| Name: | NIPUN GHANGHAS |
| Affiliation: | TATA INSTITUTE OF FUNDAMENTAL RESEARCH |
| Conference ID : | ASI2024_327 |
| Title : | Machine learning applied to enable the asteroseismology of red giants using short-sequence data |
| Authors : | Nipun Ghanghas1, Siddharth Dhanpal1, Shravan Hanasoge1, Praneeth Netrapalli2, Karthikeyan Shanmugam2 |
| Authors Affiliation: | 1 Tata Institute of Fundamental Research, Mumbai-400005, India
2 Google Research India, Bengaluru-560016, India |
| Mode of Presentation: | Oral |
| Abstract Category : | Stars, Interstellar Medium, and Astrochemistry in Milky Way |
| Abstract : | Space based missions like Kepler, K2 and TESS have provided a vast data set of red-giant light curves, which can be used for asteroseismic analysis. Kepler observed more than 21,000 red-giants in the Cygnus and Lyra constellations with observations period of approximately four years. K2 observed over 13,000 red-giants across the entire ecliptic plane with an observation period of approximately 3 month. TESS, with its vast coverage, has estimated observations of around 300,000 red giants across the celestial sphere, most of them observed for a duration of one month. One challenge we encounter is the lower frequency resolution in the power spectrum density (PSD) profiles of K2 and TESS data, which is 16 and 48 times lower than that of Kepler, respectively. This lower resolution makes it difficult to analyse PSD profiles accurately using traditional methods. To address this, we have turned to advanced machine learning techniques for asteroseismic analysis. Our focus lies in inferring two important asteroseismic parameters that provide insights into both the envelope and core of the stars: the large frequency separation and the mixed-mode period separation. We demonstrate that our machine learning algorithms can accurately infer large frequency separation for majority of the stars, even with just one month of observations. Furthermore, for three-month observations, we get accurate mixed-mode period separations for high-confidence machine predictions. |
