Huangyu Dai
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Previous Organizations / Institutes
The organizations / institutes from the scholar's other articles in the past ten years
Huangyu Dai has a total of 41 co-authors. Here are the top co-authors by number of shared publications:
- Ben Chen
Kuaishou Technology, Hangzhou, China
8 shared publications - Wei Ning
Alibaba Group, Hangzhou, China
4 shared publications - Wen Jiang
Alibaba Group, Hangzhou, China
4 shared publications - Chenyi Lei
Kuaishou Technology, Hangzhou, China
3 shared publications - Dehong Gao
School of Cybersecurity, Northwestern Polytechnical University, Xi’an, China
3 shared publications - Libin Yang
School of Cybersecurity, Northwestern Polytechnical University, Xian, China
3 shared publications
Latest Publications
- Open Access
- 15 June 2026
- Published by Association for Computing Machinery (ACM)
- p. 1583-1587
UniDGF: A Unified Detection-to-Generation Framework for Hierarchical Object Visual Recognition
- 15 June 2026
- Published by Association for Computing Machinery (ACM)
- p. 1583-1587
Open AccessUniDGF: A Unified Detection-to-Generation Framework for Hierarchical Object Visual Recognition
- Open Access
UniDGF: A Unified Detection-to-Generation Framework for Hierarchical Object Visual Recognition
Achieving visual semantic understanding requires a unified framework that simultaneously handles object detection, category prediction, and attribute recognition. However, current advanced approaches rely on global similarity and struggle to capture fine-grained category distinctions and category-specific attribute diversity, especially in large-scale e-commerce scenarios. To overcome these challenges, we introduce a detection-guided generative framework that predicts hierarchical category and attribute tokens. For each detected object, we extract refined ROI-level features and employ a BART-based generator to produce semantic tokens in a coarse-to-fine sequence covering category hierarchies and property-value pairs, with support for property-conditioned attribute recognition. Experiments on both large-scale proprietary e-commerce datasets and open-source datasets demonstrate that our approach significantly outperforms existing similarity-based pipelines and multi-stage classification systems, achieving stronger fine-grained recognition and more coherent unified inference.
Open AccessUniDGF: A Unified Detection-to-Generation Framework for Hierarchical Object Visual Recognition
Achieving visual semantic understanding requires a unified framework that simultaneously handles object detection, category prediction, and attribute recognition. However, current advanced approaches rely on global similarity and struggle to capture fine-grained category distinctions and category-specific attribute diversity, especially in large-scale e-commerce scenarios. To overcome these challenges, we introduce a detection-guided generative framework that predicts hierarchical category and attribute tokens. For each detected object, we extract refined ROI-level features and employ a BART-based generator to produce semantic tokens in a coarse-to-fine sequence covering category hierarchies and property-value pairs, with support for property-conditioned attribute recognition. Experiments on both large-scale proprietary e-commerce datasets and open-source datasets demonstrate that our approach significantly outperforms existing similarity-based pipelines and multi-stage classification systems, achieving stronger fine-grained recognition and more coherent unified inference.
