arXiv subjects

Kai Yuan

Publications and source records attributed to Kai Yuan.

At least 19 recordsLinked to original sources

arXiv2026

From Preferences to Principles: Rubric-Based Alignment for Grounded Knowledge Answers

Designing effective reward signals for open-domain question answering is challenging because high-quality responses must simultaneously satisfy multiple aspects of answer quality that are difficult to capture with a holistic scalar objective. We introduce a rubric-based reward framework that generates query-specific rubrics grounded in retrieved evidence and decomposed into multiple quality dimensions, providing fine-grained supervision during post-training. Averaged across three evaluation axes (composition, grounding, and instruction-following), our approach improves over the instruction-tuned baseline by 6.5% and over flat rubric variants by 4%, with consistent gains across all evaluation datasets. Conditioning rubrics on retrieved evidence improves factual support, while decomposing rubrics into quality-specific dimensions further improves coherence, organization, and adherence to query requirements. Our results show that grounded, multi-dimensional rubrics provide more effective reward supervision for complex open-domain question answering.

cs.CL↗

arXiv2026

Adaptive Model Compression (AMC): Saliency-Driven Resource Allocation for Ultra-Low-Power Transformer Inference

Deploying large-scale transformer models on resource-constrained edge devices remains a challenge due to the high energy and memory overhead inherent in static inference, which processes simple and complex tokens with uniform intensity. To address this, we propose Adaptive Model Compression (AMC), a saliency-driven framework that dynamically allocates hardware resources based on token importance. By implementing a multi-tier architecture, our system identifies critical high-saliency information for full-precision processing while aggressively reducing the rank and bit-width of less significant data. Experimental results demonstrate that AMC achieves a 59.2% reduction in system energy and a 2.24x increase in throughput on 45nm CMOS hardware. This approach effectively extends the battery life of mobile devices by utilizing high-definition compute only where necessary, maintaining robust performance with a marginal 3.6% accuracy trade-off.

cs.IR↗

arXiv2026

Hamilton Cycles in Semisymmetric Graphs

In light of Lov\'{a}sz's longstanding question on the existence of Hamilton paths in vertex-transitive graphs, this paper considers a natural variant: what if vertex-transitivity is relaxed, yet a high degree of symmetry--specifically edge-transitivity--is retained? To investigate this, we focus on the class of semisymmetric graphs, which are regular, edge-transitive, but not vertex-transitive. In this paper, it will be shown that every connected semisymmetric graph of order $2pq$, where $p$ and $q$ are two distinct primes contains a Hamilton cycle and that every connected cubic semisymmetric graph of order less than 3000 contains a Hamilton cycle too. Based on these observations, the following question is posed: construct a connected semisymmetric graph which has no Hamilton cycle.

math.CO↗

arXiv2026

Unifying Ranking and Generation in Query Auto-Completion via Retrieval-Augmented Generation and Multi-Objective Alignment

Query Auto-Completion (QAC) suggests query completions as users type, helping them articulate intent and reach results more efficiently. Existing approaches face fundamental challenges: traditional retrieve-and-rank pipelines have limited long-tail coverage and require extensive feature engineering, while recent generative methods suffer from hallucination and safety risks. We present a unified framework that reformulates QAC as end-to-end list generation through Retrieval-Augmented Generation (RAG) and multi-objective Direct Preference Optimization (DPO). Our approach combines three key innovations: (1) reformulating QAC as end-to-end list generation with multi-objective optimization; (2) defining and deploying a suite of rule-based, model-based, and LLM-as-judge verifiers for QAC, and using them in a comprehensive methodology that combines RAG, multi-objective DPO, and iterative critique-revision for high-quality synthetic data; (3) a hybrid serving architecture enabling efficient production deployment under strict latency constraints. Evaluation on a large-scale commercial search platform demonstrates substantial improvements: offline metrics show gains across all dimensions, human evaluation yields +0.40 to +0.69 preference scores, and a controlled online experiment achieves 5.44\% reduction in keystrokes and 3.46\% increase in suggestion adoption, validating that unified generation with RAG and multi-objective alignment provides an effective solution for production QAC. This work represents a paradigm shift to end-to-end generation powered by large language models, RAG, and multi-objective alignment, establishing a production-validated framework that can benefit the broader search and recommendation industry.

