with edge-case safety scenarios. In this work, we conduct analyses on existing real-world trajectory datasets for driving and dissect these works from the lens of driving styles, which is often intangible and non-standardized.

cs.RO↗

arXiv2024

ColPali: Efficient Document Retrieval with Vision Language Models

Documents are visually rich structures that convey information through text, but also figures, page layouts, tables, or even fonts. Since modern retrieval systems mainly rely on the textual information they extract from document pages to index documents -often through lengthy and brittle processes-, they struggle to exploit key visual cues efficiently. This limits their capabilities in many practical document retrieval applications such as Retrieval Augmented Generation (RAG). To benchmark current systems on visually rich document retrieval, we introduce the Visual Document Retrieval Benchmark ViDoRe, composed of various page-level retrieval tasks spanning multiple domains, languages, and practical settings. The inherent complexity and performance shortcomings of modern systems motivate a new concept; doing document retrieval by directly embedding the images of the document pages. We release ColPali, a Vision Language Model trained to produce high-quality multi-vector embeddings from images of document pages. Combined with a late interaction matching mechanism, ColPali largely outperforms modern document retrieval pipelines while being drastically simpler, faster and end-to-end trainable. We release models, data, code and benchmarks under open licenses at https://hf.co/vidore.

cs.IR↗

arXiv2024

Adaptive Sensitivity Analysis for Robust Augmentation against Natural Corruptions in Image Segmentation

Achieving robustness in image segmentation models is challenging due to the fine-grained nature of pixel-level classification. These models, which are crucial for many real-time perception applications, particularly struggle when faced with natural corruptions in the wild for autonomous systems. While sensitivity analysis can help us understand how input variables influence model outputs, its application to natural and uncontrollable corruptions in training data is computationally expensive. In this work, we present an adaptive, sensitivity-guided augmentation method to enhance robustness against natural corruptions. Our sensitivity analysis on average runs 10x faster and requires about 200x less storage than previous sensitivity analysis, enabling practical, on-the-fly estimation during training for a model-free augmentation policy. With minimal fine-tuning, our sensitivity-guided augmentation method achieves improved robustness on both real-world and synthetic datasets compared to state-of-the-art data augmentation techniques in image segmentation. Code implementation for this work can be found at: https://github.com/laurayuzheng/SensAug.

cs.CV↗

arXiv2021

Learning Interpretable Representations of Entanglement in Quantum Optics Experiments using Deep Generative Models

Quantum physics experiments produce interesting phenomena such as interference or entanglement, which are core properties of numerous future quantum technologies. The complex relationship between the setup structure of a quantum experiment and its entanglement properties is essential to fundamental research in quantum optics but is difficult to intuitively understand. We present a deep generative model of quantum optics experiments where a variational autoencoder is trained on a dataset of quantum optics experimental setups. In a series of computational experiments, we investigate the learned representation of our Quantum Optics Variational Auto Encoder (QOVAE) and its internal understanding of the quantum optics world. We demonstrate that the QOVAE learns an interpretable representation of quantum optics experiments and the relationship between experiment structure and entanglement. We show the QOVAE is able to generate novel experiments for highly entangled quantum states with specific distributions that match its training data. The QOVAE can learn to generate specific entangled states and efficiently search the space of experiments that produce highly entangled quantum states. Importantly, we are able to interpret how the QOVAE structures its latent space, finding curious patterns that we can explain in terms of quantum physics. The results demonstrate how we can use and understand the internal representations of deep generative models in a complex scientific domain. The QOVAE and the insights from our investigations can be immediately applied to other physical systems.

cs.LG↗

arXiv2021

QUEACO: Borrowing Treasures from Weakly-labeled Behavior Data for Query Attribute Value Extraction

We study the problem of query attribute value extraction, which aims to identify named entities from user queries as diverse surface form attribute values and afterward transform them into formally canonical forms. Such a problem consists of two phases: {named entity recognition (NER)} and {attribute value normalization (AVN)}. However, existing works only focus on the NER phase but neglect equally important AVN. To bridge this gap, this paper proposes a unified query attribute value extraction system in e-commerce search named QUEACO, which involves both two phases. Moreover, by leveraging large-scale weakly-labeled behavior data, we further improve the extraction performance with less supervision cost. Specifically, for the NER phase, QUEACO adopts a novel teacher-student network, where a teacher network that is trained on the strongly-labeled data generates pseudo-labels to refine the weakly-labeled data for training a student network. Meanwhile, the teacher network can be dynamically adapted by the feedback of the student's performance on strongly-labeled data to maximally denoise the noisy supervisions from the weak labels. For the AVN phase, we also leverage the weakly-labeled query-to-attribute behavior data to normalize surface form attribute values from queries into canonical forms from products. Extensive experiments on a real-world large-scale E-commerce dataset demonstrate the effectiveness of QUEACO.

