Jiaming Zhao
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Research Subjects
Jiaming Zhao has a total of 13 co-authors. Here are the top co-authors by number of shared publications:
- Yijiu Zhao
University of Electronic Science and Technology of China
7 shared publications - Naixin Zhou
School of Automation Engineering, University of Electronic Science and Technology of China, Chengdu, China
6 shared publications - Shibo Chen
School of Automation Engineering, University of Electronic Science and Technology of China, Chengdu, China
5 shared publications - Jiawei Cao
University of Electronic Science and Technology of China
4 shared publications - Feiyu Long
Shenzhen Institute for Advanced Study, University of Electronic Science and Technology of China 1 , Chengdu 611731
2 shared publications - Guibing Zhu
School of Shenzhen Institute for Advanced Study, University of Electronic Science and Technology of China, Chengdu, China
2 shared publications
Latest Publications
- 1 November 2026
- Published by AIP Publishing in Review of Scientific Instruments
- Vol. 97 (11)
Balanced multi-specification indirect testing of analog circuits via heterogeneous dual-level ensemble regression
Indirect testing reduces analog integrated-circuit test cost by predicting electrical specifications from low-cost response-derived features. In practical multi-specification flows, target difficulty is often uneven: easy labels can dominate mean accuracy, whereas difficult labels determine test-flow reliability. This paper formulates multi-specification indirect testing as a balanced regression problem evaluated by mean accuracy, worst-target performance, and cross-target stability. We propose Heterogeneous Dual-Level Ensemble Regression (HDLER), a target-wise framework that trains complementary base regressors independently, fuses them by stacking with feature passthrough, and adaptively blends the stacked output with base-model averaging using out-of-fold validation. On a simulated operational-amplifier dataset with 5000 samples, 58 features, and 11 specifications, HDLER achieves a mean target-wise test R2 of 0.9508 across five random seeds, outperforming stacking alone (0.9339) and simple averaging (0.7843). Relative to stacking, it improves worst-target R2 from 0.6487 to 0.6939, increases difficult-target mean R2 from 0.8264 to 0.8659, and reduces the across-target standard deviation from 0.1120 to 0.0952. A measured-data validation on commercial OPA314 devices further obtains a mean R2 of 0.9874 and a worst-target R2 of 0.9670. The results show that target-aware fusion improves multi-specification reliability when performance is limited by the weakest labels.
- 1 November 2026
- Published by AIP Publishing in Review of Scientific Instruments
- Vol. 97 (11)
Balanced multi-specification indirect testing of analog circuits via heterogeneous dual-level ensemble regression
Indirect testing reduces analog integrated-circuit test cost by predicting electrical specifications from low-cost response-derived features. In practical multi-specification flows, target difficulty is often uneven: easy labels can dominate mean accuracy, whereas difficult labels determine test-flow reliability. This paper formulates multi-specification indirect testing as a balanced regression problem evaluated by mean accuracy, worst-target performance, and cross-target stability. We propose Heterogeneous Dual-Level Ensemble Regression (HDLER), a target-wise framework that trains complementary base regressors independently, fuses them by stacking with feature passthrough, and adaptively blends the stacked output with base-model averaging using out-of-fold validation. On a simulated operational-amplifier dataset with 5000 samples, 58 features, and 11 specifications, HDLER achieves a mean target-wise test R2 of 0.9508 across five random seeds, outperforming stacking alone (0.9339) and simple averaging (0.7843). Relative to stacking, it improves worst-target R2 from 0.6487 to 0.6939, increases difficult-target mean R2 from 0.8264 to 0.8659, and reduces the across-target standard deviation from 0.1120 to 0.0952. A measured-data validation on commercial OPA314 devices further obtains a mean R2 of 0.9874 and a worst-target R2 of 0.9670. The results show that target-aware fusion improves multi-specification reliability when performance is limited by the weakest labels.