cs.IR↗

arXiv2026

Best-of-Q: Improving VLM agents with Q-function Action Ranking at Inference

Vision-Language Models (VLMs) have become powerful backbones for agents to autonomously operate in digital environments like the web and operating systems. However, these models suffer from inadaptability to fast-changing environments like the web, which can be alleviated by fine-tuning requiring expansive model training and data collection. In this work, we introduce a novel paradigm for enhancing agentic VLM policies at inference without policy retraining. Fundamentally, our approach decouples the VLM's role as a high-capacity action proposer from the final action selection mechanism. We keep the VLM policy frozen and use it to generate a set of candidate actions for a given state. Then, a lightweight, offline-trained Q-function reranks these candidates, and the agent executes the action with the highest estimated value. The main contribution is to apply the Q-function directly during inference for immediate policy improvement, and not offline to relabel data for policy retraining. We demonstrate on the academic WebVoyager benchmark that our method significantly boosts agent success rates, improving a Qwen2.5-VL-7B agent from 38.8% to 55.7% and a proprietary GPT-4.1 agent from 82.4% to 88.8%.

cs.AI↗

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

Broadband, robust, and tunable beam splitter based on topological unidirectional surface magnetoplasmons

Beam splitters are pivotal components in integrated microwave and photonic systems. However, conventional designs based on directional coupling or multi-mode interference often suffer from back scattering, frequency-dependent splitting ratios, and limited bandwidth. To overcome these limitations, here, we propose a new physical mechanism to achieve a broadband, robust, and tunable beam splitter by manipulating the mode coupling of the topological unidirectional surface magnetoplasmons (USMP) at the input and output waveguides. We show that the beam splitter not only exhibits strong robustness against obstacles but also achieves a broad bandwidth across nearly the entire USMP band with arbitrarily tunable and frequency-independent splitting ratios. Moreover, the operating band of the beam splitter can be actively tuned by adjusting the external magnetic field, while its robust and broadband characteristics are retained. Our results extend the research frontier of beam splitters and may have potential applications in integrated photonic devices and modern communication systems.

physics.optics↗

arXiv2025

Adaptive Source-Channel Coding for Multi-User Semantic and Data Communications

This paper considers a multi-user semantic and data communication (MU-SemDaCom) system, where a base station (BS) simultaneously serves users with different semantic and data tasks through a downlink multi-user multiple-input single-output (MU-MISO) channel. The coexistence of heterogeneous communication tasks, diverse channel conditions, and the requirements for digital compatibility poses significant challenges to the efficient design of MU-SemDaCom systems. To address these issues, we propose a multi-user adaptive source-channel coding (MU-ASCC) framework that adaptively optimizes deep neural network (DNN)-based source coding, digital channel coding, and superposition broadcasting. First, we employ a data-regression method to approximate the end-to-end (E2E) semantic and data distortions, for which no closed-form expressions exist. The obtained logistic formulas decompose the E2E distortion as the addition of the source and channel distortion terms, in which the logistic parameter variations are task-dependent and jointly determined by both the DNN and channel parameters. Then, based on the derived formulas, we formulate a weighted-sum E2E distortion minimization problem that jointly optimizes the source-channel coding rates, power allocation, and beamforming vectors for both the data and semantic users. Finally, an alternating optimization (AO) framework is developed, where the adaptive rate optimization is solved using the subgradient descent method, while the joint power and beamforming is addressed via the uplink-downlink duality (UDD) technique. Simulation results demonstrate that, compared with the conventional separate source-channel coding (SSCC) and deep joint source-channel coding (DJSCC) schemes that are designed for a single task, the proposed MU-ASCC scheme achieves simultaneous improvements in both the data recovery and semantic task performance.

eess.IV↗

arXiv2025

Monodromy representation of graphs

It is well-known that every vertex-transitive graph admits a representation as a coset graph. In this paper, we extend this construction by introducing monodromy graphs defined through double cosets. Our main result establishes that every graph is isomorphic to a monodromy graph, providing a new combinatorial framework for graph representation. Moreover, we show that every graph gives rise to an arc-transitive graph through its monodromy representation. Inspired by the monodromy representation of graphs, we denote an algebraic map $\mathcal{M}(G;Ω,ρ,τ)$ by $\mathcal{M}(G;U,ρ,τ)$ where $U$ is a stabiliser in $G$. As an application, we prove an enumeration theorem for orientable maps with a given monodromy group. We underscore a fundamental triad in algebraic graph theory: Where there is a graph, there is a group, an arc-transitive graph, and an orientable regular map--each arising canonically from the underlying combinatorial and algebraic structures.

