arXiv subjects

Yajun Wang

Publications and source records attributed to Yajun Wang.

At least 19 recordsLinked to original sources

arXiv2026

LLM-based User-Product Relevance Evaluation for Dynamic Product Advertising

Dynamic Product Advertising (DPA) retrieval systems are a core component of modern e-commerce advertising; however, their evaluation remains challenging. Conventional offline engagement metrics are limited by sparse, biased feedback, while online evaluation is costly and time-consuming. In this work, we propose a user--product relevance metric for offline evaluation of DPA retrieval systems. The metric first compresses the engagement history into a compact set of intent representatives via Semantic-ID-grouped, event-stratified segmentation, then assesses the relevance of a recommended product to each individual intent and aggregates them as a user-level relevance score, yielding a stable and interpretable measure of retrieval quality. To compute this metric at scale, we leverage large language models (LLMs) as semantic relevance judges, capturing nuanced functional and semantic relationships that heuristic methods often miss. We propose an agentic, human-aligned prompting framework that generates multi-category relevance guidelines and distills teacher-model judgments into an efficient student model for scalable evaluation. The resulting metric strongly correlates with offline and online business metrics, enabling faster and more reliable DPA retrieval iteration before A/B testing.

cs.IR↗

arXiv2026

Seek Before You Move: Evidence Seeking for Progress Grounding in Vision-Language Navigation

Vision-Language Navigation (VLN) requires agents to continuously ground task progress from long-horizon instructions and partial egocentric observations. Existing VLM-based navigation agents typically reason only over available observations and may remain confident even when task-relevant evidence is missing. For example, an agent may confidently proceed forward and get lost even though the landmark indicating the next turn lies outside its current field of view. We term this failure mode Progress Myopia: the agent fails to recognize unreliable progress grounding and continues acting on insufficient evidence. To address it, we propose SeekVLN, an evidence-seeking framework that couples semantic progress reasoning with active acquisition of task-relevant observations. SeekVLN is trained in two stages: First, Future-guided Reverse Generation (FRG) uses future expert actions to augment offline expert trajectories with supplementary views and evidence annotations. Supervised fine-tuning on these trajectories establishes a prior for evidence seeking and progress reasoning without additional expert interaction. However, imitation alone does not reveal whether seeking improves subsequent navigation. We therefore introduce Counterfactual Contrastive Policy Optimization (C2PO) for reinforcement fine-tuning. By comparing each evidence-seeking branch with a counterfactual direct-navigation branch from the same state, C2PO uses a contrastive reward to assign credit to seeking decisions based on subsequent navigation benefit. Experiments on simulated benchmarks show that SeekVLN achieves state-of-the-art performance, improving success rate by 12.7% and 7.5% over the base model on R2R-CE and RxR-CE, respectively. Both simulated and real-world evaluations exhibit human-like evidence-seeking behaviors for more reliable progress grounding.

cs.AI↗

arXiv2026

SetMIR: Multi-Interest Retrieval as Set Prediction

Embedding-based retrieval is at the core of industrial recommender systems, but a single user embedding is often too limited to capture a user's diverse interests. Multi-interest retrieval addresses this by using multiple user embeddings, yet existing methods still suffer from two issues: interest collapse, where different embeddings learn the same interest, and static dispatch, where serving uses a fixed retrieval budget even when some embeddings are unnecessary. We propose SetMIR, which treats multi-interest retrieval as a set prediction problem. SetMIR encodes a user's behavior history with a transformer and uses K learnable queries to decode a set of user interests, each producing a retrieval embedding and a presence score. During training, Hungarian matching assigns targets to queries one-to-one, so matched queries learn distinct interests and the presence head learns which queries are active. At serving time, SetMIR uses presence scores and query-level Non-Maximum Suppression (NMS) to issue only active, non-redundant ANN queries. On Snap's Dynamic Product Ads (DPA) data, SetMIR outperforms four learned multi-interest retrievers on every metric while issuing 33% fewer ANN queries per request. Deployed as a new retrieval source in the DPA production stack, SetMIR lifts overall CVR by 3.1%, while lifting CTR by 44% and CVR by 51% over the item-to-item retrieval source with the same item embeddings, ANN index, and retrieval quota.

