arXiv subjects

Qiang Xiao

Publications and source records attributed to Qiang Xiao.

18 recordsLinked to original sources

arXiv2026

Melo: A Production LLM-Powered Music Recommendation Agent

We describe Melo, an LLM-powered music recommendation agent deployed on NetEase Cloud Music. Melo is structured as a deterministic five-node state graph over heterogeneous tools, with a prompt- and state-machine-driven orchestration policy rather than a fine-tuned controller. At industrial scale, the bottleneck is not how smart the brain is but how the system detects and recovers from the mistakes that brain makes. Two production failure modes drove the design: entity hallucination, where the agent commits to interpretations unsupported by the live catalog or user-behavior index, and long-tail degradation, where over-constrained requests collapse to generic popular fallbacks. We address them with two complementary mechanisms. Inference-time entity grounding repurposes the production search index as a verification primitive that gates entity decisions before they propagate downstream. Reflective retry verbalizes failure reasons from a broken tool chain and feeds them into the next planning step, so the system can relax or revise constraints rather than fall back blindly. A one-month online A/B test across NetEase Cloud Music's playlist surfaces reports an over 2 pp lift in a primary playlist retention metric and a lift of over one minute in a core playlist engagement metric. Offline ablation isolates a 7.8 pp reduction in entity misidentification from the three-layer grounding stack on our evaluation set, and a triggered-session analysis on our evaluation set shows reflective retry firing on 5.8% of sessions with 59% process-level recovery. Our deployment experience suggests that progress on LLM-powered music recommendation at this scale depends as much on the named, ablatable runtime machinery that catches and corrects the brain's mistakes as on the brain itself: a hypothesis we offer for the community to test.

cs.IR↗

arXiv2026

PIANO: Personalized Reranking via Information Aggregation Node for Music Search Optimization

Unlike short-video content, music tracks have long lifecycles and lasting value. Effective music search re-ranking must therefore align the user's current query with long-term preferences while jointly optimizing Click-Through Rate (CTR) and Conversion Rate (CVR). However, existing methods suffer from two limitations: (1) sequential methods rely on item-interaction history and therefore cannot use historical search queries to tell which past preferences match the user's current search intent; (2) most listwise models optimize a single objective (e.g., CTR only), and conventional multi-objective methods balance click and conversion at the item level, ignoring how these trade-offs play out across the whole ranked list. To address these limitations, we propose PIANO, a personalized listwise re-ranking framework with two key components: (i) the Query-Driven Interest Refiner (QDIR) uses cross-attention over historical queries to align past intents with the current one; (ii) the Information Aggregation Node (IAN), a learnable [CLS]-style token, aggregates the candidate list and predicts CTR/CVR at the list level. Extensive experiments on public and industrial datasets show consistent gains over strong baselines. In online A/B tests on NetEase Cloud Music, a leading music streaming platform, PIANO achieves statistically significant improvements in CTR (+0.62%) and CVR (+4.45%).

cs.IR↗

arXiv2026

Climber-Pilot: A Non-Myopic Generative Recommendation Model Towards Better Instruction-Following

Generative retrieval has emerged as a promising paradigm in recommender systems, offering superior sequence modeling capabilities over traditional dual-tower architectures. However, in large-scale industrial scenarios, such models often suffer from inherent myopia: due to single-step inference and strict latency constraints, they tend to collapse diverse user intents into locally optimal predictions, failing to capture long-horizon and multi-item consumption patterns. Moreover, real-world retrieval systems must follow explicit retrieval instructions, such as category-level control and policy constraints. Incorporating such instruction-following behavior into generative retrieval remains challenging, as existing conditioning or post-hoc filtering approaches often compromise relevance or efficiency. In this work, we present Climber-Pilot, a unified generative retrieval framework to address both limitations. First, we introduce Time-Aware Multi-Item Prediction (TAMIP), a novel training paradigm designed to mitigate inherent myopia in generative retrieval. By distilling long-horizon, multi-item foresight into model parameters through time-aware masking, TAMIP alleviates locally optimal predictions while preserving efficient single-step inference. Second, to support flexible instruction-following retrieval, we propose Condition-Guided Sparse Attention (CGSA), which incorporates business constraints directly into the generative process via sparse attention, without introducing additional inference steps. Extensive offline experiments and online A/B testing at NetEase Cloud Music, one of the largest music streaming platforms, demonstrate that Climber-Pilot significantly outperforms state-of-the-art baselines, achieving a 4.24\% lift of the core business metric.

