Advanced Photonics Nexus, Volume. 4, Issue 4, 046016(2025)
Active learning–augmented end-to-end modeling toward fast inverse design in chirped pulse amplification
Fig. 2. (a) Proposed E2E model for simulating the CPA system. (b) Workflow of the active learning strategy.
Fig. 3. Output pulses of the E2E-AL model and SSFM-RK4. (a)–(c) Comparison of the intensity and phase in the spectral domain. (d)–(f) Comparison of the intensity and phase in the temporal domain. (g)–(i) Comparison of the intensity and phase in the temporal domain after fine-tuning compression. (j)–(l) Comparison of the intensity and phase in the spectral domain after fine-tuning compression.
Fig. 4. Intensity NRMSE and time consumption comparison between the SSFM-RK4 and E2E-AL model.
Fig. 5. Results of the active learning. The NRMSE of the E2E (red solid line) and E2E-AL (blue solid line) models under different train samples. The active learning strategy reduces the number of training samples as indicated by the yellow dashed line.
Fig. 7. Inverse design comparison between the SSFM-RK4 and E2E-AL under different target pulses. (a) and (b) Single-objective inverse design of the pulse duration. (c) and (d) Multi-objective inverse design of the pulse duration and energy. (a) and (c) Temporal intensity comparison among the target pulse, the result of the SSFM-RK4 and GA, and the result of the E2E-AL and GA. (b) and (d) Error comparison between the most matching pulse of the SSFM-RK4 and GA and the E2E-AL and GA with the target pulse in each generation.
|
|
|
Get Citation
Copy Citation Text
Helin Jiang, Guoqing Pu, Xinyi Ma, Weisheng Hu, Lilin Yi, "Active learning–augmented end-to-end modeling toward fast inverse design in chirped pulse amplification," Adv. Photon. Nexus 4, 046016 (2025)
Category: Research Articles
Received: Dec. 17, 2024
Accepted: May. 27, 2025
Published Online: Aug. 11, 2025
The Author Email: Guoqing Pu (teddyghf1994@sjtu.edu.cn), Lilin Yi (lilinyi@sjtu.edu.cn)