cs.IR↗

arXiv2026

CAMIE: Co-Engagement-Aware Multimodal Item Embeddings for Snap Dynamic Product Ads Retrieval

Item-to-item (I2I) retrieval is a core primitive in large-scale recommendation and advertising systems. In production Snap Dynamic Product Ads (DPA), I2I retrieval faces two challenges: separate visual, textual, and multimodal encoders fragment the retrieval stack, and content-only training does not align embeddings with the co-engagement behavior that drives downstream conversions. We present CAMIE, a co-engagement-aware multimodal item embedding framework for Snap DPA retrieval. CAMIE builds on LLM/MLLM backbones, using their native multimodal interfaces to represent item images and metadata in a shared embedding space. It then fine-tunes the backbone on co-engaged item pairs mined from user journeys with a symmetric in-batch InfoNCE objective. Offline, CAMIE outperforms the strongest commercial multimodal embedding model on Recall@10 and serves text-only retrieval from the same checkpoint with minimal quality loss. Online, CAMIE serves as a drop-in replacement for two deployed content-based I2I encoders, delivering +0.390% CTR / +10.832% CVR over the multimodal control, +18.958% CTR / +13.12% CVR over the text control, and +0.211% CTR / +1.911% CVR on overall DPA traffic. CAMIE is deployed in production.

cs.IR↗

arXiv2026

EGR: Embedding-Native Generative Retrieval with a Shared LLM

Generative retrieval is increasingly popular in large-scale recommendation and advertising systems, yet current methods introduce practical complications. Semantic-ID methods rely on quantization, mutable identifier vocabularies, and token-to-item grounding; embedding-based pipelines train the item encoder separately from the query generator, which limits user-item alignment. We propose EGR, an Embedding-Native Generative Retrieval framework for recommendation and advertising. EGR uses a single shared LLM to learn item representations from item metadata and user representations from interaction histories in one embedding space. Items are indexed directly as dense vectors, and user histories are encoded as dense retrieval queries. Joint contrastive training groups related items and aligns queries with their target items. We evaluate EGR on public benchmarks, industrial data, and live deployment. EGR outperforms published baselines on Amazon Reviews; on Snap DPA, it scales with data, handles cold-start items, and benefits from multimodal input. In production, EGR delivers a +2.91% conversion-rate lift, simplifying system design while improving retrieval quality and ad performance.

cs.IR↗

arXiv2026

SMART: LLM-Augmented Hybrid Retrieval for Dynamic Product Ads

Dynamic Product Ads (DPA) require retrieving relevant items from multi-million product catalogs, balancing two competing objectives: retargeting (re-surfacing known interests) and prospecting (discovering new categories). While Large Language Models (LLMs) capture semantic intent better than traditional embedding models, deploying them at scale introduces prohibitive inference costs and lexical mismatch issues. Through controlled experiments on millions of users, we demonstrate a critical retrieval decomposition: rule-generated queries excel at retargeting on a lexical BM25 index, while LLM-generated queries excel at prospecting on a dense ANN index. Building on this, we propose SMART (SeMantic-aware Adaptive ReTrieval). To manage costs, a lightweight quality gate identifies coverage gaps in initial keyword results, adaptively routing only the ~10% of users who benefit from semantic prospecting to the LLM path. Offline evaluation demonstrates that this gated approach captures the bulk of semantic prospecting gains in Relevance Score while maintaining competitive re-targeting performance at a 90% reduction in LLM costs. Finally, in a 2-week online A/B test at Snap, SMART improved the ad conversion rate by +27.6% over a strong embedding-based baseline.

cs.IR↗

arXiv2026

Multiphysics embedding localized orthogonal decomposition for thermomechanical coupling problems