cs.IR↗

arXiv2025

FLAME: A Serving System Optimized for Large-Scale Generative Recommendation with Efficiency

Generative recommendation (GR) models possess greater scaling power compared to traditional deep learning recommendation models (DLRMs), yet they also impose a tremendous increase in computational burden. Measured in FLOPs, a typical GR model's workload sits in $10^9 \sim 10^{11}$ range, roughly four orders of magnitude higher than traditional DLRMs. Delivering accurate results in a few tens of milliseconds while processing billions of such requests per day puts extreme demands on the performance of the online serving system. Therefore, for industry practitioners, the alluring gains of GR models are tempered by the formidable challenge of online deployment at scale in production services. In this work, we introduce a comprehensive solution of online serving system tailored For Large-scale GenerAtive RecoMmendation with Efficiency (FLAME). Specifically, we leveraging CPU-GPU heterogeneous hardware to decouple feature pre-processing and model computation. We encapsulated several memory optimization features as the Proximal Data Accelerator (PDA) module to make full use of limited bandwidth and storage resources, which achieves a 1.9x throughput gain and a 1.7x latency reduction. We implement the Fused Kernel Engine (FKE) module based on the functionality and interface of NVIDIA TensorRT to boost model computation, delivering a speedup ratio of 4.6x-6.1x, throughput gain ratio of 4.7x-6.3x one step further. In addition, we design the Dynamic Stream Orchestrator (DSO) module to coordinate concurrent requests, enhancing the system throughput performance with 1.3x improvement in throughput and 2.3x speed-up under non-uniform distribution of upstream candidates. Comprehensive evaluations demonstrate that our FLAME effectively supports large-scale online deployment of GR models and achieves remarkable improvements in system performance.

cs.DC↗

arXiv2025

Climber: Toward Efficient Scaling Laws for Large Recommendation Models

Transformer-based generative models have achieved remarkable success across domains with various scaling law manifestations. However, our extensive experiments reveal persistent challenges when applying Transformer to recommendation systems: (1) Transformer scaling is not ideal with increased computational resources, due to structural incompatibilities with recommendation-specific features such as multi-source data heterogeneity; (2) critical online inference latency constraints (tens of milliseconds) that intensify with longer user behavior sequences and growing computational demands. We propose Climber, an efficient recommendation framework comprising two synergistic components: the model architecture for efficient scaling and the co-designed acceleration techniques. Our proposed model adopts two core innovations: (1) multi-scale sequence extraction that achieves a time complexity reduction by a constant factor, enabling more efficient scaling with sequence length; (2) dynamic temperature modulation adapting attention distributions to the multi-scenario and multi-behavior patterns. Complemented by acceleration techniques, Climber achieves a 5.15$\times$ throughput gain without performance degradation by adopting a "single user, multiple item" batched processing and memory-efficient Key-Value caching. Comprehensive offline experiments on multiple datasets validate that Climber exhibits a more ideal scaling curve. To our knowledge, this is the first publicly documented framework where controlled model scaling drives continuous online metric growth (12.19\% overall lift) without prohibitive resource costs. Climber has been successfully deployed on Netease Cloud Music, one of China's largest music streaming platforms, serving tens of millions of users daily.

cs.IR↗

arXiv2025

Progressive Semantic Residual Quantization for Multimodal-Joint Interest Modeling in Music Recommendation

In music recommendation systems, multimodal interest learning is pivotal, which allows the model to capture nuanced preferences, including textual elements such as lyrics and various musical attributes such as different instruments and melodies. Recently, methods that incorporate multimodal content features through semantic IDs have achieved promising results. However, existing methods suffer from two critical limitations: 1) intra-modal semantic degradation, where residual-based quantization processes gradually decouple discrete IDs from original content semantics, leading to semantic drift; and 2) inter-modal modeling gaps, where traditional fusion strategies either overlook modal-specific details or fail to capture cross-modal correlations, hindering comprehensive user interest modeling. To address these challenges, we propose a novel multimodal recommendation framework with two stages. In the first stage, our Progressive Semantic Residual Quantization (PSRQ) method generates modal-specific and modal-joint semantic IDs by explicitly preserving the prefix semantic feature. In the second stage, to model multimodal interest of users, a Multi-Codebook Cross-Attention (MCCA) network is designed to enable the model to simultaneously capture modal-specific interests and perceive cross-modal correlations. Extensive experiments on multiple real-world datasets demonstrate that our framework outperforms state-of-the-art baselines. This framework has been deployed on one of China's largest music streaming platforms, and online A/B tests confirm significant improvements in commercial metrics, underscoring its practical value for industrial-scale recommendation systems.