- 1 October 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Vehicular Technology
- Vol. 75 (10) , 25361-25366
Domain Adaptation for Automatic Modulation Classification Based on Complex Temporal Convolutional Autoencoder
- 1 October 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Vehicular Technology
- Vol. 75 (10) , 25361-25366
Domain Adaptation for Automatic Modulation Classification Based on Complex Temporal Convolutional Autoencoder
- 10 April 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Very Large Scale Integration (VLSI) Systems
- Vol. 34 (7) , 2274-2283
Bayesian Model Calibration for Deep-Learning-Based Indirect Test of Analog Circuits
The indirect test offers a cost-effective alternative for the specification test of analog integrated circuits (ICs). It infers high-cost specifications from low-cost measurements through a mapping function. Deep learning provides an efficient means for modeling this mapping. However, batch-to-batch variations in IC manufacturing often lead to distribution shifts, which degrade the accuracy of such data-driven techniques. To address this, we propose a transfer learning-based framework—Bayesian model calibration (BMC)—for the indirect test. During testing, a deep neural network (DNN) is trained with early data for estimation. When a new-batch arrives, a small number of samples are conventionally tested to form a calibration set. The original estimator is updated with the new dataset via a Bayesian loss function. This loss function integrates a prior term (reflecting the early data) and a likelihood term (reflecting the new data), enabling a principled tradeoff between prior knowledge and new observations. The estimator thereby adapts its parameters to improve accuracy on the new batch. Moreover, by adjusting the relative weights of the prior and likelihood terms, the extent of model update can be controlled to match the degree of distribution shift, achieving optimal calibration. The proposed approach is evaluated on a simulated operational amplifier and two commercial IC datasets to demonstrate its advantage. The BMC can improve accuracy on new batches with minimal additional samples. Moreover, by flexibly integrating the prior knowledge and new data, it outperforms baseline methods such as direct data combination (DC) and simple fine-tuning (SFT).
- 10 April 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Very Large Scale Integration (VLSI) Systems
- Vol. 34 (7) , 2274-2283
Bayesian Model Calibration for Deep-Learning-Based Indirect Test of Analog Circuits
The indirect test offers a cost-effective alternative for the specification test of analog integrated circuits (ICs). It infers high-cost specifications from low-cost measurements through a mapping function. Deep learning provides an efficient means for modeling this mapping. However, batch-to-batch variations in IC manufacturing often lead to distribution shifts, which degrade the accuracy of such data-driven techniques. To address this, we propose a transfer learning-based framework—Bayesian model calibration (BMC)—for the indirect test. During testing, a deep neural network (DNN) is trained with early data for estimation. When a new-batch arrives, a small number of samples are conventionally tested to form a calibration set. The original estimator is updated with the new dataset via a Bayesian loss function. This loss function integrates a prior term (reflecting the early data) and a likelihood term (reflecting the new data), enabling a principled tradeoff between prior knowledge and new observations. The estimator thereby adapts its parameters to improve accuracy on the new batch. Moreover, by adjusting the relative weights of the prior and likelihood terms, the extent of model update can be controlled to match the degree of distribution shift, achieving optimal calibration. The proposed approach is evaluated on a simulated operational amplifier and two commercial IC datasets to demonstrate its advantage. The BMC can improve accuracy on new batches with minimal additional samples. Moreover, by flexibly integrating the prior knowledge and new data, it outperforms baseline methods such as direct data combination (DC) and simple fine-tuning (SFT).
- 15 January 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Instrumentation and Measurement
- Vol. 75 (00189456) , 1-10
Adaptive Data Augmentation Based on Ensemble Learning for Indirect Test of Analog Circuits
The nonparametric indirect test offers a promising strategy to reduce the testing cost of analog integrated circuits (ICs). However, as a data-driven method, its performance is constrained by imbalanced and limited training data. To address this problem, we propose an ensemble-based adaptive data augmentation (ADA) method (E-ADA). The E-ADA reshapes the original dataset toward a more uniform distribution by generating synthetic samples. Through perturbation, the number of generated samples is adapted to local data density. Multiple augmented datasets are then used to train diverse machine learning models targeting the same specification, each with a unique training set. During inference, all submodels are executed in parallel, and their outputs are averaged to yield the final estimation. This strategy improves stability across sparse and dense data intervals by correcting distribution imbalance. Meanwhile, the proposed approach enhances robustness of estimation via ensemble averaging. Moreover, diversity among submodels is enhanced by diverse datasets, further enhancing stability. Experimental results on simulation data and commercial ICs demonstrate that E-ADA improves both estimation stability and accuracy.
