| 研究課題/領域番号 |
20K14706
|
|---|
| 研究機関 | 国立天文台 |
|---|
研究代表者 |
Wu Benjamin 国立天文台, アルマプロジェクト, 特別客員研究員 (50868718)
|
|---|
| 研究期間 (年度) |
2020-04-01 – 2024-03-31
|
| キーワード | machine learning / computer vision / deep learning / astronomy / interferometry / numerical simulation |
|---|
| 研究実績の概要 |
We are exploring machine learning applications for space exploration and research and establishing deep learning computational hardware capabilities at NAOJ. We have installed new GPU server capabilities for the ALMA Project at NAOJ which can be used for accelerated machine learning.
We applied modern deep learning to two new directions: (1) astronomical interferometric imaging and (2) physics fluid simulations. This resulted in the publication of 2 papers at top machine learning conferences (NeurIPS, AAAI).
More specifically, we developed a new "neural interferometry" imaging method using neural implicit representations. We also incorporated signal processing techniques into physics-informed neural networks, greatly improving simulation accuracy and speed.
|
| 現在までの達成度 (区分) |
現在までの達成度 (区分)
2: おおむね順調に進展している
理由
The GPU server upgrades have proceeded smoothly.
For rover localization, data generation was completed and initial convolutional neural networks were prototyped. However, we pivoted on the research topic due to new opportunities and collaborators, pursuing deep learning applications in fluid simulations (physics-informed neural networks) and also astronomical imaging (neural interferometry). This led to two new papers published at high-impact machine learning conferences. Due to this reprioritization, our original project on rover localization was delayed.
|
| 今後の研究の推進方策 |
I plan to continue testing different neural network architectures for the rover localization project on our finalized datasets. We may include newer techniques such as vision transformers, inspired from our recent publications.
If new opportunities arise to make impacts in adjacent fields such as physics simulations or astronomical imaging, we will also continue to expand upon those models.
|
