Searcharxiv⌕ Search

arXiv subjects

Tao Yu

Publications and source records attributed to Tao Yu.

At least 145 records · Page 8Linked to original sources
arXiv2025

Digital Twin-Enabled Blockage-Aware Dynamic mmWave Multi-Hop V2X Communication

Millimeter wave (mmWave) technology in vehicle-to-everything (V2X) communication offers unprecedented data rates and low latency, but faces significant reliability challenges due to signal blockages and limited range. This paper introduces a novel system for managing dynamic multi-hop mmWave V2X communications in complex blocking environments. We present a system architecture that integrates a mobility digital twin (DT) with the multi-hop routing control plane, providing a comprehensive, real-time view of the network and its surrounding traffic environment. This integration enables the control plane to make informed routing decisions based on rich contextual data about vehicles, infrastructure, and potential signal blockages. Leveraging this DT-enhanced architecture, we propose an advanced routing algorithm that combines high-precision environmental data with trajectory prediction to achieve blockage-aware mmWave multi-hop V2X routing. Our algorithm anticipates network topology changes and adapts topology dynamically to maintain reliable connections. We evaluate our approach through proof-of-concept simulations using a mobility DT of the Nishishinjuku area. Results demonstrate that our DT-enabled routing strategy significantly outperforms conventional methods in maintaining reliable mmWave V2X connections across various traffic scenarios, including fully connected and mixed traffic environments.

cs.NI↗
arXiv2025

Generative Representational Instruction Tuning

All text-based language problems can be reduced to either generation or embedding. Current models only perform well at one or the other. We introduce generative representational instruction tuning (GRIT) whereby a large language model is trained to handle both generative and embedding tasks by distinguishing between them through instructions. Compared to other open models, our resulting GritLM 7B sets a new state of the art on the Massive Text Embedding Benchmark (MTEB) and outperforms all models up to its size on a range of generative tasks. By scaling up further, GritLM 8x7B outperforms all open generative language models that we tried while still being among the best embedding models. Notably, we find that GRIT matches training on only generative or embedding data, thus we can unify both at no performance loss. Among other benefits, the unification via GRIT speeds up Retrieval-Augmented Generation (RAG) by > 60% for long documents, by no longer requiring separate retrieval and generation models. Models, code, etc. are freely available at https://github.com/ContextualAI/gritlm.

cs.CL↗
arXiv2025

AgentTrek: Agent Trajectory Synthesis via Guiding Replay with Web Tutorials

Graphical User Interface (GUI) agents can automate complex tasks across digital environments, but their development is hindered by the scarcity of high-quality trajectory data for training. Existing approaches rely on expensive human annotation, making them unsustainable at scale. We propose AgentTrek, a scalable data synthesis pipeline that generates web agent trajectories by leveraging publicly available tutorials. Our three-stage method: (1) automatically harvests and filters tutorial-like texts from the internet using a specialized classification model, (2) transforms these texts into structured task specifications with step-by-step instructions, and (3) employs a visual-language model (VLM) agent to execute these instructions in real environments, while a VLM-based evaluator verifies trajectory correctness. The synthesized trajectories encompass multiple modalities, including text-based HTML observations with function-calling API actions, and vision-based screenshot observations with pixel-level actions. This multimodal data, enriched with chain-of-thought reasoning, enables agents to achieve state-of-the-art performance on both textual web browsing benchmarks (e.g., WebArena) and visual web grounding and browsing benchmarks (e.g., ScreenSpot Web and Multimodal Mind2Web). Furthermore, our fully automated approach significantly reduces data collection costs, achieving a cost of just $0.55 per high-quality trajectory without human annotators. Our work demonstrates that guided replay using web tutorials is a practical and scalable strategy for training advanced GUI agents, paving the way for more capable and autonomous digital assistants.

cs.CL↗
arXiv2025

Stochastic Rounding for LLM Training: Theory and Practice

As the parameters of Large Language Models (LLMs) have scaled to hundreds of billions, the demand for efficient training methods -- balancing faster computation and reduced memory usage without sacrificing accuracy -- has become more critical than ever. In recent years, various mixed precision strategies, which involve different precision levels for optimization components, have been proposed to increase training speed with minimal accuracy degradation. However, these strategies often require manual adjustments and lack theoretical justification. In this work, we leverage stochastic rounding (SR) to address numerical errors of training with low-precision representation. We provide theoretical analyses of implicit regularization and convergence under the Adam optimizer when SR is utilized. With the insights from these analyses, we extend previous BF16 + SR strategy to be used in distributed settings, enhancing the stability and performance for large scale training. Empirical results from pre-training models with up to 6.7B parameters, for the first time, demonstrate that our BF16 with SR strategy outperforms (BF16, FP32) mixed precision strategies, achieving better validation perplexity, up to $1.54\times$ higher throughput, and $30\%$ less memory usage.

