Wenping Zhu

Wenping Zhu

Auto-generated profile, based on publicly available information

Wenping Zhu

Tsinghua University

Research Focus & Metrics

Academic Career Length

9Years (2018-Present)

(Calculated once Scilit publication data is synced: current year − year of first publication + 1)

Representative Publications

View All Publications
25 January 2022 in IEEE Transactions on Computer-Aided Design of Integrated Circuits and Systems
Verified
Journal Article
Closed access
ALAng Li
Ang Li
HMHuiyu Mo
Huiyu Mo
WZWenping Zhu
Wenping Zhu
QLQiang Li
Qiang Li
SYShouyi Yin
Shouyi Yin
LLLeibo Liu
Leibo Liu
Convolutional neural network (CNN) has demonstrated great success in pattern recognition scenarios at the cost of nearly billions of parameters and consequent convolution operations. Various dedicated hardware designs are proposed to accelerate the CNN computation in more energy-efficient manners. Especially, the bit-serial accelerator (BSA) is one of the most effective approaches on resource-limited platforms by eliminating zero-bit computations. However, the irregular distribution and varying number of effectual (nonzero) bits in weights significantly cause hardware underutilization, impeding further performance improvement of state-of-the-art BSAs. To address this issue, BitCluster, a hardware-friendly quantization method, is proposed to make each weight with the identical number of effectual bits for load-balanced computation. Considering distinct sensitivities to weight precision in different neural layers, layer-level BitCluster is proposed to design further for fine-grained weight quantization. It systematically determines the layerwise quantization configurations, which significantly improve the overall performance with <1% accuracy loss. BitCluster is comprehensively evaluated on a BitCluster-compatible BSA design by taking six mainstream CNN models as benchmarks. The experimental results show that the BitCluster-based BSA achieves $1.6\times $ higher hardware utilization and $3.4\times $ speedup on average than state-of-the-art BSAs, with $5\times $ better energy efficiency on average.

27 August 2018 in IEEE Transactions on Multimedia
Verified
Journal Article
Closed access
HMHuiyu Mo
Huiyu Mo
LLLeibo Liu
Leibo Liu
WZWenping Zhu
Wenping Zhu
SYShouyi Yin
Shouyi Yin
SWShaojun Wei
Shaojun Wei
Face alignment is a critical task in many multimedia and vision applications that use face-based algorithms. Recent research has focused on achieving efficient initialization to improve performance; however, the use of facial attributes and the extent of their correlation with initialization have not been fully exploited. This paper presents a lightweight method called expression- and pose-based adaptive initialization (EXPAI), in which facial attributes, i.e., expression and pose information, are used as priors. This approach can significantly improve face alignment performance. In addition, reliable expression and head pose information can be derived simultaneously in the same framework. First, an expression- and pose-based template dictionary is formed by augmenting the mean shape across three degrees of freedom, thereby substantially improving the robustness of the initial shape with respect to large head pose variations. Second, each of the templates corresponding to an image of interest, which are jointly determined using a shape-constrained multiclass classifier and binary classifiers, is assigned a pretrained confidence coefficient. The initial shape that is thus generated for subsequent cascaded regression is more adaptive and enables higher accuracy. Furthermore, EXPAI enables initialization with significantly increased computational efficiency because of its independence from the original dataset. The experimental results obtained on the widely used 300-W dataset show that our method achieves very competitive performance compared with that of the state-of-the-art methods. In particular, for the challenging subset of 300-W, EXPAI reduces errors by more than 14% compared with coarse-to-fine shape searching (CFSS), which currently exhibits the best performance among regression-based approaches. Furthermore, a speed increase of more than 10x compared with CFSS is achieved.

25 November 2019 in IEEE Transactions on Circuits and Systems for Video Technology
Verified
Journal Article
Closed access
HMHuiyu Mo
Huiyu Mo
LLLeibo Liu
Leibo Liu
WZWenping Zhu
Wenping Zhu
QLQiang Li
Qiang Li
HLHong Liu
Hong Liu
SWShaojun Wei
Shaojun Wei
Face detection and alignment are two fundamental tasks for facial applications and the corresponding accelerators have been designed to enable energy-efficient acceleration. However, these dedicated accelerators are always designed separately, thereby ignoring the inherent correlation between face detection and alignment and causing additional communication and area overhead. Based on this motivation, a multi-task cascaded convolutional networks (MTCNN) algorithm-based accelerator is presented in this work to support both face detection and alignment for multiple faces. First, multiply-accumulate (MAC) operations and memory access of the magnification process in the resize module are reduced by 22.8% and 24.8% on average, respectively, when compared with those of similar methods. Second, clustering non-maximum suppression (C-NMS) is proposed to significantly reduce the intersection over union computation and eliminate the hardware-inference sorting process in NMS, yielding a 16.0% speedup in the overall process. Third, an efficient pipeline architecture is proposed to implement a complexity-and memory-intensive proposal network of MTCNN in a more computationally efficient manner, with 38.3% less memory capacity than a similar solution. Meanwhile, only approximately half of the multipliers are needed to achieve the same throughput with high pipeline utilization. Fourth, considering the variable number of faces in each input, a batch schedule mechanism is proposed to improve the fully-connected layer hardware utilization by 16.7% on average in the batch process. Based on a simulation with the TSMC 28 nm CMOS process, this accelerator consumes only 10.9ms at 400 MHz to simultaneously process 5 faces. The power efficiency reaches 4.80 TOPS/W, which is 227.4× higher than that of the state-of-the-art solution.

