贝叶斯优化
第五部分：案例研究
EN

案例研究

前四部分在测试函数上构建方法，所选的测试函数便于把每个想法展示清楚。本部分把这些方法用于真实问题，每章一个，并保留测试函数略去的细节：一次评估的实际代价、噪声的来源、实践者在第一次查询之前必须做出的选择，以及已发表的研究在实际问题上应用这些方法时有何发现。

其中两章使用实测数据，图中的优化器面对的是真实结果而非公式：一是分类器在真实数据集上的验证误差，二是一组经过完全枚举的化学反应的产率。另一章用按已发表研究校准的模拟行走者重演人在回路中的外骨骼调节，并明确说明哪些部分属于模拟。最后一章的问题之所以真实，是因为读者本人就是问题的一部分：读者通过比较不同版本来增强一张照片，方法则从中学习读者的眼光。

本部分以第三部分与第四部分为前提，各章可以独立阅读。

本部分各章

  1. 22 为分类器调参

    基于实测结果的贝叶斯优化。第一个问题是支持向量机在两个超参数上的验证误差地形：读者先亲自搜索，再观看网格搜索、随机搜索与贝叶斯搜索在同一地形上回放；第二个问题是梯度提升树模型七个超参数上的搜索记录，并给出代价、噪声与留出数据上的得分。

  2. 23 优化化学反应

    在已发表且全部实测的反应数据集上回放贝叶斯优化：配体、碱、溶剂、浓度与温度共 1,728 种组合，每种组合都有实测产率。读者先在同一数据上参与化学家们做过的优化游戏，再观察优化器在独热编码或描述符编码、不同批量大小下的表现，并与随机选择、50 位有记录的化学家以及论文本身的运行结果比较。

  3. 24 人在回路中调节外骨骼

    在按已发表研究校准的模拟行走者上重演人在回路的外骨骼调节：一次两分钟代谢估计的价值；一次调节会话，读者凭感觉调节装置，并在相同的行走时间内与贝叶斯优化、进化策略和偏好优化比较；以及仍在适应中的人对所有这些方法的影响。

  4. 25 通过比较增强照片

    在一张真实照片上以六项调整进行偏好贝叶斯优化：读者通过二选一或沿直线滑动来增强照片，模型从中学习读者的品味；本章还将一次会话所揭示的现象（一致性、漂移、对提问顺序的依赖）与已发表的研究相对照。