- 15 January 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Instrumentation and Measurement
- Vol. 75 (00189456) , 1-10
Adaptive Data Augmentation Based on Ensemble Learning for Indirect Test of Analog Circuits
The nonparametric indirect test offers a promising strategy to reduce the testing cost of analog integrated circuits (ICs). However, as a data-driven method, its performance is constrained by imbalanced and limited training data. To address this problem, we propose an ensemble-based adaptive data augmentation (ADA) method (E-ADA). The E-ADA reshapes the original dataset toward a more uniform distribution by generating synthetic samples. Through perturbation, the number of generated samples is adapted to local data density. Multiple augmented datasets are then used to train diverse machine learning models targeting the same specification, each with a unique training set. During inference, all submodels are executed in parallel, and their outputs are averaged to yield the final estimation. This strategy improves stability across sparse and dense data intervals by correcting distribution imbalance. Meanwhile, the proposed approach enhances robustness of estimation via ensemble averaging. Moreover, diversity among submodels is enhanced by diverse datasets, further enhancing stability. Experimental results on simulation data and commercial ICs demonstrate that E-ADA improves both estimation stability and accuracy.
- 26 September 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Internet of Things Journal
- Vol. 12 (23) , 51516-51532
MIIF-Net: An Automatic Modulation Classifier Based on Multidimensional Images
Automatic modulation classification (AMC) plays a vital role in the Internet of Things (IoT) and cognitive radio (CR) to ensure communication security, but the AMC technology is suffering from low signal-to-noise ratio (SNR), nonideal defects in communication systems, and so forth. In order to suppress the impact of interferences on AMC and improve classification performance, we propose a novel multidimensional image-based AMC method: multidimensional image information fusion network (MIIF-Net). Based on the analysis of signal model, this work focuses on suppressing noise and carrier frequency offset (CFO) to improve the classification accuracy of AMC. Motivated by multidimensional feature strategy, modulation signals are converted and encoded into multidimensional images to introduce multidomain information and suppress CFO. From the perspective of image enhancement, a novel noise suppression method is proposed, which does not rely on any prior information about signals. In addition, the residual network (ResNet) is utilized as encoder to extract distinctive feature in different dimensions, and channel attention mechanism (CAM) is employed to fuse multidimensional features. The multilayer perceptron (MLP) is utilized for modulation classification in MIIF-Net. MIIF-Net is evaluated in public dataset. Simulation results demonstrate that the proposed MIIF-Net achieves superior performance, especially under low SNR levels. MIIF-Net is further evaluated in practical acquired data. The experimental result shows that MIIF-Net exhibits the consistent advantage under low SNR levels.
- 26 September 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Internet of Things Journal
- Vol. 12 (23) , 51516-51532
MIIF-Net: An Automatic Modulation Classifier Based on Multidimensional Images
Automatic modulation classification (AMC) plays a vital role in the Internet of Things (IoT) and cognitive radio (CR) to ensure communication security, but the AMC technology is suffering from low signal-to-noise ratio (SNR), nonideal defects in communication systems, and so forth. In order to suppress the impact of interferences on AMC and improve classification performance, we propose a novel multidimensional image-based AMC method: multidimensional image information fusion network (MIIF-Net). Based on the analysis of signal model, this work focuses on suppressing noise and carrier frequency offset (CFO) to improve the classification accuracy of AMC. Motivated by multidimensional feature strategy, modulation signals are converted and encoded into multidimensional images to introduce multidomain information and suppress CFO. From the perspective of image enhancement, a novel noise suppression method is proposed, which does not rely on any prior information about signals. In addition, the residual network (ResNet) is utilized as encoder to extract distinctive feature in different dimensions, and channel attention mechanism (CAM) is employed to fuse multidimensional features. The multilayer perceptron (MLP) is utilized for modulation classification in MIIF-Net. MIIF-Net is evaluated in public dataset. Simulation results demonstrate that the proposed MIIF-Net achieves superior performance, especially under low SNR levels. MIIF-Net is further evaluated in practical acquired data. The experimental result shows that MIIF-Net exhibits the consistent advantage under low SNR levels.
