Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Hunting for "Oddballs" with Machine Learning: Detecting Anomalous Exoplanets Using a Deep-Learned Low-Dimensional Representation of Transit Spectra with Autoencoders

Alexander Roman, Emilie Panek, Roy T. Forestano, Eyup B. Unlu, Katia Matcheva, Konstantin T. Matchev

TL;DR

This study tackles detecting anomalous exoplanet atmospheres in large spectral datasets by combining autoencoder-driven latent representations with multiple anomaly detectors (reconstruction loss, one-class SVM, K-means, LOF). The authors show that performing anomaly detection in latent space consistently outperforms analyses in the raw spectral space, with K-means in latent space delivering the most robust performance across noise levels up to 50 ppm. This approach enables fast, scalable pre-screening for large surveys like Ariel, reducing computational demands while preserving sensitivity to CO2-rich outliers. Limitations include focusing on a single anomaly type and simplified atmospheric physics; extending to additional molecules and more realistic clouds/hazes will be important future directions.

Abstract

This study explores the application of autoencoder-based machine learning techniques for anomaly detection to identify exoplanet atmospheres with unconventional chemical signatures using a low-dimensional data representation. We use the Atmospheric Big Challenge (ABC) database, a publicly available dataset with over 100,000 simulated exoplanet spectra, to construct an anomaly detection scenario by defining CO2-rich atmospheres as anomalies and CO2-poor atmospheres as the normal class. We benchmarked four different anomaly detection strategies: Autoencoder Reconstruction Loss, One-Class Support Vector Machine (1 class-SVM), K-means Clustering, and Local Outlier Factor (LOF). Each method was evaluated in both the original spectral space and the autoencoder's latent space using Receiver Operating Characteristic (ROC) curves and Area Under the Curve (AUC) metrics. To test the performance of the different methods under realistic conditions, we introduced Gaussian noise levels ranging from 10 to 50 ppm. Our results indicate that anomaly detection is consistently more effective when performed within the latent space across all noise levels. Specifically, K-means clustering in the latent space emerged as a stable and high-performing method. We demonstrate that this anomaly detection approach is robust to noise levels up to 30 ppm (consistent with realistic space-based observations) and remains viable even at 50 ppm when leveraging latent space representations. On the other hand, the performance of the anomaly detection methods applied directly in the raw spectral space degrades significantly with increasing the level of noise. This suggests that autoencoder-driven dimensionality reduction offers a robust methodology for flagging chemically anomalous targets in large-scale surveys where exhaustive retrievals are computationally prohibitive.

Hunting for "Oddballs" with Machine Learning: Detecting Anomalous Exoplanets Using a Deep-Learned Low-Dimensional Representation of Transit Spectra with Autoencoders

TL;DR

This study tackles detecting anomalous exoplanet atmospheres in large spectral datasets by combining autoencoder-driven latent representations with multiple anomaly detectors (reconstruction loss, one-class SVM, K-means, LOF). The authors show that performing anomaly detection in latent space consistently outperforms analyses in the raw spectral space, with K-means in latent space delivering the most robust performance across noise levels up to 50 ppm. This approach enables fast, scalable pre-screening for large surveys like Ariel, reducing computational demands while preserving sensitivity to CO2-rich outliers. Limitations include focusing on a single anomaly type and simplified atmospheric physics; extending to additional molecules and more realistic clouds/hazes will be important future directions.

Abstract

This study explores the application of autoencoder-based machine learning techniques for anomaly detection to identify exoplanet atmospheres with unconventional chemical signatures using a low-dimensional data representation. We use the Atmospheric Big Challenge (ABC) database, a publicly available dataset with over 100,000 simulated exoplanet spectra, to construct an anomaly detection scenario by defining CO2-rich atmospheres as anomalies and CO2-poor atmospheres as the normal class. We benchmarked four different anomaly detection strategies: Autoencoder Reconstruction Loss, One-Class Support Vector Machine (1 class-SVM), K-means Clustering, and Local Outlier Factor (LOF). Each method was evaluated in both the original spectral space and the autoencoder's latent space using Receiver Operating Characteristic (ROC) curves and Area Under the Curve (AUC) metrics. To test the performance of the different methods under realistic conditions, we introduced Gaussian noise levels ranging from 10 to 50 ppm. Our results indicate that anomaly detection is consistently more effective when performed within the latent space across all noise levels. Specifically, K-means clustering in the latent space emerged as a stable and high-performing method. We demonstrate that this anomaly detection approach is robust to noise levels up to 30 ppm (consistent with realistic space-based observations) and remains viable even at 50 ppm when leveraging latent space representations. On the other hand, the performance of the anomaly detection methods applied directly in the raw spectral space degrades significantly with increasing the level of noise. This suggests that autoencoder-driven dimensionality reduction offers a robust methodology for flagging chemically anomalous targets in large-scale surveys where exhaustive retrievals are computationally prohibitive.
Paper Structure (12 sections, 2 equations, 11 figures, 2 tables)

This paper contains 12 sections, 2 equations, 11 figures, 2 tables.

Table of Contents

  1. Introduction
  2. Dataset description
  3. The original ABC database
  4. Modified ABC database and data preprocessing
  5. Methods
  6. Dimensionality reduction methods
  7. Anomaly detection methods
  8. Results
  9. Performance in different spaces
  10. Effect of noise
  11. Method comparison
  12. Conclusions and discussion

Figures (11)

  • Figure 1: Distributions of selected parameters in the database used in this study (after Forestano_2023). The blue histograms represent the distributions over the full ABC database, while the orange histograms represent the distributions within the subset left after the cuts discussed in the text.
  • Figure 2: Left panel: Scatter plot of the temperature versus the log concentration of CO$_2$ (after Forestano_2023). The gray points represent the normal population and the red points represent the anomalous population. Right panel: Histogrammed distributions of planets according to their log-concentration of CO$_2$: the normal population (gray histogram) and the anomalous population (red histogram).
  • Figure 3: Comparison of normal spectra (left panel) and anomalous spectra (right panel). For better readability, here we show only 100 normalized spectra of each population.
  • Figure 4: Reconstruction comparison for normal (left) and anomalous (right) spectra (after normalization). For normal data, the autoencoder closely matches the original spectrum. For anomalous data, the reconstruction fails to capture key features, showing larger differences. The spectra pictured here are 'ideal' and have not had any noise added.
  • Figure 5: A visualization of the architecture of the autoencoder network. Each colored block represents a layer of the network and the shaded band indicates a ReLU activation function, present on all layers except the output.
  • ...and 6 more figures