9 December 2022 in IEEE Journal of Solid-State Circuits
Verified
Journal Article
Closed access
YZYihong Zhu
Yihong Zhu
WZWenping Zhu
Wenping Zhu
CLChongyang Li
Chongyang Li
MZMin Zhu
Min Zhu
CDChenchen Deng
Chenchen Deng
LLLeibo Liu
Leibo Liu
Post-quantum cryptography (PQC) is investigated to replace the classical public cryptography algorithms, which would be completely broken by large-scale quantum computers. However, current PQC schemes have completely different mathematical foundations and parameter sets, which makes the implementation of unified PQC processor extremely challenging. To address this issue, an agile PQC processor, RePQC, is proposed in this work to support schemes on multiple mathematical problems. First, the hierarchical calculation framework, ranging from algorithm level, task level, and coefficient level, is proposed to achieve desirable flexibility and energy efficiency. Second, a hybrid processing element array is built to support arithmetic and logical operations simultaneously, while algorithm-hardware co-design is utilized in task-level schedulers to further improve the algorithm-oriented energy efficiency. Finally, parallelism exploration and algorithm-level computation transformation is further utilized to optimize the configuration on RePQC for higher throughput. Fabricated in a 28-nm process, RePQC achieves the energy efficiency of 3.4 uJ/Op and the throughput of 48 kOPS, which is $2\times $ and $23\times $ higher than the state-of-the-art work, respectively. To the best of our knowledge, RePQC is the first silicon-proven PQC processor for different mathematical problems.

25 October 2024 in IEEE Journal of Solid-State Circuits
Verified
Journal Article
Closed access
YZYihong Zhu
Yihong Zhu
WZWenping Zhu
Wenping Zhu
YOYi Ouyang
Yi Ouyang
JSJunwen Sun
Junwen Sun
QZQi Zhao
Qi Zhao
LLLeibo Liu
Leibo Liu
Post-quantum cryptography (PQC) is currently being standardized to replace the existing public-key cryptography for data security in the era of quantum computing. PQC algorithms exhibit considerable diversity in their underlying mathematical problems, storage requirements, and computational patterns, thus complicating unified PQC architecture design. To address this issue, a unified PQC domain-specific accelerator (DSA), post-quantum processing unit (PQPU), is proposed to address the trade-off between performance and flexibility at the algorithm, architecture, and circuit levels. First, a task-clustering-based framework is proposed to enable task-level parallel execution by utilizing the inherent parallelism and common functions across different PQC algorithms. Second, a region-based dynamically updated task path (TP) is constructed to facilitate automatic task-dependency management, with agile control flow and minimized overheads. Finally, algorithm-hardware co-optimizations are proposed in each task cluster to improve throughput and energy efficiency. Fabricated in a 28-nm process, PQPU has energy efficiency and throughput of 4.4 $\mu$ J/Op and 69.4 kOPS. The energy-delay product (EDP) and throughput achieved are 19.3% and 44.6% compared to the state-of-the-art design, respectively. To the best of our knowledge, PQPU is the first silicon-proven PQC accelerator that supports all valid schemes in NIST PQC standardization.

11 May 2020 in IEEE Transactions on Multimedia
Verified
Journal Article
Closed access
HMHuiyu Mo
Huiyu Mo
LLLeibo Liu
Leibo Liu
WZWenping Zhu
Wenping Zhu
QLQiang Li
Qiang Li
SYShouyi Yin
Shouyi Yin
ELEric Li
Eric Li
The mnemonic descent method (MDM) algorithm is the first end-to-end recurrent convolutional system for high-accuracy face alignment. However, the heavy computational complexity and high memory access demands make it difficult to satisfy the requirements of real-time applications. To address this problem, the improved MDM (I-MDM) algorithm is proposed for efficient hardware mplementation based on several hardwareoriented optimizations. First, a patch merging mechanism is introduced to dynamically cluster and eliminate redundant landmarks, which significantly reduces computational complexity with minimal accuracy loss. Second, a dedicated convolutional layer is inserted to halve the number of computations and memory access of the subsequent fully connected layer, yielding a 4.42% decrease in the failure rate. Third, a lightweight preprocessing method named dual regressors is proposed to reinitialize face images, which can greatly improve the overall accuracy. Meanwhile, compared with the similar method, the DR method can reduce computations and memory storage by nearly 99.9%. Overall and compared to the MDM algorithm, I-MDM not only reduces the number of computations by 23.5% but also decreases the failure rate by 17.9% on the 300W test set. Based on the proposed I-MDM algorithm, an I-MDM-based hardwired accelerator is presented using TSMC 65 nm CMOS process. First, compared with similar solutions, the gradient calculation operation is rearranged and loaded pixels are reused in HoG feature extraction to eliminate all division operations and 25% off-chip memory access. Second, patch-independent central activations are used to enable patch-level pipelined operations, yielding a 2x speed-up in the overall process. This accelerator achieves 460 GOPS/W energy efficiency at 330 MHz, which is 38x higher than that of the most recent face alignment accelerator with the same process.