math.CO↗

arXiv2025

Video-Language Critic: Transferable Reward Functions for Language-Conditioned Robotics

Natural language is often the easiest and most convenient modality for humans to specify tasks for robots. However, learning to ground language to behavior typically requires impractical amounts of diverse, language-annotated demonstrations collected on each target robot. In this work, we aim to separate the problem of what to accomplish from how to accomplish it, as the former can benefit from substantial amounts of external observation-only data, and only the latter depends on a specific robot embodiment. To this end, we propose Video-Language Critic, a reward model that can be trained on readily available cross-embodiment data using contrastive learning and a temporal ranking objective, and use it to score behavior traces from a separate actor. When trained on Open X-Embodiment data, our reward model enables 2x more sample-efficient policy training on Meta-World tasks than a sparse reward only, despite a significant domain gap. Using in-domain data but in a challenging task generalization setting on Meta-World, we further demonstrate more sample-efficient training than is possible with prior language-conditioned reward models that are either trained with binary classification, use static images, or do not leverage the temporal information present in video data.

cs.RO↗

arXiv2025

Unifi3D: A Study on 3D Representations for Generation and Reconstruction in a Common Framework

Following rapid advancements in text and image generation, research has increasingly shifted towards 3D generation. Unlike the well-established pixel-based representation in images, 3D representations remain diverse and fragmented, encompassing a wide variety of approaches such as voxel grids, neural radiance fields, signed distance functions, point clouds, or octrees, each offering distinct advantages and limitations. In this work, we present a unified evaluation framework designed to assess the performance of 3D representations in reconstruction and generation. We compare these representations based on multiple criteria: quality, computational efficiency, and generalization performance. Beyond standard model benchmarking, our experiments aim to derive best practices over all steps involved in the 3D generation pipeline, including preprocessing, mesh reconstruction, compression with autoencoders, and generation. Our findings highlight that reconstruction errors significantly impact overall performance, underscoring the need to evaluate generation and reconstruction jointly. We provide insights that can inform the selection of suitable 3D models for various applications, facilitating the development of more robust and application-specific solutions in 3D generation. The code for our framework is available at https://github.com/isl-org/unifi3d.

cs.GR↗

arXiv2025

Adaptive Source-Channel Coding for Semantic Communications

Semantic communications (SemComs) have emerged as a promising paradigm for joint data and task-oriented transmissions, combining the demands for both the bit-accurate delivery and end-to-end (E2E) distortion minimization. However, current joint source-channel coding (JSCC) in SemComs is not compatible with the existing communication systems and cannot adapt to the variations of the sources or the channels, while separate source-channel coding (SSCC) is suboptimal in the finite blocklength regime. To address these issues, we propose an adaptive source-channel coding (ASCC) scheme for SemComs over parallel Gaussian channels, where the deep neural network (DNN)-based semantic source coding and conventional digital channel coding are separately deployed and adaptively designed. To enable efficient adaptation between the source and channel coding, we first approximate the E2E data and semantic distortions as functions of source coding rate and bit error ratio (BER) via logistic regression, where BER is further modeled as functions of signal-to-noise ratio (SNR) and channel coding rate. Then, we formulate the weighted sum E2E distortion minimization problem for joint source-channel coding rate and power allocation over parallel channels, which is solved by the successive convex approximation. Finally, simulation results demonstrate that the proposed ASCC scheme outperforms typical deep JSCC and SSCC schemes for both the single- and parallel-channel scenarios while maintaining full compatibility with practical digital systems.

cs.IT↗

arXiv2025

AHSG: Adversarial Attack on High-level Semantics in Graph Neural Networks

Adversarial attacks on Graph Neural Networks aim to perturb the performance of the learner by carefully modifying the graph topology and node attributes. Existing methods achieve attack stealthiness by constraining the modification budget and differences in graph properties. However, these methods typically disrupt task-relevant primary semantics directly, which results in low defensibility and detectability of the attack. In this paper, we propose an Adversarial Attack on High-level Semantics for Graph Neural Networks (AHSG), which is a graph structure attack model that ensures the retention of primary semantics. By combining latent representations with shared primary semantics, our model retains detectable attributes and relational patterns of the original graph while leveraging more subtle changes to carry out the attack. Then we use the Projected Gradient Descent algorithm to map the latent representations with attack effects to the adversarial graph. Through experiments on robust graph deep learning models equipped with defense strategies, we demonstrate that AHSG outperforms other state-of-the-art methods in attack effectiveness. Additionally, using Contextual Stochastic Block Models to detect the attacked graph further validates that our method preserves the primary semantics of the graph.