cs.LG↗
arXiv2025

Systematic Outliers in Large Language Models

Outliers have been widely observed in Large Language Models (LLMs), significantly impacting model performance and posing challenges for model compression. Understanding the functionality and formation mechanisms of these outliers is critically important. Existing works, however, largely focus on reducing the impact of outliers from an algorithmic perspective, lacking an in-depth investigation into their causes and roles. In this work, we provide a detailed analysis of the formation process, underlying causes, and functions of outliers in LLMs. We define and categorize three types of outliers-activation outliers, weight outliers, and attention outliers-and analyze their distributions across different dimensions, uncovering inherent connections between their occurrences and their ultimate influence on the attention mechanism. Based on these observations, we hypothesize and explore the mechanisms by which these outliers arise and function, demonstrating through theoretical derivations and experiments that they emerge due to the self-attention mechanism's softmax operation. These outliers act as implicit context-aware scaling factors within the attention mechanism. As these outliers stem from systematic influences, we term them systematic outliers. Our study not only enhances the understanding of Transformer-based LLMs but also shows that structurally eliminating outliers can accelerate convergence and improve model compression. The code is avilable at https://github.com/an-yongqi/systematic-outliers.

cs.CL↗
arXiv2025

MM-RLHF: The Next Step Forward in Multimodal LLM Alignment

Despite notable advancements in Multimodal Large Language Models (MLLMs), most state-of-the-art models have not undergone thorough alignment with human preferences. This gap exists because current alignment research has primarily achieved progress in specific areas (e.g., hallucination reduction), while the broader question of whether aligning models with human preferences can systematically enhance MLLM capability remains largely unexplored. To this end, we introduce MM-RLHF, a dataset containing $\mathbf{120k}$ fine-grained, human-annotated preference comparison pairs. This dataset represents a substantial advancement over existing resources, offering superior size, diversity, annotation granularity, and quality. Leveraging this dataset, we propose several key innovations to improve both the quality of reward models and the efficiency of alignment algorithms. Notably, we introduce a Critique-Based Reward Model, which generates critiques of model outputs before assigning scores, offering enhanced interpretability and more informative feedback compared to traditional scalar reward mechanisms. Additionally, we propose Dynamic Reward Scaling, a method that adjusts the loss weight of each sample according to the reward signal, thereby optimizing the use of high-quality comparison pairs. Our approach is rigorously evaluated across $\mathbf{10}$ distinct dimensions and $\mathbf{27}$ benchmarks, with results demonstrating significant and consistent improvements in model performance. Specifically, fine-tuning LLaVA-ov-7B with MM-RLHF and our alignment algorithm leads to a $\mathbf{19.5}$% increase in conversational abilities and a $\mathbf{60}$% improvement in safety. We have open-sourced the preference dataset, reward model, training and evaluation code, as well as reward modeling and safety benchmarks. For more details, please visit our project page: https://mm-rlhf.github.io.

cs.CL↗
arXiv2025

CIRCUIT: A Benchmark for Circuit Interpretation and Reasoning Capabilities of LLMs

The role of Large Language Models (LLMs) has not been extensively explored in analog circuit design, which could benefit from a reasoning-based approach that transcends traditional optimization techniques. In particular, despite their growing relevance, there are no benchmarks to assess LLMs' reasoning capability about circuits. Therefore, we created the CIRCUIT dataset consisting of 510 question-answer pairs spanning various levels of analog-circuit-related subjects. The best-performing model on our dataset, GPT-4o, achieves 48.04% accuracy when evaluated on the final numerical answer. To evaluate the robustness of LLMs on our dataset, we introduced a unique feature that enables unit-test-like evaluation by grouping questions into unit tests. In this case, GPT-4o can only pass 27.45% of the unit tests, highlighting that the most advanced LLMs still struggle with understanding circuits, which requires multi-level reasoning, particularly when involving circuit topologies. This circuit-specific benchmark highlights LLMs' limitations, offering valuable insights for advancing their application in analog integrated circuit design.

cs.LG↗
arXiv2025

Learn-by-interact: A Data-Centric Framework for Self-Adaptive Agents in Realistic Environments