cs.CL↗

arXiv2021

Graph-based Multilingual Product Retrieval in E-commerce Search

Nowadays, with many e-commerce platforms conducting global business, e-commerce search systems are required to handle product retrieval under multilingual scenarios. Moreover, comparing with maintaining per-country specific e-commerce search systems, having a universal system across countries can further reduce the operational and computational costs, and facilitate business expansion to new countries. In this paper, we introduce a universal end-to-end multilingual retrieval system, and discuss our learnings and technical details when training and deploying the system to serve billion-scale product retrieval for e-commerce search. In particular, we propose a multilingual graph attention based retrieval network by leveraging recent advances in transformer-based multilingual language models and graph neural network architectures to capture the interactions between search queries and items in e-commerce search. Offline experiments on five countries data show that our algorithm outperforms the state-of-the-art baselines by 35% recall and 25% mAP on average. Moreover, the proposed model shows significant increase of conversion/revenue in online A/B experiments and has been deployed in production for multiple countries.

cs.CL↗

arXiv2020

Neural Message Passing on High Order Paths

Graph neural network have achieved impressive results in predicting molecular properties, but they do not directly account for local and hidden structures in the graph such as functional groups and molecular geometry. At each propagation step, GNNs aggregate only over first order neighbours, ignoring important information contained in subsequent neighbours as well as the relationships between those higher order connections. In this work, we generalize graph neural nets to pass messages and aggregate across higher order paths. This allows for information to propagate over various levels and substructures of the graph. We demonstrate our model on a few tasks in molecular property prediction.

cs.LG↗

arXiv2020

Graph Deconvolutional Generation

Graph generation is an extremely important task, as graphs are found throughout different areas of science and engineering. In this work, we focus on the modern equivalent of the Erdos-Renyi random graph model: the graph variational autoencoder (GVAE). This model assumes edges and nodes are independent in order to generate entire graphs at a time using a multi-layer perceptron decoder. As a result of these assumptions, GVAE has difficulty matching the training distribution and relies on an expensive graph matching procedure. We improve this class of models by building a message passing neural network into GVAE's encoder and decoder. We demonstrate our model on the specific task of generating small organic molecules

cs.LG↗

arXiv2016

Reviewer Integration and Performance Measurement for Malware Detection

We present and evaluate a large-scale malware detection system integrating machine learning with expert reviewers, treating reviewers as a limited labeling resource. We demonstrate that even in small numbers, reviewers can vastly improve the system's ability to keep pace with evolving threats. We conduct our evaluation on a sample of VirusTotal submissions spanning 2.5 years and containing 1.1 million binaries with 778GB of raw feature data. Without reviewer assistance, we achieve 72% detection at a 0.5% false positive rate, performing comparable to the best vendors on VirusTotal. Given a budget of 80 accurate reviews daily, we improve detection to 89% and are able to detect 42% of malicious binaries undetected upon initial submission to VirusTotal. Additionally, we identify a previously unnoticed temporal inconsistency in the labeling of training datasets. We compare the impact of training labels obtained at the same time training data is first seen with training labels obtained months later. We find that using training labels obtained well after samples appear, and thus unavailable in practice for current training data, inflates measured detection by almost 20 percentage points. We release our cluster-based implementation, as well as a list of all hashes in our evaluation and 3% of our entire dataset.

cs.CR↗

arXiv2015

Brewing Analytics Quality for Cloud Performance

Cloud computing has become increasingly popular. Many options of cloud deployments are available. Testing cloud performance would enable us to choose a cloud deployment based on the requirements. In this paper, we present an innovative process, implemented in software, to allow us to assess the quality of the cloud performance data. The process combines performance data from multiple machines, spanning across user experience data, workload performance metrics, and readily available system performance data. Furthermore, we discuss the major challenges of bringing raw data into tidy data formats in order to enable subsequent analysis, and describe how our process has several layers of assessment to validate the quality of the data processing procedure. We present a case study to demonstrate the effectiveness of our proposed process, and conclude our paper with several future research directions worth investigating.

cs.PF↗