Multiscale thermomechanical problems in highly heterogeneous media are challenging because the elastic, thermal, and coupling coefficients may vary on unresolved spatial scales. We propose a multiphysics-embedding localized orthogonal decomposition (ME-LOD) method in which displacement and temperature correctors are generated by a coupled static operator. The corrector problems are localized to coarse-grid patches and solved in the kernel of a projective quasi-interpolation operator. We prove uniform inf-sup stability on the global fine-scale kernel and on all zero-extension patch kernels, establish exponential decay of the coupled correctors and the resulting multiscale basis functions, and derive spatial approximation and fully discrete reduction estimates. Numerical experiments demonstrate that, for the tested periodic, random, and high-contrast coefficient fields, ME-LOD attains smaller errors than the comparison method at the same coarse resolution and patch size and can reach a prescribed accuracy with fewer oversampling layers. Although each coupled local corrector is more expensive than a decoupled corrector, the improved localization yields a favorable overall accuracy-to-cost balance in the reported tests.

math.NA↗

arXiv2026

Topological spectral gap, multiscale Weyl's law, and homogenization in high-contrast PDEs

This paper introduces a unified abstract variational and topological framework to characterize the spectral gap and eigenvalue distribution in high-contrast multiscale partial differential equations (PDEs). We rigorously prove that the exact location of the spectral gap is universally determined by the dimension of the local null space associated with the high-contrast inclusions. For systems with infinite-dimensional kernels, this location is strictly determined by the topological Betti numbers. Furthermore, we establish a multiscale Weyl's law via a spectral decoupling theorem, demonstrating that as the contrast approaches infinity, the multiscale spectrum bifurcates into two independent components: the Dirichlet spectrum of the background matrix and the internal Neumann spectrum of the inclusions. Using spectral homogenization theory, we also show that in the limit of vanishing periodicity, this expanding topological eigenspace asymptotically spans the entire spectral space of the macroscopic homogenized operator. These theoretical results are comprehensively verified through numerical experiments on diffusion, elasticity, fourth-order plate, Maxwell, and grad-div operators.

math.NA↗

arXiv2026

Advancing Graph Few-Shot Learning via In-Context Learning

Graph few-shot learning, which aims to classify nodes from novel classes with only a few labeled examples, is a widely studied problem in graph learning. However, existing methods often face two key limitations. First, the predominant graph few-shot learning paradigm relies on supervised tasks, failing to leverage the vast number of unlabeled nodes in the graph. Second, many approaches require complex task adaptation or fine-tuning during inference, limiting their efficiency and applicability. Inspired by the powerful in-context learning capabilities of large language models, we propose a novel model named VISION for adVancIng graph few-Shot learning via In-cOntext LearNing to address these challenges. Our model reframes graph few-shot learning as a fine-tuning-free sequence reasoning problem. At its core is a context-aware network that initializes nodes with role embeddings and employs a dual-context fusion module to synergistically integrate local topological structures and global task-level dependencies. This allows our model to dynamically generate class-aware representations for the query set conditioned on the support set context in a single forward pass. To effectively train our model, we introduce an unsupervised task generator that creates structure-adaptive features and constructs diverse pseudo-tasks from abundant unlabeled data. Our method unifies unsupervised meta-learning with graph in-context learning, achieving efficient inference. Extensive experiments on multiple benchmark datasets demonstrate the superiority of our model. Our public code can be found

cs.AI↗

arXiv2026

Decoupling scales via localized subspace iteration and temporal splitting for multiscale parabolic equations

Simulating diffusion in heterogeneous media presents a significant computational challenge, as resolving microscopic physical scales traditionally demands excessively fine computational grids. To overcome this barrier, we extend the Localized Subspace Iteration (LSI) framework to multiscale parabolic equations. The proposed method constructs optimal, low-dimensional trial spaces by iteratively approximating the dominant eigenspaces of local inverse operators via Localized Standard Subspace Iteration (LSSI) or Localized Krylov Subspace Iteration (LKSI). Because these LSI basis functions are inherently tailored to capture the slow-decaying, low-frequency modes of the parabolic solution, they naturally suppress error accumulation over long-term integration. To further improve computational efficiency, we decouple the basis construction into an offline phase and implement a contrast-independent, partially explicit temporal splitting scheme for online time-stepping. By explicitly advancing the dominant macroscopic modes while implicitly treating high-frequency microscopic corrections, this scheme guarantees stability without imposing restrictive time-step constraints. We establish rigorous a priori error estimates in both the energy and $L^2$ norms. Numerical experiments illustrate the accuracy and efficiency of the LSI framework, particularly highlighting the LKSI method's advantages in handling high-contrast, complex multiscale media.