Autonomous agents powered by large language models (LLMs) have the potential to enhance human capabilities, assisting with digital tasks from sending emails to performing data analysis. The abilities of existing LLMs at such tasks are often hindered by the lack of high-quality agent data from the corresponding environments they interact with. We propose Learn-by-interact, a data-centric framework to adapt LLM agents to any given environments without human annotations. Learn-by-interact synthesizes trajectories of agent-environment interactions based on documentations, and constructs instructions by summarizing or abstracting the interaction histories, a process called backward construction. We assess the quality of our synthetic data by using them in both training-based scenarios and training-free in-context learning (ICL), where we craft innovative retrieval approaches optimized for agents. Extensive experiments on SWE-bench, WebArena, OSWorld and Spider2-V spanning across realistic coding, web, and desktop environments show the effectiveness of Learn-by-interact in various downstream agentic tasks -- baseline results are improved by up to 12.2\% for ICL with Claude-3.5 and 19.5\% for training with Codestral-22B. We further demonstrate the critical role of backward construction, which provides up to 14.0\% improvement for training. Our ablation studies demonstrate the efficiency provided by our synthesized data in ICL and the superiority of our retrieval pipeline over alternative approaches like conventional retrieval-augmented generation (RAG). We expect that Learn-by-interact will serve as a foundation for agent data synthesis as LLMs are increasingly deployed at real-world environments.

cs.LG↗
arXiv2025

Spin Quenching and Transport by Hidden Dzyaloshinskii-Moriya Interactions

Explicit interactions, \textit{e.g.}, dipolar and exchange couplings, usually govern magnetization dynamics. Some interactions may be hidden from the global crystal symmetry. We report that in a large class of \textit{uniaxial} antiferromagnets, a \textit{hidden} Dzyaloshinskii-Moriya interaction with retaining global inversion symmetry quenches the spin of magnon along the Néel vector ${\bf n}$, thus forbidding its angular-momentum flow. Some magnon spins, termed ``nodal" and ``corner" spins, survive when they distribute \textit{singularly} at the hot spots, i.e., high-symmetric degeneracy points in the Brillouin zone, and are protected by crystal symmetries. The biased magnetic field along ${\bf n}$ broadens such distributions, allowing bulk spin transport with unique signatures in the magnetic field and temperature dependencies. This explains recent experiments and highlights the role of hidden interaction.

cond-mat.mes-hall↗
arXiv2024

Momentum Approximation in Asynchronous Private Federated Learning

Asynchronous protocols have been shown to improve the scalability of federated learning (FL) with a massive number of clients. Meanwhile, momentum-based methods can achieve the best model quality in synchronous FL. However, naively applying momentum in asynchronous FL algorithms leads to slower convergence and degraded model performance. It is still unclear how to effective combinie these two techniques together to achieve a win-win. In this paper, we find that asynchrony introduces implicit bias to momentum updates. In order to address this problem, we propose momentum approximation that minimizes the bias by finding an optimal weighted average of all historical model updates. Momentum approximation is compatible with secure aggregation as well as differential privacy, and can be easily integrated in production FL systems with a minor communication and storage cost. We empirically demonstrate that on benchmark FL datasets, momentum approximation can achieve $1.15 \textrm{--}4\times$ speed up in convergence compared to naively combining asynchronous FL with momentum.

cs.LG↗
arXiv2024

EVOR: Evolving Retrieval for Code Generation

Recently the retrieval-augmented generation (RAG) has been successfully applied in code generation. However, existing pipelines for retrieval-augmented code generation (RACG) employ static knowledge bases with a single source, limiting the adaptation capabilities of Large Language Models (LLMs) to domains they have insufficient knowledge of. In this work, we develop a novel pipeline, EVOR, that employs the synchronous evolution of both queries and diverse knowledge bases. On two realistic settings where the external knowledge is required to solve code generation tasks, we compile four new datasets associated with frequently updated libraries and long-tail programming languages, named EVOR-BENCH. Extensive experiments demonstrate that EVOR achieves two to four times of execution accuracy compared to other methods such as Reflexion (Shinn et al., 2024), DocPrompting (Zhou et al., 2023), etc. We demonstrate that EVOR is flexible and can be easily combined with them to achieve further improvement. Further analysis reveals that EVOR benefits from the synchronous evolution of queries and documents and the diverse information sources in the knowledge base. We hope that our studies will inspire more insights into the design of advanced RACG pipelines in future research. Our model, code, and data are available at https://arks-codegen.github.io.

cs.CL↗
arXiv2024

Receiver operating characteristic curve analysis with non-ignorable missing disease status

This article considers the receiver operating characteristic (ROC) curve analysis for medical data with non-ignorable missingness in the disease status. In the framework of the logistic regression models for both the disease status and the verification status, we first establish the identifiability of model parameters, and then propose a likelihood method to estimate the model parameters, the ROC curve, and the area under the ROC curve (AUC) for the biomarker. The asymptotic distributions of these estimators are established. Via extensive simulation studies, we compare our method with competing methods in the point estimation and assess the accuracy of confidence interval estimation under various scenarios. To illustrate the application of the proposed method in practical data, we apply our method to the National Alzheimer's Coordinating Center data set.

stat.ME↗
arXiv2024

Local interaction simulation approach for the acoustic wave equation with perfectly matched layer