cs.IR↗

arXiv2025

Programmable skyrmions for robust communication and intelligent sensing

The recently observed plasmonic skyrmions, as electromagnetic counterparts of topologically stable quasiparticles, hold significant promise as novel carriers for robust information transfer and manipulation of nontrivial light-matter interactions. However, their practical applications has been hindered by the lack of flexible tuning devices to encode these topological structures. Here, we present a programmable plasmonic skyrmion platform capable of encoding diverse skyrmion topologies, including Neel-type skyrmions and merons. Based on unprecedented ultra-fast coding feature, we synthesize harmonic skyrmions in the temporal dimension and, for the first time, applied skyrmions in communication and sensing applications. Specifically, we achieved highly robust and multi-channel wireless communications by using programmable topological skyrmions, providing a promising platform for communication in turbulent noise channels and extreme conditions. Furthermore, we implemented intelligent sensing across twenty animal models on the same platform, achieving high recognition accuracy. This methodology offers programmable and temporal insights into the skyrmions for their practical applications in next-generation wireless communication and intelligent sensing.

physics.optics↗

arXiv2024

LiRank: Industrial Large Scale Ranking Models at LinkedIn

We present LiRank, a large-scale ranking framework at LinkedIn that brings to production state-of-the-art modeling architectures and optimization methods. We unveil several modeling improvements, including Residual DCN, which adds attention and residual connections to the famous DCNv2 architecture. We share insights into combining and tuning SOTA architectures to create a unified model, including Dense Gating, Transformers and Residual DCN. We also propose novel techniques for calibration and describe how we productionalized deep learning based explore/exploit methods. To enable effective, production-grade serving of large ranking models, we detail how to train and compress models using quantization and vocabulary compression. We provide details about the deployment setup for large-scale use cases of Feed ranking, Jobs Recommendations, and Ads click-through rate (CTR) prediction. We summarize our learnings from various A/B tests by elucidating the most effective technical approaches. These ideas have contributed to relative metrics improvements across the board at LinkedIn: +0.5% member sessions in the Feed, +1.76% qualified job applications for Jobs search and recommendations, and +4.3% for Ads CTR. We hope this work can provide practical insights and solutions for practitioners interested in leveraging large-scale deep ranking systems.

cs.LG↗

arXiv2023

Bootstrapping Contrastive Learning Enhanced Music Cold-Start Matching

We study a particular matching task we call Music Cold-Start Matching. In short, given a cold-start song request, we expect to retrieve songs with similar audiences and then fastly push the cold-start song to the audiences of the retrieved songs to warm up it. However, there are hardly any studies done on this task. Therefore, in this paper, we will formalize the problem of Music Cold-Start Matching detailedly and give a scheme. During the offline training, we attempt to learn high-quality song representations based on song content features. But, we find supervision signals typically follow power-law distribution causing skewed representation learning. To address this issue, we propose a novel contrastive learning paradigm named Bootstrapping Contrastive Learning (BCL) to enhance the quality of learned representations by exerting contrastive regularization. During the online serving, to locate the target audiences more accurately, we propose Clustering-based Audience Targeting (CAT) that clusters audience representations to acquire a few cluster centroids and then locate the target audiences by measuring the relevance between the audience representations and the cluster centroids. Extensive experiments on the offline dataset and online system demonstrate the effectiveness and efficiency of our method. Currently, we have deployed it on NetEase Cloud Music, affecting millions of users. Code will be released in the future.

cs.IR↗

arXiv2022

Critical diffraction of irregular structure detonations and their predictability from experimentally obtained D-K data

The present work reports new experiments of detonation diffraction in a 2D channel configuration in stoichiometric mixtures of ethylene, ethane, and methane with oxygen as oxidizer. The flow field details are obtained using high-speed schlieren near the critical conditions of diffraction. The critical initial pressure for successful diffraction is reported for the ethylene, ethane and methane mixtures. The flow field details revealed that the lateral portion of the wave results in a zone of quenched ignition. The dynamics of the laterally diffracting shock front are found in good agreement with the recent model developed by Radulescu et al. (Physics of Fluids 2021). The model provides noticeable improvement over the local models using Whitham's characteristic rule and Wescott, Bdzil and Stewart's model for weakly curved reactive shocks. These models provide a link between the critical channel height and the critical wave curvature. The critical channel heights and global curvatures are found in very good agreement with the critical curvatures measured independently by Xiao and Radulescu (Combust. Flame 2020) in quasi-steady experiments in exponential horns for three mixtures tested. Furthermore, critical curvature data obtained by others in the literature was found to provide a good prediction of critical diffraction in 2D. These findings suggest that the critical diffraction of unstable detonations may be well predicted by a model based on the maximum curvature of the detonation front, where the latter is to be measured experimentally and account for the role of the cellular structure in the burning mechanism. This finding provides support to the view that models for unstable detonations at a meso-scale larger than the cell size, i.e., hydrodynamic average models, are meaningful.