第五部分参考文献

本部分各章共引用 60 篇文献。

  1. Abram, S. J., Poggensee, K. L., Sánchez, N., Simha, S. N., Finley, J. M., Collins, S. H., and Donelan, J. M. (2022). General variability leads to specific adaptation toward optimal movement policies. Current Biology. 第 24 章
  2. Adesiji, A. D., Wang, J., Kuo, C.-S., and Brown, K. A. (2026). Benchmarking self-driving labs. Digital Discovery. 第 23 章
  3. Alpaydin, E., and Kaynak, C. (1998). Optical Recognition of Handwritten Digits. UCI Machine Learning Repository, data set, CC BY 4.0. doi:10.24432/C50P49. 非同行评审 第 22 章
  4. Bergstra, J., and Bengio, Y. (2012). Random Search for Hyper-Parameter Optimization. Journal of Machine Learning Research. 第 22 章
  5. Cawley, G. C., and Talbot, N. L. C. (2010). On Over-fitting in Model Selection and Subsequent Selection Bias in Performance Evaluation. Journal of Machine Learning Research. 第 22 章
  6. Chang, S., Kim, C.-Y., and Cho, Y. S. (2017). Sequential effects in preference decision: Prior preference assimilates current preference. PLOS ONE. 第 25 章
  7. Chong, T. L. H., Shen, I.-C., Sato, I., and Igarashi, T. (2021). Interactive Optimization of Generative Image Modelling using Sequential Subspace Search and Content-based Guidance. Computer Graphics Forum. doi:10.1111/cgf.14188. 第 25 章
  8. Clites, T. R., Shepherd, M. K., Ingraham, K. A., Wontorcik, L., and Rouse, E. J. (2021). Understanding patient preference in prosthetic ankle stiffness. Journal of NeuroEngineering and Rehabilitation. 第 24 章
  9. Cortes, C., and Vapnik, V. (1995). Support-Vector Networks. Machine Learning. 第 22 章
  10. Deneault, J. R., Kim, W., Kim, J., Gu, Y., Chang, J., Maruyama, B., Myung, J. I., and Pitt, M. A. (2025). Preferential Bayesian optimization improves the efficiency of printing objects with subjective qualities. Digital Discovery. 第 23 章
  11. Díaz, M. A., Nuzzo, S., Flynn, L., Beckerle, P., Verstraten, T., and De Pauw, K. (2026). User preference in the personalized control of an ankle prosthesis: a case study. Journal of NeuroEngineering and Rehabilitation. doi:10.1186/s12984-026-01931-w. 第 24 章
  12. Ding, Y., Kim, M., Kuindersma, S., and Walsh, C. J. (2018). Human-in-the-Loop Optimization of Hip Assistance with a Soft Exosuit during Walking. Science Robotics. 第 24 章
  13. Feurer, M., and Hutter, F. (2019). Hyperparameter Optimization. Automated Machine Learning: Methods, Systems, Challenges. 第 22 章
  14. Friedman, J. H. (2001). Greedy Function Approximation: A Gradient Boosting Machine. The Annals of Statistics. 第 22 章
  15. Ginsbourger, D., Le Riche, R., and Carraro, L. (2010). Kriging Is Well-Suited to Parallelize Optimization. Computational Intelligence in Expensive Optimization Problems. 第 23 章
  16. Gupta, R., Hartford, J., and Liu, B. (2025). LLMs for Bayesian Optimization in Scientific Domains: Are We There Yet? Findings of the Association for Computational Linguistics: EMNLP 2025. 第 23 章
  17. Hansen, N., and Ostermeier, A. (2001). Completely Derandomized Self-Adaptation in Evolution Strategies. Evolutionary Computation. 第 24 章
  18. Hopkins, M., Reeber, E., Forman, G., and Suermondt, J. (1999). Spambase. UCI Machine Learning Repository, data set, CC BY 4.0. doi:10.24432/C53G6X. 非同行评审 第 22 章
  19. Hsu, C.-W., Chang, C.-C., and Lin, C.-J. (2003). A Practical Guide to Support Vector Classification. Department of Computer Science, National Taiwan University. 非同行评审 第 22 章
  20. Hutter, F., Hoos, H., and Leyton-Brown, K. (2014). An Efficient Approach for Assessing Hyperparameter Importance. International Conference on Machine Learning. 第 22 章
  21. Ingraham, K. A., Remy, C. D., and Rouse, E. J. (2022). The role of user preference in the customized control of robotic exoskeletons. Science Robotics. 第 24 章
  22. Iwai, K., Kumagae, Y., Koyama, Y., Hamasaki, M., and Goto, M. (2025). Constrained Preferential Bayesian Optimization and Its Application in Banner Ad Design. Proceedings of the Thirty-Fourth International Joint Conference on Artificial Intelligence. 第 25 章
