Research suggesting the possibility of habitable exoplanets has truly ushered in a paradigm shift in cosmology. Yet, without certainty, there can be no truth. We are yet to identify which of these detected planets have the ability to foster life or are already inhabited.
Over the years, several potentially habitable exoplanets have been hypothesised by applying various Earth-based criteria. Some have even introduced new habitability metrics using modelling and supervised learning approaches. However, manually scanning thousands of planets and identifying those comparable to Earth is a time-consuming task. Furthermore, as the number of identified exoplanets continually increases, identifying those rare atypical examples by describing them in terms of planetary properties, types, populations, and, ultimately, habitability potential necessitates knowledge of many planetary parameters gleaned from observations; requiring hours of expensive telescope time.
This is precisely where artificial intelligence (AI) makes a world of difference!
Led by Prof Snehanshu Saha of BITS Pilani, Goa and Dr Margarita Safonova of Indian Institute of Astrophysics (IIA), an autonomous institute of the Department of Science & Technology, Government of India, a group of researchers that included Prof Santonu Sarkar, Jyotirmoy Sarkar (a doctoral student) and Kartik Bhatia (an undergraduate student), have devised a novel AI-based approach to identify potentially habitable planets with a high probability.
The team was able to detect and shortlist 6 potentially habitable planets, which are a subset of the 60 optimistic habitable candidates identified by Abel Mendez and his group in the PHL-EC database. This, according to the scientists, is a criitcal validation of the AI-based method. In addition, the team also proposed nearly 8000 candidate planets that showed anomalous features like Earth.
Analytics India Magazine caught up with Prof Snehanshu Saha to delve deeper into their research and understand its impact on cosmology.
Solving the anomaly detection problem with AI
As per the Planetary Habitability Laboratory’s Exoplanets Catalogue, only a small fraction of all known planets are habitable. They’ve further categorised the list to optimistic (24 planets) and conservative (36 planets). As the number of habitable exoplanets is significantly lesser than non-habitable planets, habitable candidates can be thought of as anomalies. In this study, the team professed the idea of equating the habitability detection problem to an anomaly detection problem, keeping Earth as the only known anomaly and categorising potentially habitable planets as anomalies as well. This will essentially allow researchers to detect anomalies with an unsupervised machine learning (ML) approach via a novel clustering algorithm. Keeping Earth as the point of reference, they can estimate the prerequisites for habitability. However, this would require a quick screening tool.
Powered by AI, the Multi-Stage Memetic Binary Tree Anomaly Identifier (MSMBTAI) is based on a novel multi-stage memetic algorithm (MSMA) that is predicated on viewing Earth as an anomaly. The AI identifies anomalies and extends them to an unsupervised multi-stage multi-version memetic clustering algorithm (MSMVMCA) to detect potentially habitable exoplanets based on those anomalies. The results are cross-checked with the Planetary Habitability Laboratory’s habitable exoplanet catalogue with both optimistic and conservative lists of potentially habitable exoplanets.
The algorithm builds upon the generic notion of a meme (using imitation to transfer an idea or knowledge from one person to another) to indicate cross-cultural evolution in posterity and subsequently induce new learning mechanisms as generations pass. This way, it can act as a quick screening tool for evaluating habitability perspectives from observed properties. Using the proposed technique, the team was able to identify a few planets that exhibited similar anomalous characteristics like Earth. Interestingly, the new anomalous candidate detection technique showed similar results without using surface temperature as a feature.
Though the quest to find potentially-habitable planets outside of Earth doesn’t solely depend on AI/ML methods, combining these techniques with domain knowledge can undoubtedly make it easier to analyse exoplanets in the future. Garnering promising results, the MSMVMCA-based anomaly detection method is definitely another “giant leap” for the scientific community and humanity at large.