cs.LG↗

arXiv2025

Regular embedding of simple hypergraphs

Regular hypermaps with underlying simple hypergraphs are analysed. We obtain an algorithm to classify the regular embeddings of simple hypergraphs with given order, and determine the automorphism groups of regular embedding of simple hypergraphs with prime square order.

math.CO↗

arXiv2025

Cayley hyper-digraphs and Cayley hypermaps

A Cayley hyper-digraph is a directed hypergraph that its automorphism group contains a subgroup acting regularly on vertices and a Cayley hypermap is a hypermap whose automorphism group contains a subgroup which induces regular action on the hypervertex set. In this paper, we study Cayley hyper-digraphs and construct Cayley hypermaps which have high level of symmetry. Our main goal is to present the general theory so as to make it clear to study Cayley hypermaps.

math.CO↗

arXiv2025

Linear hypermaps--modelling linear hypergraphs on surfaces

A hypergraph is linear if each pair of distinct vertices appears in at most one common edge. We say $\varGamma=(V,E)$ is an associated graph of a linear hypergraph $\mathcal{H}=(V, X)$ if for any $x\in X$, the induced subgraph $\varGamma[x]$ is a cycle, and for any $e\in E$, there exists a unique edge $y\in X$ such that $e\subseteq y$. A linear hypermap $\mathcal{M}$ is a $2$-cell embedding of a connected linear hypergraph $\mathcal{H}$'s associated graph $\varGamma$ on a compact connected surface, such that for any edge $x\in E(\mathcal{H})$, $\varGamma[x]$ is the boundary of a $2$-cell and for any $e\in E(\varGamma)$, $e$ is incident with two distinct $2$-cells. In this paper, we introduce linear hypermaps to model linear hypergraphs on surfaces and regular linear hypermaps modelling configurations on the surfaces. As an application, we classify regular linear hypermaps on the sphere and determine the total number of proper regular linear hypermaps of genus 2 to 101.

math.CO↗

arXiv2025

Generating Physically Realistic and Directable Human Motions from Multi-Modal Inputs

This work focuses on generating realistic, physically-based human behaviors from multi-modal inputs, which may only partially specify the desired motion. For example, the input may come from a VR controller providing arm motion and body velocity, partial key-point animation, computer vision applied to videos, or even higher-level motion goals. This requires a versatile low-level humanoid controller that can handle such sparse, under-specified guidance, seamlessly switch between skills, and recover from failures. Current approaches for learning humanoid controllers from demonstration data capture some of these characteristics, but none achieve them all. To this end, we introduce the Masked Humanoid Controller (MHC), a novel approach that applies multi-objective imitation learning on augmented and selectively masked motion demonstrations. The training methodology results in an MHC that exhibits the key capabilities of catch-up to out-of-sync input commands, combining elements from multiple motion sequences, and completing unspecified parts of motions from sparse multimodal input. We demonstrate these key capabilities for an MHC learned over a dataset of 87 diverse skills and showcase different multi-modal use cases, including integration with planning frameworks to highlight MHC's ability to solve new user-defined tasks without any finetuning.

cs.RO↗

arXiv2024

Robust multimode interference and conversion in topological unidirectional surface magnetoplasmons

We have theoretically investigated surface magnetoplasmons (SMPs) in a yttrium-iron-garnet (YIG) sandwiched waveguide. The dispersion demonstated that this waveguide can support topological unidirectional SMPs. Based on unidirectional SMPs, magnetically controllable multimode interference (MMI) is verified in both symmetric and asymmetric waveguides. Due to the coupling between the modes along two YIG-air interfaces, the asymmetric waveguide supports a unidirectional even mode within a single-mode frequency range. Moreover, these modes are topological protected when disorder is introduced. Utilizing robust unidirectional SMPs MMI (USMMI), tunable splitters have been achieved. It has been demonstrated that mode conversion between different modes can be realized. These results provide many degrees of freedom to manipulate topological waves.

physics.optics↗