  23. Kalinin, S. V., Liu, Y., Biswas, A., Duscher, G., Pratiush, U., Roccapriore, K., Ziatdinov, M., and Vasudevan, R. (2024). Human-in-the-loop: The future of Machine Learning in Automated Electron Microscopy. Microscopy Today. doi:10.1093/mictod/qaad096. 第 23 章
  24. Keerthi, S. S., and Lin, C.-J. (2003). Asymptotic Behaviors of Support Vector Machines with Gaussian Kernel. Neural Computation. 第 22 章
  25. Kim, G., and Sergi, F. (2026). Validation of Dynamic Bayesian Optimization for Human-in-the-Loop Optimization of Exoskeleton Control at User-Driven Walking Speed. bioRxiv. 预印本 第 24 章
  26. Klein, A., Falkner, S., Bartels, S., Hennig, P., and Hutter, F. (2017). Fast Bayesian Optimization of Machine Learning Hyperparameters on Large Datasets. Artificial Intelligence and Statistics. 第 22 章
  27. Koyama, Y., and Igarashi, T. (2018). Computational Design with Crowds. Computational Interaction. 第 25 章
  28. Koyama, Y., Sato, I., Sakamoto, D., and Igarashi, T. (2017). Sequential line search for efficient visual design optimization by crowds. ACM Transactions on Graphics. 第 25 章
  29. Koyama, Y., Sato, I., and Goto, M. (2020). Sequential Gallery for Interactive Visual Design Optimization. ACM Transactions on Graphics 39(4) (SIGGRAPH 2020). 第 25 章
  30. Kutulakos, Z., and Slade, P. (2024). Simulating human-in-the-loop optimization of exoskeleton assistance to compare optimization algorithm performance. bioRxiv. 预印本 第 24 章
  31. Lee, U. H., Shetty, V. S., Franks, P. W., Tan, J., Evangelopoulos, G., Ha, S., and Rouse, E. J. (2023). User preference optimization for control of ankle exoskeletons using sample efficient active learning. Science Robotics. 第 24 章
  32. Li, L., Jamieson, K., DeSalvo, G., Rostamizadeh, A., and Talwalkar, A. (2018). Hyperband: A Novel Bandit-Based Approach to Hyperparameter Optimization. Journal of Machine Learning Research. 第 22 章
  33. Liu, C., Ling, S., and Jacobson, A. (2026b). GimmBO: Interactive Generative Image Model Merging via Bayesian Optimization. ACM Transactions on Graphics. doi:10.1145/3811293. 第 25 章
  34. Liu, X.-Y., Li, G., Wang, W., and Hou, Z.-G. (2026d). Personalized Lower-limb Exoskeleton Assistance via Preference-based Bayesian Optimization. arXiv. 预印本 第 24 章
  35. Maberry, A., and Martin, A. E. (2026). Just Noticeable Difference of Impedance Parameters While Walking in an Ankle Exoskeleton. IEEE Transactions on Neural Systems and Rehabilitation Engineering. 第 24 章
  36. Mikkola, P., Todorović, M., Järvi, J., Rinke, P., and Kaski, S. (2020). Projective Preferential Bayesian Optimization. International Conference on Machine Learning. 第 25 章
  37. Ou, C., Buschek, D., Mayer, S., and Butz, A. (2022). The Human in the Infinite Loop: A Case Study on Revealing and Explaining Human-AI Interaction Loop Failures. Mensch und Computer 2022. 第 25 章
  38. Ou, C., Mayer, S., and Butz, A. (2023). The Impact of Expertise in the Loop for Exploring Machine Rationality. IUI 2023. 第 25 章
  39. Pedregosa, F., Varoquaux, G., Gramfort, A., Michel, V., Thirion, B., Grisel, O., … Duchesnay, É. (2011). Scikit-learn: Machine Learning in Python. Journal of Machine Learning Research. 第 22 章
  40. Poggensee, K. L., and Collins, S. H. (2021). How adaptation, training, and customization contribute to benefits from exoskeleton assistance. Science Robotics. 第 24 章
  41. Pombo, M., Brielmann, A. A., and Pelli, D. G. (2023). The intrinsic variance of beauty judgment. Attention, Perception, & Psychophysics. 第 25 章
  42. Pratiush, U., Roccapriore, K. M., Liu, Y., Duscher, G., Ziatdinov, M., and Kalinin, S. V. (2025). Building Workflows for Interactive Human in the Loop Automated Experiment (hAE) in STEM-EELS. Digital Discovery. doi:10.1039/d5dd00033e. 第 23 章