Simulation of the acoustic wave equation plays an important role in various applications, including audio engineering, medical imaging, and fluid dynamics. However, the complexity of the propagation medium can pose challenges, such as the infinite computing region and the interface conditions between different media. In this paper, we construct a method for simulating acoustic wave propagation based on the local interaction simulation approach (LISA) and the perfectly matched layer (PML). This method can simulate wave propagation in a finite computing region and overcome the smoothing process at the interface between different media. Numerical examples demonstrate the effectiveness of this approach.

math.NA↗
arXiv2024

IREE Oriented Active RIS-Assisted Green communication System with Outdated CSI

The rapid evolution of communication technologies has spurred a growing demand for energy-efficient network architectures and performance metrics. Active Reconfigurable Intelligent Surfaces (RIS) are emerging as a key component in green network architectures. Compared to passive RIS, active RIS are equipped with amplifiers on each reflecting element, allowing them to simultaneously reflect and amplify signals, thereby overcoming the double multiplicative fading in the phase response, and improving both system coverage and performance. Additionally, the Integrated Relative Energy Efficiency (IREE) metric, as introduced in [1], addresses the dynamic variations in traffic and capacity over time and space, enabling more energy-efficient wireless systems. Building on these advancements, this paper investigates the problem of maximizing IREE in active RIS-assisted green communication systems. However, acquiring perfect Channel State Information (CSI) in practical systems poses significant challenges and costs. To address this, we derive the average achievable rate based on outdated CSI and formulated the corresponding IREE maximization problem, which is solved by jointly optimizing beamforming at both the base station and RIS. Given the non-convex nature of the problem, we propose an Alternating Optimization Successive Approximation (AOSO) algorithm. By applying quadratic transform and relaxation techniques, we simplify the original problem and alternately optimize the beamforming matrices at the base station and RIS. Furthermore, to handle the discrete constraints of the RIS reflection coefficients, we develop a successive approximation method. Experimental results validate our theoretical analysis of the algorithm's convergence , demonstrating the effectiveness of the proposed algorithm and highlighting the superiority of IREE in enhancing the performance of green communication networks.

eess.SP↗
arXiv2024

VLKEB: A Large Vision-Language Model Knowledge Editing Benchmark

Recently, knowledge editing on large language models (LLMs) has received considerable attention. Compared to this, editing Large Vision-Language Models (LVLMs) faces extra challenges from diverse data modalities and complicated model components, and data for LVLMs editing are limited. The existing LVLM editing benchmark, which comprises three metrics (Reliability, Locality, and Generality), falls short in the quality of synthesized evaluation images and cannot assess whether models apply edited knowledge in relevant content. Therefore, we employ more reliable data collection methods to construct a new Large $\textbf{V}$ision-$\textbf{L}$anguage Model $\textbf{K}$nowledge $\textbf{E}$diting $\textbf{B}$enchmark, $\textbf{VLKEB}$, and extend the Portability metric for more comprehensive evaluation. Leveraging a multi-modal knowledge graph, our image data are bound with knowledge entities. This can be further used to extract entity-related knowledge, which constitutes the base of editing data. We conduct experiments of different editing methods on five LVLMs, and thoroughly analyze how do they impact the models. The results reveal strengths and deficiencies of these methods and hopefully provide insights for future research. The codes and dataset are available at: https://github.com/VLKEB/VLKEB.

cs.CL↗
arXiv2024

Internet of Federated Digital Twins (IoFDT): Connecting Twins Beyond Borders for Society 5.0

The concept of digital twin (DT), which enables the creation of a programmable, digital representation of physical systems, is expected to revolutionize future industries and will lie at the heart of the vision of a future smart society, namely, Society 5.0, in which high integration between cyber (digital) and physical spaces is exploited to bring economic and societal advancements. However, the success of such a DT-driven Society 5.0 requires a synergistic convergence of artificial intelligence and networking technologies into an integrated, programmable system that can coordinate DT networks to effectively deliver diverse Society 5.0 services. Prior works remain restricted to either qualitative study, simple analysis or software implementations of a single DT, and thus, they cannot provide the highly synergistic integration of digital and physical spaces as required by Society 5.0. In contrast, this paper envisions a novel concept of an Internet of Federated Digital Twins (IoFDT) that holistically integrates heterogeneous and physically separated DTs representing different Society 5.0 services within a single framework and system. For this concept of IoFDT, we first introduce a hierarchical architecture that integrates federated DTs through horizontal and vertical interactions, bridging cyber and physical spaces to unlock new possibilities. Then, we discuss challenges of realizing IoFDT, highlighting the intricacies across communication, computing, and AI-native networks while also underscoring potential innovative solutions. Subsequently, we elaborate on the importance of the implementation of a unified IoFDT platform that integrates all technical components and orchestrates their interactions, emphasizing the necessity of practical experimental platforms with a focus on real-world applications in areas like smart mobility.

cs.AI↗