physics.flu-dyn↗

arXiv2022

Directly wireless communication of human minds via non-invasive brain-computer-metasurface platform

Brain-computer interfaces (BCIs), invasive or non-invasive, have projected unparalleled vision and promise for assisting patients in need to better their interaction with the surroundings. Inspired by the BCI-based rehabilitation technologies for nerve-system impairments and amputation, we propose an electromagnetic brain-computer-metasurface (EBCM) paradigm, regulated by human's cognition by brain signals directly and non-invasively. We experimentally show that our EBCM platform can translate human's mind from evoked potentials of P300-based electroencephalography to digital coding information in the electromagnetic domain non-invasively, which can be further processed and transported by an information metasurface in automated and wireless fashions. Directly wireless communications of the human minds are performed between two EBCM operators with accurate text transmissions. Moreover, several other proof-of-concept mind-control schemes are presented using the same EBCM platform, exhibiting flexibly-customized capabilities of information processing and synthesis like visual-beam scanning, wave modulations, and pattern encoding.

cs.IT↗

arXiv2021

On the self-similarity of diffracting gaseous detonations and the critical channel width problem

One strategy for arresting propagating detonation waves in pipes is by imposing a sudden area enlargement, which provides a rapid lateral divergence of the gases in the reaction zone and attenuates the leading shock. For sufficiently small tube diameter, the detonation decays to a deflagration and the shock decays to negligible strengths. This is known as the critical tube diameter problem. In the present study, we provide a closed form model to predict the detonation quenching for 2D channels. Whitham's geometric shock dynamics, coupled with a shock evolution law based on shocks sustained by a constant source obtained by the shock change equations of Radulescu, is shown to capture the lateral shock dynamics response to the failure wave originating at the expansion corner. A criterion for successful detonation transmission to open space is that the lateral strain rate provided by the failure wave not exceed the critical strain rate of steady curved detonations. Using the critical lateral strain rate obtained by He and Clavin, a closed form solution is obtained for the critical channel opening permitting detonation transmission. The predicted critical channel width is found in very good agreement with our recent experiments and simulations of diffracting H$_2$/O$_2$/Ar detonations.

physics.flu-dyn↗

arXiv2020

Reprogrammable plasmonic topological insulators with ultrafast control

Topological photonics has revolutionized our understanding of light propagation, but most of current studies are focused on designing a static photonic structure. Developing a dynamic photonic topological platform to switch multiple topological functionalities at ultrafast speed is still a great challenge. Here we demonstrate an ultrafast reprogrammable plasmonic topological insulator, where the topological propagation route can be dynamically steered at nanosecond-level switching time, namely more than 10^7 times faster than the current state-of-the-art. This orders-of-magnitude improvement is achieved by using ultrafast electronic switches in an innovative way to implement the programmability. Due to the flexible programmability, many existing photonic topological functionalities can be integrated into this agile topological platform. Our work brings the current studies of photonic topological insulators to a digital and intelligent era, which could boost the development of intelligent and ultrafast photoelectric devices with built-in topological protection.

physics.app-ph↗

arXiv2020

Dynamics of Hydrogen-oxygen-argon cellular detonations with a constant mean lateral strain rate

The present work revisits the problem of modelling the real gaseous detonation dynamics at the macro-scale by simple steady one-dimensional (1D) models. Experiments of detonations propagating in channels with exponentially expanding cross-sections were conducted in the H2/O2/Ar reactive system. Steady detonation waves were obtained at the macro-scale, with cellular structures characterized by reactive transverse waves. For all the mixtures studied, the dependence of the mean detonation speed was found to be in excellent agreement with first principles predictions of quasi-1D detonation dynamics with lateral strain rate predicted from detailed chemical kinetic models. This excellent agreement departs from the earlier experiments of Radulescu and Borzou (2018) in more unstable detonations. The excellent agreement is likely due to the much longer reaction zone lengths of argon diluted hydrogen-oxygen detonations at low pressures, as compared with the characteristic induction zone lengths. While the cellular instability modifies the detonation induction zone, the detonation dynamics at the macro-scale are arguably controlled by its hydrodynamic thickness. Near the limit, minor discrepancy is observed, with the experimental detonations typically continuing to propagate to slightly higher lateral strain rates and higher velocity deficits.

