Periodic Reporting for period 4 - ExoAI (Deciphering super-Earths using Artificial Intelligence)
Berichtszeitraum: 2022-07-01 bis 2023-06-30
Understanding the origins and prevalence of life in the universe are some of society's oldest and most fundamental pursuits. The study of exoplanets - i.e. planets orbiting other stars – offers one of the most promising avenues of answering fundamental questions of planet formation, astrobiology and extra-terrestrial habitability. Contrary to expectations, super-Earths (planets roughly between 1 - 10 Earth masses) are, in fact, the most abundant planets. In other words, our galaxy is one of Earths and slightly larger planets, not giants.
Understanding their natures requires a step change in data analysis and modelling.
We require a step change in the data analysis of current exoplanet observations to answer these questions. In short, the more sensitive we are to the faintest signals, the more information we can glean from the light of these foreign worlds. Recent developments in machine learning and artificial intelligence (AI) have revolutionised many areas of industry and science. The ExoAI project focused on the development of state-of-the-art AI solutions to existing issues in data analysis of exoplanetary observations as well as the theoretical modelling of exoplanet atmospheres. Using machine learning, we can better understand the behaviour of instruments such as the Hubble Space Telescope (HST) and disentangle the faint planetary signatures from the systematic noise of the instrument. Similarly, on the theoretical side, we have developed AI algorithms that speed up traditionally computationally intensive simulations of the exoplanet atmospheres. So-called ‘inverse models’ translate the observations into a description of the physical characteristics of the exoplanet. By making these models significantly faster, we were able to incorporate more realistic physics and chemistry than what was possible before. These AI-powered solutions help us to study these exotic worlds using a consistent methodology and gain insight into the population of super-Earths in far greater detail than previously possible. The key results of this project were the development and release of the TauREx 3 atmospheric modelling software, which has been downloaded 64,000 times to date and led to over 250 publications in the field already. Another key result was our full analysis of all HST exoplanet atmospheric data. By analysing all HST data coherently, we were able to answer long-standing questions of exoplanet formation and evolution.
Understanding their natures requires a step change in data analysis and modelling.
We require a step change in the data analysis of current exoplanet observations to answer these questions. In short, the more sensitive we are to the faintest signals, the more information we can glean from the light of these foreign worlds. Recent developments in machine learning and artificial intelligence (AI) have revolutionised many areas of industry and science. The ExoAI project focused on the development of state-of-the-art AI solutions to existing issues in data analysis of exoplanetary observations as well as the theoretical modelling of exoplanet atmospheres. Using machine learning, we can better understand the behaviour of instruments such as the Hubble Space Telescope (HST) and disentangle the faint planetary signatures from the systematic noise of the instrument. Similarly, on the theoretical side, we have developed AI algorithms that speed up traditionally computationally intensive simulations of the exoplanet atmospheres. So-called ‘inverse models’ translate the observations into a description of the physical characteristics of the exoplanet. By making these models significantly faster, we were able to incorporate more realistic physics and chemistry than what was possible before. These AI-powered solutions help us to study these exotic worlds using a consistent methodology and gain insight into the population of super-Earths in far greater detail than previously possible. The key results of this project were the development and release of the TauREx 3 atmospheric modelling software, which has been downloaded 64,000 times to date and led to over 250 publications in the field already. Another key result was our full analysis of all HST exoplanet atmospheric data. By analysing all HST data coherently, we were able to answer long-standing questions of exoplanet formation and evolution.
The ExoAI project's outstanding achievement includes the detection of water vapour in the atmosphere of a temperate super-Earth/warm-Neptune, K2-18b (Tsiaras et al. 2019, Nature Astronomy). This discovery is crucial for understanding potentially habitable planets in our solar system. The project has also published studies on hot-Neptunes and hot-Jupiters, establishing a catalogue of uniformly analysed exoplanets and shedding light on planet formation and atmospheric parameters (Tisaras et al. 2018, Changeat et al. 2022, Edwards et al. 2022).
To improve data analysis, ExoAI pioneered the use of deep learning techniques like LSTM neural networks and transformers (Morvan et al. 2020, 2021, 2022). These algorithms efficiently extract faint planetary signals from instrumental noise, surpassing traditional approaches. The TauREx 3 framework (Al-Rafaie et al. 2020, 2021, 2022), a sophisticated atmospheric modelling tool, enables faster analysis and uniformity across the exoplanet community for better population insights. It has led to over 250 publications in the field and was downloaded 64,000 times.