- Open Access
- 10 November 2025
- Published by Association for Computing Machinery (ACM)
- p. 4695-4699
H-PRM: A Pluggable Hotword Pre-Retrieval Module for Various Speech Recognition Systems
- 10 November 2025
- Published by Association for Computing Machinery (ACM)
- p. 4695-4699
Open AccessH-PRM: A Pluggable Hotword Pre-Retrieval Module for Various Speech Recognition Systems
- Open Access
- 10 November 2025
- Published by Association for Computing Machinery (ACM)
- p. 1788-1797
UniECS: Unified Multimodal E-Commerce Search Framework with Gated Cross-modal Fusion
- 10 November 2025
- Published by Association for Computing Machinery (ACM)
- p. 1788-1797
Open AccessUniECS: Unified Multimodal E-Commerce Search Framework with Gated Cross-modal Fusion
- 22 August 2024
- Published by Springer Nature
- p. 55-71
Robust Interaction-Based Relevance Modeling for Online e-Commerce Search
- 22 August 2024
- Published by Springer Nature
- p. 55-71
Robust Interaction-Based Relevance Modeling for Online e-Commerce Search
- 13 August 2024
- Published by Elsevier in Expert Systems with Applications
- Vol. 258
LLMs-based machine translation for E-commerce
- 13 August 2024
- Published by Elsevier in Expert Systems with Applications
- Vol. 258
LLMs-based machine translation for E-commerce
- Open Access
- 13 May 2024
- Published by Association for Computing Machinery (ACM)
- p. 670-673
General2Specialized LLMs Translation for E-commerce
Existing Neural Machine Translation (NMT) models mainly handle translation in the general domain, while overlooking domains with special writing formulas, such as e-commerce and legal documents. Taking e-commerce as an example, the texts usually include amounts of domain-related words and have more grammar problems, which leads to inferior performances of current NMT methods. To address these problems, we collect two domain-related resources, including a set of term pairs (aligned Chinese-English bilingual terms) and a parallel corpus annotated for the e-commerce domain. Furthermore, we propose a two-step fine-tuning paradigm (named G2ST) with self-contrastive semantic enhancement to transfer one general NMT model to the specialized NMT model for e-commerce. The paradigm can be used for the NMT models based on Large language models (LLMs). Extensive evaluations on real e-commerce titles demonstrate the superior translation quality and robustness of our G2ST approach, as compared with state-of-the-art NMT models such as LLaMA, Qwen, GPT-3.5, and even GPT-4.
- 13 May 2024
- Published by Association for Computing Machinery (ACM)
- p. 670-673
Open AccessGeneral2Specialized LLMs Translation for E-commerce
Existing Neural Machine Translation (NMT) models mainly handle translation in the general domain, while overlooking domains with special writing formulas, such as e-commerce and legal documents. Taking e-commerce as an example, the texts usually include amounts of domain-related words and have more grammar problems, which leads to inferior performances of current NMT methods. To address these problems, we collect two domain-related resources, including a set of term pairs (aligned Chinese-English bilingual terms) and a parallel corpus annotated for the e-commerce domain. Furthermore, we propose a two-step fine-tuning paradigm (named G2ST) with self-contrastive semantic enhancement to transfer one general NMT model to the specialized NMT model for e-commerce. The paradigm can be used for the NMT models based on Large language models (LLMs). Extensive evaluations on real e-commerce titles demonstrate the superior translation quality and robustness of our G2ST approach, as compared with state-of-the-art NMT models such as LLaMA, Qwen, GPT-3.5, and even GPT-4.
- Open Access
General2Specialized LLMs Translation for E-commerce
Existing Neural Machine Translation (NMT) models mainly handle translation in the general domain, while overlooking domains with special writing formulas, such as e-commerce and legal documents. Taking e-commerce as an example, the texts usually include amounts of domain-related words and have more grammar problems, which leads to inferior performances of current NMT methods. To address these problems, we collect two domain-related resources, including a set of term pairs (aligned Chinese-English bilingual terms) and a parallel corpus annotated for the e-commerce domain. Furthermore, we propose a two-step fine-tuning paradigm (named G2ST) with self-contrastive semantic enhancement to transfer one general NMT model to the specialized NMT model for e-commerce. The paradigm can be used for the NMT models based on Large language models (LLMs). Extensive evaluations on real e-commerce titles demonstrate the superior translation quality and robustness of our G2ST approach, as compared with state-of-the-art NMT models such as LLaMA, Qwen, GPT-3.5, and even GPT-4.
Open AccessGeneral2Specialized LLMs Translation for E-commerce
Existing Neural Machine Translation (NMT) models mainly handle translation in the general domain, while overlooking domains with special writing formulas, such as e-commerce and legal documents. Taking e-commerce as an example, the texts usually include amounts of domain-related words and have more grammar problems, which leads to inferior performances of current NMT methods. To address these problems, we collect two domain-related resources, including a set of term pairs (aligned Chinese-English bilingual terms) and a parallel corpus annotated for the e-commerce domain. Furthermore, we propose a two-step fine-tuning paradigm (named G2ST) with self-contrastive semantic enhancement to transfer one general NMT model to the specialized NMT model for e-commerce. The paradigm can be used for the NMT models based on Large language models (LLMs). Extensive evaluations on real e-commerce titles demonstrate the superior translation quality and robustness of our G2ST approach, as compared with state-of-the-art NMT models such as LLaMA, Qwen, GPT-3.5, and even GPT-4.