math.NA↗

arXiv2025

Green Emergency Communications in RIS- and MA-Assisted Multi-UAV SAGINs: A Partially Observable Reinforcement Learning Approach

In post-disaster space-air-ground integrated networks (SAGINs), terrestrial infrastructure is often impaired, and unmanned aerial vehicles (UAVs) must rapidly restore connectivity for mission-critical ground terminals in cluttered non-line-of-sight (NLoS) urban environments. To enhance coverage, UAVs employ movable antennas (MAs), while reconfigurable intelligent surfaces (RISs) on surviving high-rises redirect signals. The key challenge is communication-limited partial observability, leaving each UAV with a narrow, fast-changing neighborhood view that destabilizes value estimation. Existing multi-agent reinforcement learning (MARL) approaches are inadequate--non-communication methods rely on unavailable global critics, heuristic sharing is brittle and redundant, and learnable protocols (e.g., CommNet, DIAL) lose per-neighbor structure and aggravate non-stationarity under tight bandwidth. To address partial observability, we propose a spatiotemporal A2C where each UAV transmits prior-decision messages with local state, a compact policy fingerprint, and a recurrent belief, encoded per neighbor and concatenated. A spatial discount shapes value targets to emphasize local interactions, while analysis under one-hop-per-slot latency explains stable training with delayed views. Experimental results show our policy outperforms IA2C, ConseNet, FPrint, DIAL, and CommNet--achieving faster convergence, higher asymptotic reward, reduced Temporal-Difference(TD)/advantage errors, and a better communication throughput-energy trade-off.

eess.SY↗

arXiv2025

Efficient Spectral Efficiency Maximization Design for IRS-aided MIMO Systems

Driven by the growing demand for higher spectral efficiency in wireless communications, intelligent reflecting surfaces (IRS) have attracted considerable attention for their ability to dynamically reconfigure the propagation environment. This work addresses the spectral efficiency maximization problem in IRS-assisted multiple-input multiple-output (MIMO) systems, which involves the joint optimization of the transmit precoding matrix and the IRS phase shift configuration. This problem is inherently challenging due to its non-convex nature. To tackle it effectively, we introduce a computationally efficient algorithm, termed ADMM-APG, which integrates the alternating direction method of multipliers (ADMM) with the accelerated projected gradient (APG) method. The proposed framework decomposes the original problem into tractable subproblems, each admitting a closed-form solution while maintaining low computational complexity. Simulation results demonstrate that the ADMM-APG algorithm consistently surpasses existing benchmark methods in terms of spectral efficiency and computational complexity, achieving significant performance gains across a range of system configurations.

cs.IT↗

arXiv2025

Graph Few-Shot Learning via Adaptive Spectrum Experts and Cross-Set Distribution Calibration

Graph few-shot learning has attracted increasing attention due to its ability to rapidly adapt models to new tasks with only limited labeled nodes. Despite the remarkable progress made by existing graph few-shot learning methods, several key limitations remain. First, most current approaches rely on predefined and unified graph filters (e.g., low-pass or high-pass filters) to globally enhance or suppress node frequency signals. Such fixed spectral operations fail to account for the heterogeneity of local topological structures inherent in real-world graphs. Moreover, these methods often assume that the support and query sets are drawn from the same distribution. However, under few-shot conditions, the limited labeled data in the support set may not sufficiently capture the complex distribution of the query set, leading to suboptimal generalization. To address these challenges, we propose GRACE, a novel Graph few-shot leaRning framework that integrates Adaptive spectrum experts with Cross-sEt distribution calibration techniques. Theoretically, the proposed approach enhances model generalization by adapting to both local structural variations and cross-set distribution calibration. Empirically, GRACE consistently outperforms state-of-the-art baselines across a wide range of experimental settings. Our code can be found here.

