Hongyu Wang
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
Hongyu Wang has a total of 25 co-authors. Here are the top co-authors by number of shared publications:
- Baochang Zhang
Beihang University
1 shared publication - Bingqian Li
Gaoling School of Artificial Intelligence, Renmin University of China, Beijing, China
1 shared publication - Chunhui Liu
Institute of Unmanned System, Beihang University, Beijing, China
1 shared publication - Daoyuan Wang
Meituan Group, Beijing, China
1 shared publication - Dong Xu
Xiaohongshu Technology Co., Ltd., Shanghai, China
1 shared publication - Duona Zhang
North China University of Technology
1 shared publication
Latest Publications
- Open Access
- 19 July 2026
- Published by Association for Computing Machinery (ACM)
- p. 4578-4582
RedGR: Unified Generative Retrieval for Recommendation in REDnote
- 19 July 2026
- Published by Association for Computing Machinery (ACM)
- p. 4578-4582
Open AccessRedGR: Unified Generative Retrieval for Recommendation in REDnote
- Open Access
- 21 October 2024
- Published by Association for Computing Machinery (ACM)
- p. 5015-5022
Sequence-level Semantic Representation Fusion for Recommender Systems
With the rapid development of recommender systems, there is increasing side information that can be employed to improve the recommendation performance. Specially, we focus on the utilization of the associated textual data of items (e.g., product title) and study how text features can be effectively fused with ID features in sequential recommendation. In this paper, we propose a novel Te xt-I D semantic fusion approach for sequential Rec ommendation, namely TedRec. The core idea is to conduct a sequence-level semantic fusion approach by integrating global contexts. The key strategy lies in that we transform the text and ID embeddings by Fourier Transform from time domain to frequency domain. In the frequency domain, the global sequential characteristics of the original sequences are inherently aggregated into the transformed representations, so that we can employ simple multiplicative operations to effectively fuse the two kinds of item features. Our fusion approach can be proved to have the same effects of contextual convolution, so as to achieving sequence-level semantic fusion. Further, we propose to enhance the discriminability of the text embeddings from the text encoder, by adaptively injecting positional information via a mixture-of-experts (MoE) modulation method. Both offline and online experiments demonstrate the effectiveness of our approach.
- 21 October 2024
- Published by Association for Computing Machinery (ACM)
- p. 5015-5022
Open AccessSequence-level Semantic Representation Fusion for Recommender Systems
With the rapid development of recommender systems, there is increasing side information that can be employed to improve the recommendation performance. Specially, we focus on the utilization of the associated textual data of items (e.g., product title) and study how text features can be effectively fused with ID features in sequential recommendation. In this paper, we propose a novel Te xt-I D semantic fusion approach for sequential Rec ommendation, namely TedRec. The core idea is to conduct a sequence-level semantic fusion approach by integrating global contexts. The key strategy lies in that we transform the text and ID embeddings by Fourier Transform from time domain to frequency domain. In the frequency domain, the global sequential characteristics of the original sequences are inherently aggregated into the transformed representations, so that we can employ simple multiplicative operations to effectively fuse the two kinds of item features. Our fusion approach can be proved to have the same effects of contextual convolution, so as to achieving sequence-level semantic fusion. Further, we propose to enhance the discriminability of the text embeddings from the text encoder, by adaptively injecting positional information via a mixture-of-experts (MoE) modulation method. Both offline and online experiments demonstrate the effectiveness of our approach.
- 8 November 2021
- Published by Institute of Electrical and Electronics Engineers (IEEE)
- p. 1-6
A Computing-in-memory Scheme with Series Bit-cell in STT-MRAM for Efficient Multi-bit Analog Multiplication
Computing-in-memory (CIM) is widely studied to solve the Von Neumann bottleneck, which improves energy-efficient computing. In this work, we propose a CIM with series bit-cell (SBCIM) scheme, which can perform the multi-bit analog multiplication in spin-transfer torque magnetic random access memory (STT-MRAM). Utilizing the proposed bit-cell structure consisting of three transistors and one magnetic tunnel junction (MTJ), multiple bit-cells on a column can be connected in series to overcome the inherent low on/off ratio of MTJ in CIM. Then, by converting the input data into a current injected into bit-line and the bit-line voltage into time-domain, the multiplication of two 2-bit numbers is implemented. In addition, a difference gain (DG) circuit is also designed to increase the difference of the signal representing the multiplication value, which achieves the multiplication of two 3-bit numbers. Simulation results show that the signal margin in our scheme is 10~200 times higher than that of conventional memory array in CIM schemes. Meanwhile, compared with the Spintronic Processing Unit (SPU) scheme, the delay and energy of Multiply Accumulate (MAC) operation in SBCIM scheme have been improved by 60 times and 8.8 times under 3-bit precision, respectively.
