Using autoencoders and deep transfer learning to determine the stellar parameters of 286 CARMENES M dwarfs (2405.08703v1)

Abstract: Deep learning (DL) techniques are a promising approach among the set of methods used in the ever-challenging determination of stellar parameters in M dwarfs. In this context, transfer learning could play an important role in mitigating uncertainties in the results due to the synthetic gap (i.e. difference in feature distributions between observed and synthetic data). We propose a feature-based deep transfer learning (DTL) approach based on autoencoders to determine stellar parameters from high-resolution spectra. Using this methodology, we provide new estimations for the effective temperature, surface gravity, metallicity, and projected rotational velocity for 286 M dwarfs observed by the CARMENES survey. Using autoencoder architectures, we projected synthetic PHOENIX-ACES spectra and observed CARMENES spectra onto a new feature space of lower dimensionality in which the differences between the two domains are reduced. We used this low-dimensional new feature space as input for a convolutional neural network to obtain the stellar parameter determinations. We performed an extensive analysis of our estimated stellar parameters, ranging from 3050 to 4300 K, 4.7 to 5.1 dex, and -0.53 to 0.25 dex for Teff, logg, and [Fe/H], respectively. Our results are broadly consistent with those of recent studies using CARMENES data, with a systematic deviation in our Teff scale towards hotter values for estimations above 3750 K. Furthermore, our methodology mitigates the deviations in metallicity found in previous DL techniques due to the synthetic gap. We consolidated a DTL-based methodology to determine stellar parameters in M dwarfs from synthetic spectra, with no need for high-quality measurements involved in the knowledge transfer. These results suggest the great potential of DTL to mitigate the differences in feature distributions between the observations and the PHOENIX-ACES spectra.

• The paper employs autoencoders and deep transfer learning to enhance stellar parameter estimation for M dwarfs.
• Its methodology integrates feature extraction from synthetic spectra with fine-tuning on observed CARMENES data to bridge the synthetic gap.
• Accurate predictions of temperature, gravity, and metallicity for 286 M dwarfs offer promising applications in astronomy and exoplanet studies.

Estimating Stellar Parameters of M Dwarfs Using Autoencoders and Transfer Learning

Overview

The task of determining stellar parameters of M dwarfs—such as effective temperature ($T_{\rm eff}$), surface gravity (log\,g), and metallicity ([M/H])—is crucial yet challenging due to their inherent faintness and strong magnetic activity. This paper proposes a clever solution employing deep learning techniques, specifically autoencoders and deep transfer learning (DTL), to tackle this problem.

Methodology

Data Preparation

The paper uses high-resolution spectra from both observed CARMENES data and synthetic PHOENIX-ACES models. The CARMENES survey observes M dwarfs in both visible and near-infrared (NIR) ranges, which are then corrected and preprocessed to ensure high-quality spectra.

Autoencoders for Feature Extraction

Autoencoders (AEs) are a type of neural network used for dimensionality reduction. Here’s a quick breakdown of their application in this paper:

• An autoencoder compresses the input spectra into a lower-dimensional representation, known as the latent space.
• Initially trained on synthetic PHOENIX-ACES spectra, the autoencoder extracts relevant spectral features.

Deep Transfer Learning

To address the differences between synthetic and observed spectra (known as the synthetic gap), the authors use a fine-tuning approach in DTL:

• The autoencoder network, initially trained on synthetic spectra, is fine-tuned using observed CARMENES spectra.
• This step significantly reduces discrepancies, bringing synthetic and observed data closer in the latent space, making it effective for further analysis.

Convolutional Neural Networks (CNNs) for Parameter Estimation

For predicting the stellar parameters, the paper employs one-dimensional CNNs:

• The compressed representations from the autoencoder serve as input to CNNs.
• Multiple CNN models are trained to independently estimate each stellar parameter, resulting in robust predictions.

Numerical Results

The proposed methodology shows promising results in estimating stellar parameters for 286 CARMENES M dwarfs:

• Effective temperatures range from 3050 to 4300 K.
• Surface gravity values range between 4.7 and 5.1 dex.
• Metallicities vary from -0.53 to 0.25 dex.

Comparisons with recent studies reveal a good correlation, noting a systematic deviation toward hotter effective temperatures above 3750 K. This suggests that while the methodology is effective, further improvements can be made, particularly for higher temperature estimations.

Discussion and Implications

Practical Applications

• Astronomy and Exoplanet Studies: Accurate determination of stellar parameters is key for exoplanet detection and characterization.
• Astronomical Surveys: Large surveys like CARMENES can benefit from automated tools to process vast amounts of data efficiently.

Theoretical Implications

• Feature Extraction: The success of autoencoders in capturing relevant features could lead to broader applications in other astronomical datasets.
• Mitigating Synthetic Gaps: This paper’s approach to addressing discrepancies between synthetic and observed data could inspire similar techniques in other fields facing analogous challenges.

Future Directions

The paper highlights potential areas for future research:

• Larger and More Diverse Data: Increasing the sample size and diversity could help refine the models further.
• Improved Synthetic Models: Enhancing synthetic spectra to better mimic observed spectra will reduce the synthetic gap.
• Advanced Neural Networks: Exploring other types of neural networks and architectures could boost performance further.

Conclusion

This paper presents an innovative application of autoencoders and deep transfer learning for determining the stellar parameters of M dwarfs. By fine-tuning autoencoder models with observed data and employing CNNs for parameter prediction, the paper makes significant strides in addressing challenges posed by the synthetic gap. This methodology not only demonstrates effective performance but also opens up new avenues for utilizing deep learning in astronomical research. As AI continues to evolve, its integration into astrophysics promises to yield even more sophisticated and accurate tools for exploring the universe.