- 22 September 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Instrumentation and Measurement
- Vol. 74 (00189456) , 1-11
Data-Driven Soft Fault Diagnosis for Analog Circuits Based on Contrastive Learning
Analog circuits are fundamental to modern electrical systems, yet soft fault diagnosis of analog circuits remains challenging due to their nonlinear characteristics and the increasing scale of circuit integration. To address this issue, this article presents a novel data-driven fault diagnosis method utilizing contrastive learning (CL). In contrast to conventional data-driven approaches, the proposed method employs a self-supervised CL framework for representation learning, which captures informative features related to circuit faults. A temporal information-based encoder, leveraging 1-D convolutional neural networks (1D-CNNs) and a self-attention (SA) mechanism, is designed to enhance the extraction of latent patterns from raw time series data without relying on traditional signal processing techniques, thereby minimizing information loss and the need for expert knowledge. The encoder, trained using CL, is followed by a supervised two-layer neural network (NN) for fault classification. Simulation results on two benchmark filter circuits demonstrate the efficacy of the proposed method. Furthermore, the robustness of the encoder is validated under different faulty degrees, underscoring its reliability in various diagnostic scenarios.
- 22 September 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Instrumentation and Measurement
- Vol. 74 (00189456) , 1-11
Data-Driven Soft Fault Diagnosis for Analog Circuits Based on Contrastive Learning
Analog circuits are fundamental to modern electrical systems, yet soft fault diagnosis of analog circuits remains challenging due to their nonlinear characteristics and the increasing scale of circuit integration. To address this issue, this article presents a novel data-driven fault diagnosis method utilizing contrastive learning (CL). In contrast to conventional data-driven approaches, the proposed method employs a self-supervised CL framework for representation learning, which captures informative features related to circuit faults. A temporal information-based encoder, leveraging 1-D convolutional neural networks (1D-CNNs) and a self-attention (SA) mechanism, is designed to enhance the extraction of latent patterns from raw time series data without relying on traditional signal processing techniques, thereby minimizing information loss and the need for expert knowledge. The encoder, trained using CL, is followed by a supervised two-layer neural network (NN) for fault classification. Simulation results on two benchmark filter circuits demonstrate the efficacy of the proposed method. Furthermore, the robustness of the encoder is validated under different faulty degrees, underscoring its reliability in various diagnostic scenarios.
- 5 February 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Very Large Scale Integration (VLSI) Systems
- Vol. 33 (5) , 1224-1234
A Model Splitting Approach to Improve Reliability and Accuracy for Alternate Test of Analog/Mixed-Signal Circuits
Machine learning-based alternate test of analog/mixed-signal integrated circuits (ICs) has been widely studied in the last decade, which has the benefits of simplifying test equipment and decreasing test costs. However, due to low reliability and accuracy, it is hard to adopt the alternate test technique in the industry. In this article, a model splitting approach (MDSP approach) is proposed to improve the reliability and accuracy of the alternate test. The machine learning-based estimation model is “split” into two models with “complementary” performance (a “positive” model and a “negative” model). The “positive” model generates estimations that are no smaller than label values, while the “negative” model outputs estimations that are no larger than label values. Estimations with excessive differences between two models are identified as suspected estimations with large errors and filtered out. The rest results of “complementary” models are averaged to generate the final estimations. By comparing estimations of two models, the estimations with large error are filtered out effectively, and the estimation accuracy is improved significantly by fusing the results of two estimators. The MDSP approach is investigated with data from the commercial analog-to-digital converter and operational amplifier (OP). Results demonstrated that the proposed approach can improve test reliability and accuracy significantly.
- 5 February 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Very Large Scale Integration (VLSI) Systems
- Vol. 33 (5) , 1224-1234
A Model Splitting Approach to Improve Reliability and Accuracy for Alternate Test of Analog/Mixed-Signal Circuits
Machine learning-based alternate test of analog/mixed-signal integrated circuits (ICs) has been widely studied in the last decade, which has the benefits of simplifying test equipment and decreasing test costs. However, due to low reliability and accuracy, it is hard to adopt the alternate test technique in the industry. In this article, a model splitting approach (MDSP approach) is proposed to improve the reliability and accuracy of the alternate test. The machine learning-based estimation model is “split” into two models with “complementary” performance (a “positive” model and a “negative” model). The “positive” model generates estimations that are no smaller than label values, while the “negative” model outputs estimations that are no larger than label values. Estimations with excessive differences between two models are identified as suspected estimations with large errors and filtered out. The rest results of “complementary” models are averaged to generate the final estimations. By comparing estimations of two models, the estimations with large error are filtered out effectively, and the estimation accuracy is improved significantly by fusing the results of two estimators. The MDSP approach is investigated with data from the commercial analog-to-digital converter and operational amplifier (OP). Results demonstrated that the proposed approach can improve test reliability and accuracy significantly.