Moreover, the project explores deep learning's potential in the retrieval process, accelerating data analysis and allowing incorporation of more complex models (Zingales & Waldmann 2018, Yip et al. 2021, 2023). We have also conducted the first study of Explainable AI (XAI) in how and if deep learning algorithms learn the underlying atmospheric physics from example (Yip et al. 2021). This is an important fundamental stepping stone for the field as we demonstrated that learning does indeed occur in deep learning approaches. The ExoAI project's pioneering efforts in data analysis and atmospheric modeling open new frontiers in exoplanet research, advancing our knowledge of distant worlds and their potential for habitability.
Beyond planetary science, ExoAI's algorithms find diverse applications, from methane detection on Mars to identifying ammonia clouds on Saturn (Waldmann & Griffith 2019, Nature Astronomy), to detecting illegal fishing activity in South East Asia and to detecting Satellites orbiting our Earth (ERC-PoC 2023). Our cross-disciplinary applications demonstrate the versatility and impact of AI-driven solutions in diverse scientific domains.
To improve data analysis, ExoAI pioneered the use of deep learning techniques like LSTM neural networks and transformers (Morvan et al. 2020, 2021, 2022). These algorithms efficiently extract faint planetary signals from instrumental noise, surpassing traditional approaches. The TauREx 3 framework (Al-Rafaie et al. 2020, 2021, 2022), a sophisticated atmospheric modelling tool, enables faster analysis and uniformity across the exoplanet community for better population insights. It has led to over 250 publications in the field and was downloaded 64,000 times.
Moreover, the project explores deep learning's potential in the retrieval process, accelerating data analysis and allowing incorporation of more complex models (Zingales & Waldmann 2018, Yip et al. 2021, 2023). We have also conducted the first study of Explainable AI (XAI) in how and if deep learning algorithms learn the underlying atmospheric physics from example (Yip et al. 2021). This is an important fundamental stepping stone for the field as we demonstrated that learning does indeed occur in deep learning approaches. The ExoAI project's pioneering efforts in data analysis and atmospheric modeling open new frontiers in exoplanet research, advancing our knowledge of distant worlds and their potential for habitability.
Beyond planetary science, ExoAI's algorithms find diverse applications, from methane detection on Mars to identifying ammonia clouds on Saturn (Waldmann & Griffith 2019, Nature Astronomy), to detecting illegal fishing activity in South East Asia and to detecting Satellites orbiting our Earth (ERC-PoC 2023). Our cross-disciplinary applications demonstrate the versatility and impact of AI-driven solutions in diverse scientific domains.
Data analysis and interpretation in the field of exoplanets have been treated as separate sub-fields, leading to significant issues, especially in low signal-to-noise scenarios. The instrument model used to de-trend data can impact the final interpretation due to accidental over- or under-correction of the observed signal. To address this, a holistic framework was developed, merging data analysis, instrumental, and atmospheric models into a unified model. This approach allows for a comprehensive understanding of uncertainties and biases that were previously hidden. Key components include a redevelopment of TauREx 3 as a framework with a plugin system, state-of-the-art AI techniques for data de-trending and atmospheric modeling, and differentiable atmospheric forward models to propagate gradients and noise through the analysis pipeline efficiently.
Another focus was on the explainability of deep learning approaches in atmospheric modeling. The black-box nature of neural networks and the difficulty in interpreting their decision-making have been a criticism. In response, the project introduced explainable AI techniques to understand the workings of deep learning in atmospheric retrieval. This work laid a foundation for future developments in the field.
With the JWST telescope launch having been delayed several times, we are now receiving the Early Release Science data, which is of ground-breaking quality. We will build on the successful developments of this project to apply TauREx 3 and data analysis techniques to the JWST exoplanet data in the future.
Another focus was on the explainability of deep learning approaches in atmospheric modeling. The black-box nature of neural networks and the difficulty in interpreting their decision-making have been a criticism. In response, the project introduced explainable AI techniques to understand the workings of deep learning in atmospheric retrieval. This work laid a foundation for future developments in the field.
With the JWST telescope launch having been delayed several times, we are now receiving the Early Release Science data, which is of ground-breaking quality. We will build on the successful developments of this project to apply TauREx 3 and data analysis techniques to the JWST exoplanet data in the future.