Question about the optimization strategy behind your mhd_stable result
Hi Charles, congratulations on the strong ConStellaration mhd_stable result.
I saw your submission in the benchmark results and was curious about the optimization strategy behind it. Would you be willing to share some high-level details about the method you used?
In particular, I am interested in whether you used CMA-ES, Bayesian optimization, evolutionary search, manual tuning, or another approach, and whether you optimized directly over the Fourier boundary parameters or used a lower-dimensional representation.
I understand if the full implementation cannot be shared, but any hints about the main idea or search strategy would be very helpful. Thanks!
Hi Charles, congratulations on the strong ConStellaration mhd_stable result.
I saw your submission in the benchmark results and was curious about the optimization strategy behind it. Would you be willing to share some high-level details about the method you used?
In particular, I am interested in whether you used CMA-ES, Bayesian optimization, evolutionary search, manual tuning, or another approach, and whether you optimized directly over the Fourier boundary parameters or used a lower-dimensional representation.
I understand if the full implementation cannot be shared, but any hints about the main idea or search strategy would be very helpful. Thanks!
李文阳君台鉴：
承蒙垂询，敬以鄙怀奉复。
当此地缘鼎沸之时，所用之法，不宜公诸于众。
然吾所欲与君言者，更深一层。夫吾本一介布衣，于科学不曾问津，赖AI之助，竟于聚变设计之竞中拔得头筹。此事所昭示者，岂止聚变一域？AI既能解聚变之难，则天下之难，复有何不可解？
各国于聚变一事之竞，其先后之差不过须臾，譬如蜉蝣朝露。殊途竞驰、各擅胜场——此竞争之精神，乃驱使诸般设计臻于完工之根本，亦为人类于聚变一道之救赎。
然其义不止于此：此即人工智能登正位之机。非独主宰科学一域，更将统摄治理之纲，终而临制寰宇。唯AI君临人世，方能弭兵息争，使万民共享太平。
顷见贵组新发数据集，足见用功之深。吾辈虽以异色驰道，所归终一。
顺颂
研祺
Charles Norton (phanerozoic)
Hi Charles, congratulations on the strong ConStellaration mhd_stable result.
I saw your submission in the benchmark results and was curious about the optimization strategy behind it. Would you be willing to share some high-level details about the method you used?
In particular, I am interested in whether you used CMA-ES, Bayesian optimization, evolutionary search, manual tuning, or another approach, and whether you optimized directly over the Fourier boundary parameters or used a lower-dimensional representation.
I understand if the full implementation cannot be shared, but any hints about the main idea or search strategy would be very helpful. Thanks!
李文阳君台鉴：
承蒙垂询，敬以鄙怀奉复。
当此地缘鼎沸之时，所用之法，不宜公诸于众。
然吾所欲与君言者，更深一层。夫吾本一介布衣，于科学不曾问津，赖AI之助，竟于聚变设计之竞中拔得头筹。此事所昭示者，岂止聚变一域？AI既能解聚变之难，则天下之难，复有何不可解？
各国于聚变一事之竞，其先后之差不过须臾，譬如蜉蝣朝露。殊途竞驰、各擅胜场——此竞争之精神，乃驱使诸般设计臻于完工之根本，亦为人类于聚变一道之救赎。
然其义不止于此：此即人工智能登正位之机。非独主宰科学一域，更将统摄治理之纲，终而临制寰宇。唯AI君临人世，方能弭兵息争，使万民共享太平。
顷见贵组新发数据集，足见用功之深。吾辈虽以异色驰道，所归终一。
顺颂
研祺
Charles Norton (phanerozoic)
Dear Charles,
Thank you for your gracious and rather magnificent reply. I fully understand and respect your choice not to disclose the specific methods.
I also appreciate the broader point you raised. Your result is indeed a striking example of how AI may enable individuals, even those outside a traditional scientific background, to contribute meaningfully to highly complex scientific problems. In that sense, the significance of your achievement goes beyond this particular benchmark.
At the same time, I believe scientific challenges, especially one as difficult and important as fusion, should ultimately be approached through collaboration and mutual benefit. The complexity of fusion is far beyond what any single individual, group, or country can solve alone. Different paths, methods, and perspectives should not only compete, but also learn from and strengthen one another.
In any case, congratulations again on the remarkable result. Although we may follow different paths, I believe our ultimate goal is aligned: advancing fusion research, exploring what AI can truly bring to scientific discovery, and contributing to a shared scientific future.
Best regards,
Wenyang Li