  43. Probst, P., Boulesteix, A.-L., and Bischl, B. (2019). Tunability: Importance of Hyperparameters of Machine Learning Algorithms. Journal of Machine Learning Research. 第 22 章
  44. Ranković, B., Griffiths, R.-R., and Schwaller, P. (2026). Large language models as uncertainty-calibrated optimizers for experimental discovery. Nature Machine Intelligence. doi:10.1038/s42256-026-01283-z. 第 23 章
  45. Schäfer, N., Zhao, G., Li, B., Kupnik, M., Seyfarth, A., Beckerle, P., and Grimmer, M. (2026). User preference-based human-in-the-loop tuning of exoskeleton assistance during walking. npj Biomedical Innovations. doi:10.1038/s44385-026-00085-7. 第 24 章
  46. Schoinas, E., Rastogi, A., Carter, A., Granley, J., and Beyeler, M. (2025). Evaluating Deep Human-in-the-Loop Optimization for Retinal Implants Using Sighted Participants. 2025 47th Annual International Conference of the IEEE Engineering in Medicine and Biology Society (EMBC). doi:10.1109/embc58623.2025.11253762. 第 25 章
  47. Selinger, J. C., and Donelan, J. M. (2014). Estimating instantaneous energetic cost during non-steady-state gait. Journal of Applied Physiology. 第 24 章
  48. Shields, B. J. (2021). EDBO: Experimental Design via Bayesian Optimization. GitHub repository, MIT License; direct arylation data in experiments/data/direct_arylation, commit 9b41eac. 软件 第 23 章
  49. Shields, B. J., and Li, J. (2020). EvML: Expert versus Machine Learning. GitHub repository, MIT License; reaction optimization game records and analysis, commit 9fb4655. 软件 第 23 章
  50. Shields, B. J., Stevens, J., Li, J., Parasram, M., Damani, F., Alvarado, J. I. M., … Doyle, A. G. (2021). Bayesian reaction optimization as a tool for chemical synthesis. Nature. 第 23 章
  51. Siivola, E., Vehtari, A., Vanhatalo, J., González, J., and Andersen, M. R. (2018). Correcting Boundary Over-Exploration Deficiencies in Bayesian Optimization with Virtual Derivative Sign Observations. 2018 IEEE 28th International Workshop on Machine Learning for Signal Processing (MLSP). 第 22 章
  52. Slade, P., Kochenderfer, M. J., Delp, S. L., and Collins, S. H. (2022). Personalizing exoskeleton assistance while walking in the real world. Nature. 第 24 章
  53. Snoek, J., Larochelle, H., and Adams, R. P. (2012). Practical Bayesian Optimization of Machine Learning Algorithms. Advances in Neural Information Processing Systems 25 (NeurIPS 2012). 第 22 章
  54. Tucker, M., Cheng, M., Novoseller, E., Cheng, R., Yue, Y., Burdick, J. W., and Ames, A. D. (2020a). Human Preference-Based Learning for High-dimensional Optimization of Exoskeleton Walking Gaits. IROS 2020. 第 24 章
  55. Tucker, M., Novoseller, E., Kann, C., Sui, Y., Yue, Y., Burdick, J. W., and Ames, A. D. (2020b). Preference-Based Learning for Exoskeleton Gait Optimization. 2020 IEEE International Conference on Robotics and Automation (ICRA). 第 24 章
  56. Turner, R., Eriksson, D., McCourt, M., Kiili, J., Laaksonen, E., Xu, Z., and Guyon, I. (2021). Bayesian Optimization is Superior to Random Search for Machine Learning Hyperparameter Tuning: Analysis of the Black-Box Optimization Challenge 2020. NeurIPS 2020 Competition and Demonstration Track. 第 22 章
  57. Xie, Q., Astudillo, R., Frazier, P. I., Scully, Z., and Terenin, A. (2024). Cost-aware Bayesian Optimization via the Pandora's Box Gittins Index. NeurIPS 2024. 第 22 章
  58. Xie, Q., Cai, L., Terenin, A., Frazier, P. I., and Scully, Z. (2026). Cost-aware Stopping for Bayesian Optimization. International Conference on Machine Learning. 第 22 章
  59. Zhang, J., Fiers, P., Witte, K. A., Jackson, R. W., Poggensee, K. L., Atkeson, C. G., and Collins, S. H. (2017). Human-in-the-loop optimization of exoskeleton assistance during walking. Science. 第 24 章
  60. Zhang, R., Zhu, X., Pourebadi Khotbehsara, M., Dao, W., Bıyık, E., and Culbertson, H. (2026b). Vibrotactile Preference Learning: Uncertainty-Aware Preference Learning for Personalized Vibration Feedback. UMAP 2026 (per Semantic Scholar). 第 25 章