cs.LG↗

arXiv2025

Continuous variable quantum communication with 40 pairs of entangled sideband

Constructing large-scale quantum resources is an important foundation for further improving the efficiency and scalability of quantum communication. Here, we present an efficient extraction and stable control scheme of 40 pairs of entangled sideband modes from the squeezed light by specially designing optical parametric oscillator. Utilizing the low-loss optical frequency comb control technology and the local cross-correlation algorithm, we model and manage the efficient separation process of the entangled sidebands modes facilitated by the optical filtering cavities, a maximum entanglement level of 6.5 dB is achieved. The feasibility of large-capacity quantum dense coding based on these entangled sideband modes is proved experimentally, which is of great significance for optimizing the utilization of quantum resources, thereby contributing to the advancement of large-capacity quantum communication networks and enabling the realization of more secure and efficient quantum communication systems.

quant-ph↗

arXiv2025

Reservoir-engineered squeezed lasing through the parametric coupling

We report the first experimental demonstration of squeezed lasing in a reservoir-engineered optical parametric oscillator (OPO). The OPO provides a basis of squeezed states and parametric amplification in lasing emission, whose vacuum reservoir is coupled to a squeezed vacuum generated by a second OPO. With a precisely controlled squeezing angle and strong squeezing injection, the parametric interaction in the first OPO is exponentially enhanced. It successfully circumvents the decoherence in the system, and eliminates the undesired noise of spontaneous photon emission in the OPO. As a result, the amplified parametric process simultaneously reserves the coherence and quantum properties in the first OPO, and yields a -6.1 dB squeezed laser in optical domain with a narrow linewidth and high brightness. Our work sheds light on potential applications of squeezed lasing in quantum metrology and quantum optics.

quant-ph↗

arXiv2025

Quantum-enhanced laser phase noise filter

Quantum noise is the fundamental limit of laser phase noise filter. We cannot realize the effective quantum-enhanced phase noise suppression through simply utilizing amplitude noise suppression scheme. Here, we present the first experimental demonstration of a quantum-enhanced laser phase noise filter, achieved by employing a noise ellipse rotation phase noise readout technique combined with an excess amplitude noise suppression scheme. We address the primary limitations in the extracting of laser phase noise, and make the quantum enhancement via squeezed vacuum injection feasible. A maximum of 5 dB quantum-enhanced phase noise suppression is realized across the Fourier frequencies from 5 kHz to 60 kHz. The demonstration unlocks the application of squeezed vacuum state in laser phase noise suppression.

quant-ph↗

arXiv2025

Bright squeezed light in the kilohertz frequency band

The largely technical noise of a free running laser is the fundamental limit for preparation of a bright squeezed light, especially within MHz band. We construct a universal and complete theoretical model for nonclassical stabilization, and propose a novel bright squeezed light generation scheme with assistance of a hybrid noise stabilization technology. The scheme assimilates a broadband passive noise stabilization to successfully extend the feedback frequency bandwidth of a nonclassical active stabilization to MHz band. Meanwhile, the technical noise suppression magnitude is further improved by 9 dB, which creates the prerequisite conditions for preparing a broadband bright squeezed light. Finally, a -5.5 dB bright squeezed light with a power of 1 mW and a squeezing bandwidth among kHz to MHz band was generated. The experimental results agree well with the theoretical one. To the best of our knowledge, it is the first demonstration of a milliwatt-order bright squeezed light across the kHz frequency band. The demonstration sheds light on potential applications of bright squeezed light in quantum metrology and might enable new concepts in the future.

quant-ph↗

arXiv2025

Laser intensity noise suppression for space-borne gravitational wave mission

Laser intensity noise is a main limitation of measurement and sensing mission represented by gravitational wave detection. We develop a noise decomposition model and design the core elements of the feedback loop independently based on the analysis results. We construct a fiber amplifier system with ultra-low intensity noise in the 0.1 mHz-1 Hz frequency band by the employment of an optoelectronic feedback loop that is specially designed. The study provides experimental basis and technologies for precise measurement and sensing system at ultra-low frequency.

physics.optics↗