physics.flu-dyn↗

arXiv2020

Role of instability on the limits of laterally strained detonation waves

The present work examines the role of instability and diffusive phenomena in controlling the limits of detonations subject to lateral strain rates. Experiments were conducted in hydrocarbon-oxygen mixtures with varying levels of cellular instability. These detonations were propagated in channels with exponentially enlarging cross-sections, following the recent works of Radulescu \& Borzou (2018) and Xiao \& Radulescu (2020). Steady detonation waves were obtained at the macro-scale, with the near-limit reaction zone structures characterized by significant unreacted gas pockets. The turbulent flame burning velocity of these pockets was evaluated to be 30 m/s to 70 m/s, which is larger than the theoretical laminar value by a factor of 2 to 7. For all the mixtures tested, the characteristic $D-K$ relationships, relating the detonation mean propagation speed with lateral flow divergence, were obtained directly from experiments and as well from the generalized ZND model with lateral strain rates using detailed chemical kinetics. The results showed that the degree of departure between experiments and the theoretical predictions increases significantly with the detonation instability level. As compared to the laminar ZND wave, the more unstable detonations are much more detonable than the more stable detonations, with substantially larger limiting divergence rates and maximum velocity deficits. Such enhanced detonability with detonation instability can be manifested in the significantly enhanced global rates of energy release with the notably suppressed thermal character of ignition for the more unstable detonations. This globally enhanced burning mechanism is found to be realized by the intensified auto-ignition assisted by the turbulent diffusive burning of the unreacted gas pockets, substantially shortening the characteristic reaction zone lengths.

physics.flu-dyn↗

arXiv2020

Effect of boundary layer losses on 2D detonation cellular structures

We evaluate the effect of boundary layer losses on two-dimensional H2/O2/Ar cellular detonations obtained in narrow channels. The experiments provide the details of the cellular structure and the detonation speed deficits from the ideal CJ speed. We model the effect of the boundary layer losses by incorporating the flow divergence in the third dimension due to the negative boundary layer displacement thickness, modeled using Mirels' theory. The cellular structures obtained numerically with the resulting quasi-2D formulation of the reactive Euler equations with two-step chain-branching chemistry are found in excellent agreement with experiment, both in terms of cell dynamics and velocity deficits, provided the boundary layer constant of Mirels is modified by a factor of 2. A significant increase in the cell size is found with increasing velocity deficit. This is found to be very well captured by the induction zone increase in slower detonations due to the lower temperatures in the induction zone.

physics.flu-dyn↗

arXiv2013

Target Mass Monitoring and Instrumentation in the Daya Bay Antineutrino Detectors

The Daya Bay experiment measures sin^2 2θ_13 using functionally identical antineutrino detectors located at distances of 300 to 2000 meters from the Daya Bay nuclear power complex. Each detector consists of three nested fluid volumes surrounded by photomultiplier tubes. These volumes are coupled to overflow tanks on top of the detector to allow for thermal expansion of the liquid. Antineutrinos are detected through the inverse beta decay reaction on the proton-rich scintillator target. A precise and continuous measurement of the detector's central target mass is achieved by monitoring the the fluid level in the overflow tanks with cameras and ultrasonic and capacitive sensors. In addition, the monitoring system records detector temperature and levelness at multiple positions. This monitoring information allows the precise determination of the detectors' effective number of target protons during data taking. We present the design, calibration, installation and in-situ tests of the Daya Bay real-time antineutrino detector monitoring sensors and readout electronics.

physics.ins-det↗

arXiv2012

Leakage Tests of the Stainless Steel Vessels of the Antineutrino Detectors in the Daya Bay Reactor Neutrino Experiment

The antineutrino detectors in the Daya Bay reactor neutrino experiment are liquid scintillator detectors designed to detect low energy particles from antineutrino interactions with high efficiency and low backgrounds. Since the antineutrino detector will be installed in a water Cherenkov cosmic ray veto detector and will run for 3 to 5 years, ensuring water tightness is critical to the successful operation of the antineutrino detectors. We choose a special method to seal the detector. Three leak checking methods have been employed to ensure the seal quality. This paper will describe the sealing method and leak testing results.

physics.ins-det↗