Z. A. 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
Research Subjects
Z. A. Wang has a total of 29 co-authors. Here are the top co-authors by number of shared publications:
Alessandro Cappelli
LightOn
1 shared publicationC. Y. Qiao
State Key Laboratory of Nuclear Physics and Technology, School of Physics, Peking University, Beijing 100871, China
1 shared publicationCharles Brossollet
LightOn
1 shared publicationFabien Moreau
LightOn
1 shared publication- Faguang Yan
State Key Laboratory for Superlattices and Microstructures, Institute of Semiconductors, Chinese Academy of Sciences 1 , Beijing 100083, China
1 shared publication - Florent Krzakala
École Polytechnique Fédérale de Lausanne (EPFL) IdePHICS Lab CH-1015 Lausanne Switzerland
1 shared publication
Latest Publications
- Open Access
- 15 May 2026
- Published by National Academy of Sciences in Proceedings of the National Academy of Sciences
- Vol. 123 (20)
Streamlined optical training of large-scale modern deep learning architectures with direct feedback alignment
Modern deep learning relies nearly exclusively on dedicated electronic hardware accelerators. Photonic approaches, with low consumption and high operation speed, are increasingly considered for inference but, to date, remain mostly limited to relatively basic tasks. Simultaneously, the problem of training deep and complex neural networks, overwhelmingly performed through backpropagation, remains a significant limitation to the size and, consequently, the performance of current architectures and a major compute and energy bottleneck. Here, we experimentally implement a versatile and scalable training algorithm, called direct feedback alignment, on a hybrid electronic-photonic platform. An optical processing unit performs large-scale random matrix multiplications, which is the central operation of this algorithm. We perform optical training of modern deep learning architectures, including Transformers, with more than 1B parameters, and obtain good performances on language, vision, and diffusion-based generative tasks. We study the scaling of the training time and demonstrate a potential advantage of our hybrid opto-electronic approach for ultra-deep and wide neural networks, thus opening a promising route to sustain the exponential growth of modern artificial intelligence beyond traditional von Neumann approaches.
- 15 May 2026
- Published by National Academy of Sciences in Proceedings of the National Academy of Sciences
- Vol. 123 (20)
Open AccessStreamlined optical training of large-scale modern deep learning architectures with direct feedback alignment
Modern deep learning relies nearly exclusively on dedicated electronic hardware accelerators. Photonic approaches, with low consumption and high operation speed, are increasingly considered for inference but, to date, remain mostly limited to relatively basic tasks. Simultaneously, the problem of training deep and complex neural networks, overwhelmingly performed through backpropagation, remains a significant limitation to the size and, consequently, the performance of current architectures and a major compute and energy bottleneck. Here, we experimentally implement a versatile and scalable training algorithm, called direct feedback alignment, on a hybrid electronic-photonic platform. An optical processing unit performs large-scale random matrix multiplications, which is the central operation of this algorithm. We perform optical training of modern deep learning architectures, including Transformers, with more than 1B parameters, and obtain good performances on language, vision, and diffusion-based generative tasks. We study the scaling of the training time and demonstrate a potential advantage of our hybrid opto-electronic approach for ultra-deep and wide neural networks, thus opening a promising route to sustain the exponential growth of modern artificial intelligence beyond traditional von Neumann approaches.
- 22 August 2022
- Published by American Physical Society (APS) in Physical Review C
- Vol. 106 (2) , L021304
Bayesian approach to heterogeneous data fusion of imperfect fission yields for augmented evaluations
We demonstrate that Bayesian machine learning can be used to treat the vast amount of experimental fission data which are noisy, incomplete, discrepant, and correlated. To supply the application needs, the two-dimensional cumulative fission yields (CFY) of neutron-induced fission of are evaluated for energy dependencies and uncertainty qualifications by cross-experiment data fusion. For independent fission yields (IFY) with very few experimental data, the heterogeneous data fusion of CFY and IFY is employed to interpolate the energy dependence. This work shows that Bayesian data fusion can facilitate the maximum utilization of imperfect raw nuclear data.- 22 August 2022
- Published by American Physical Society (APS) in Physical Review C
- Vol. 106 (2) , L021304
Bayesian approach to heterogeneous data fusion of imperfect fission yields for augmented evaluations
We demonstrate that Bayesian machine learning can be used to treat the vast amount of experimental fission data which are noisy, incomplete, discrepant, and correlated. To supply the application needs, the two-dimensional cumulative fission yields (CFY) of neutron-induced fission of are evaluated for energy dependencies and uncertainty qualifications by cross-experiment data fusion. For independent fission yields (IFY) with very few experimental data, the heterogeneous data fusion of CFY and IFY is employed to interpolate the energy dependence. This work shows that Bayesian data fusion can facilitate the maximum utilization of imperfect raw nuclear data.- Open Access
Selectively controlled ferromagnets by electric fields in van der Waals ferromagnetic heterojunctions
Charge transfer plays a key role at the interfaces of heterostructures, which can affect electronic structures and ultimately the physical properties of the materials. However, charge transfer is difficult to manipulate externally once the interface formed. Here, we report electrically tunable charge transfer in Fe3GeTe2/Cr2Ge2Te6/Fe3GeTe2 all-magnetic van der Waals heterostructures, which can be exploited to selectively modify the magnetic properties of the top or bottom Fe3GeTe2 electrodes. The directional charge transfer from metallic Fe3GeTe2 to semiconducting Cr2Ge2Te6 remarkably modifies magnetic anisotropy energy of Fe3GeTe2, leading to the dramatically suppressed coercivity. The electrically selective control of ferromagnets demonstrated in this study could stimulate the development of spintronic devices based on van der Waals magnets.
Open AccessSelectively controlled ferromagnets by electric fields in van der Waals ferromagnetic heterojunctions
Charge transfer plays a key role at the interfaces of heterostructures, which can affect electronic structures and ultimately the physical properties of the materials. However, charge transfer is difficult to manipulate externally once the interface formed. Here, we report electrically tunable charge transfer in Fe3GeTe2/Cr2Ge2Te6/Fe3GeTe2 all-magnetic van der Waals heterostructures, which can be exploited to selectively modify the magnetic properties of the top or bottom Fe3GeTe2 electrodes. The directional charge transfer from metallic Fe3GeTe2 to semiconducting Cr2Ge2Te6 remarkably modifies magnetic anisotropy energy of Fe3GeTe2, leading to the dramatically suppressed coercivity. The electrically selective control of ferromagnets demonstrated in this study could stimulate the development of spintronic devices based on van der Waals magnets.