- Open Access
- 7 September 2021
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Access
- Vol. 9, 126307-126318
Joint Bernoulli Filtering and MIMO Processing for Detection of Moving Targets in Shallow Ocean Environments
A joint multiple-input multiple-output (MIMO) processing framework is proposed to exploit spatial diversity and moving cues for the enhancement of target detection in a shallow ocean environment. Orthogonal signals are transmitted to illuminate different aspects of a target and beamforming operation is carried out over the received data for estimating target bearing. The target range is achieved by a replica correlation integrator where the beamformer outputs are matched with transmitted signals. After meanshift clustering algorithm is carried out over the bearing-range spectrums to generate the clutter centers, the potential trajectories of targets can be tracked by an improved Bernoulli filter with inputs of these centers. The at-sea experimental results have shown the effectiveness of the joint processing framework in MIMO detection of moving targets.
- 7 September 2021
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Access
- Vol. 9, 126307-126318
Open AccessJoint Bernoulli Filtering and MIMO Processing for Detection of Moving Targets in Shallow Ocean Environments
A joint multiple-input multiple-output (MIMO) processing framework is proposed to exploit spatial diversity and moving cues for the enhancement of target detection in a shallow ocean environment. Orthogonal signals are transmitted to illuminate different aspects of a target and beamforming operation is carried out over the received data for estimating target bearing. The target range is achieved by a replica correlation integrator where the beamformer outputs are matched with transmitted signals. After meanshift clustering algorithm is carried out over the bearing-range spectrums to generate the clutter centers, the potential trajectories of targets can be tracked by an improved Bernoulli filter with inputs of these centers. The at-sea experimental results have shown the effectiveness of the joint processing framework in MIMO detection of moving targets.
- Open Access
- 13 October 2020
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Access
- Vol. 8, 187948-187958
Deconvolved Conventional Beamforming and Adaptive Cubature Kalman Filter Based Distant Speech Perception System
A spatial-temporal processing framework integrated of speech enhancement and speech tracking is proposed in this paper for distant speech perception. First, weak speech signals are enhanced by the deconvolved conventional beamforming (DCBF) with a microphone array. By virtue of the narrow beamwidth and low sidelobes of the DCBF, the competing sources can be effectively suppressed without introducing extra speech distortion. Second, with the accurate bearing provided by the DCBF, the Cubature Kalman filter can be utilized to track the speech source of interest. By introducing a scaling factor in the current statistical motion model, a new tracking algorithm is proposed which is suitable for both maneuvering and nonmaneuvering speech sources. The introduced scaling factor can be adaptively adjusted to improve the tracking performance of the proposed algorithm for different motion models. Numerical results show that the proposed algorithm can provide better tracking performance than the conventional one. In particular, the tracking root mean square error can be reduced by half for some cases.
- 13 October 2020
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Access
- Vol. 8, 187948-187958
Open AccessDeconvolved Conventional Beamforming and Adaptive Cubature Kalman Filter Based Distant Speech Perception System
A spatial-temporal processing framework integrated of speech enhancement and speech tracking is proposed in this paper for distant speech perception. First, weak speech signals are enhanced by the deconvolved conventional beamforming (DCBF) with a microphone array. By virtue of the narrow beamwidth and low sidelobes of the DCBF, the competing sources can be effectively suppressed without introducing extra speech distortion. Second, with the accurate bearing provided by the DCBF, the Cubature Kalman filter can be utilized to track the speech source of interest. By introducing a scaling factor in the current statistical motion model, a new tracking algorithm is proposed which is suitable for both maneuvering and nonmaneuvering speech sources. The introduced scaling factor can be adaptively adjusted to improve the tracking performance of the proposed algorithm for different motion models. Numerical results show that the proposed algorithm can provide better tracking performance than the conventional one. In particular, the tracking root mean square error can be reduced by half for some cases.