- 8 November 2021
- Published by Institute of Electrical and Electronics Engineers (IEEE)
- p. 1-6
A Computing-in-memory Scheme with Series Bit-cell in STT-MRAM for Efficient Multi-bit Analog Multiplication
Computing-in-memory (CIM) is widely studied to solve the Von Neumann bottleneck, which improves energy-efficient computing. In this work, we propose a CIM with series bit-cell (SBCIM) scheme, which can perform the multi-bit analog multiplication in spin-transfer torque magnetic random access memory (STT-MRAM). Utilizing the proposed bit-cell structure consisting of three transistors and one magnetic tunnel junction (MTJ), multiple bit-cells on a column can be connected in series to overcome the inherent low on/off ratio of MTJ in CIM. Then, by converting the input data into a current injected into bit-line and the bit-line voltage into time-domain, the multiplication of two 2-bit numbers is implemented. In addition, a difference gain (DG) circuit is also designed to increase the difference of the signal representing the multiplication value, which achieves the multiplication of two 3-bit numbers. Simulation results show that the signal margin in our scheme is 10~200 times higher than that of conventional memory array in CIM schemes. Meanwhile, compared with the Spintronic Processing Unit (SPU) scheme, the delay and energy of Multiply Accumulate (MAC) operation in SBCIM scheme have been improved by 60 times and 8.8 times under 3-bit precision, respectively.
- Open Access
- 4 February 2020
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Access
- Vol. 8, 27097-27105
Modulated Autocorrelation Convolution Networks for Automatic Modulation Classification Based on Small Sample Set
For modulation classification, hand-crafted approaches can generalize well from a few samples, yet deep learning algorithms require millions of samples to achieve the superior performance with purely data-driven manner. However for many practical problems only with small sample set (SSS) available, there still remains a challenge for deep learning. In this paper, we employ deep learning to solve the modulation classification task in a more practical setting, particularly suffering from the SSS problem and with low signal-to-noise ratios (SNRs). Novel modulated autocorrelation convolution networks (MACNs) are introduced to capture periodic representation for automatic modulation classification (AMC). In MACNs, modulated communication signals are classified with the periodic local features under an autocorrelation convolution criterion. Modulation filters are utilized to enhance the capacity of the convolution filters and compress the model. On a challenging SSS learning task in low SNRs, MACNs achieve state-of-the-art performance that outperforms the existing algorithms for AMC, while compressing the size of required storage space of convolutional filters by a factor of 8 compared with convolution neural networks (CNNs).
- 4 February 2020
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Access
- Vol. 8, 27097-27105
Open AccessModulated Autocorrelation Convolution Networks for Automatic Modulation Classification Based on Small Sample Set
For modulation classification, hand-crafted approaches can generalize well from a few samples, yet deep learning algorithms require millions of samples to achieve the superior performance with purely data-driven manner. However for many practical problems only with small sample set (SSS) available, there still remains a challenge for deep learning. In this paper, we employ deep learning to solve the modulation classification task in a more practical setting, particularly suffering from the SSS problem and with low signal-to-noise ratios (SNRs). Novel modulated autocorrelation convolution networks (MACNs) are introduced to capture periodic representation for automatic modulation classification (AMC). In MACNs, modulated communication signals are classified with the periodic local features under an autocorrelation convolution criterion. Modulation filters are utilized to enhance the capacity of the convolution filters and compress the model. On a challenging SSS learning task in low SNRs, MACNs achieve state-of-the-art performance that outperforms the existing algorithms for AMC, while compressing the size of required storage space of convolutional filters by a factor of 8 compared with convolution neural networks (CNNs).