arXiv subjects

Kai Hou Yip

Publications and source records attributed to Kai Hou Yip.

At least 19 recordsLinked to original sources

arXiv2026

Graph neural networks for exoplanet atmospheres

Calculating disequilibrium chemistry in exoplanet atmospheres remains a significant computational bottleneck in atmospheric retrievals. The increasing observational precision from facilities such as JWST and the Ariel mission requires including disequilibrium chemistry in these analyses. Previous studies have demonstrated that neural networks can emulate kinetic chemistry, although their spatial inductive bias does not align with the topology of chemical reaction networks. This study introduces a graph neural network surrogate that represents chemical species as nodes and temperature-dependent reaction rates as edges, thereby enabling information propagation along physically meaningful chemical pathways. The model is trained on atmospheres generated using the Venot+2020 chemical scheme and Guillot temperature-pressure profiles. The GNN accurately reconstructs disequilibrium abundances across the sampled parameter space and reduces the mean abundance error by a factor of approximately 3 compared to the previous U-Net model. When applied to transmission spectra, most predictions fall within the observational precision expected for JWST and Ariel, with only about 7% of test atmospheres exceeding a 20 ppm mean spectral error. Performance variations are primarily observed in chemically transitional regimes near a carbon-to-oxygen ratio of one and at low temperatures. An evaluation of the boundary-case planet WASP-39b demonstrates effective performance under a moderate domain shift. Perturbation analysis indicates that disturbances propagate along chemical connectivity rather than spatial adjacency, confirming that the architecture captures the structure of reaction networks. These results suggest that GNN surrogates provide accurate, computationally efficient predictions of disequilibrium chemistry, facilitating integration into the atmospheric retrieval pipeline.

astro-ph.IM↗

arXiv2026

Operational Range Bounding in Spectroscopy: A Safety Cage Framework for Machine Learning Models

Ensuring the reliability of black-box machine learning models in safety-critical space missions remains a significant challenge, particularly when ground-truth is unavailable for validation. Although machine learning models offer a powerful means to augment standard pipelines by extracting transmission spectra from complex exoplanetary light curves, their susceptibility to unmodelled instrument anomalies, stellar activity, and domain shifts introduces unquantified risks. This study evaluates a modular safety cage architecture that operates as a parallel monitoring layer to assess the validity of a prediction without modifying the underlying estimator. By monitoring different runtime indicators, including uncertainty quantification, out-of-domain detection, and influence functions, the framework constrains the model's operational domain to a verified region. A controlled evaluation is conducted under both in-domain and cross-domain conditions, using datasets from the 2019 and 2021 editions of the Ariel Data Challenges. The results reveal that model failure is multifaceted and that no single indicator captures all failure modes, demonstrating the need for indicator fusion. The application of safety-driven rejection strategies shows that a modest 20% reduction in data coverage results in error reductions between 45% and 65% across different domains and evaluation metrics. Using a formalised coverage-risk framework, a systematic analysis of indicator combinations is performed to identify configurations that maximise risk-ranking accuracy and optimise the trade-off between data coverage and scientific performance. Safety cages provide a transparent mechanism for detecting unreliable predictions and represent a critical step towards the safe deployment of data-driven models in scientific applications, such as astrophysics, where ground truth is seldom available.

cs.LG↗

arXiv2026

Traceable Spectral Inference via Influence Functions: Efficient Data Attribution and Error Proxies for the Ariel Mission

Interpretability is critical for machine learning models deployed in scientific space missions such as ESA's Ariel, where ground truth is unavailable during operations and physical plausibility must be assessed. While most explainable AI methods focus on feature attribution, this work investigates training data attribution through influence functions and introduces three key contributions for operational spectroscopy pipelines. First, influence is reformulated in terms of prediction rather than loss, enabling label-free deployment. Second, by leveraging the closed-form ridge solution of an Extreme Learning Machine, infinitesimal prediction influence is efficiently computed. Third, an influence-based conservative error proxy is derived by propagating training residuals through the influence sensitivities. Evaluated against simulated spectra, the proposed proxy correlates strongly with scale and shape-based spectral errors. Furthermore, influence functions enable the identification of the most influential samples and the approximation of the most harmful ones. Together, these results suggest that this approach can serve as an operational framework for scientific machine learning.

cs.LG↗

arXiv2026

A public dataset of Ariel simulated observations for developing exoplanetary atmosphere data reduction pipelines

Detecting and characterising exoplanet atmospheres remains challenging because atmospheric signals can be comparable to residual noise and instrumental/astrophysical systematics. Spectral features span from a few ppm for small planets up to $\sim 10^3$ ppm for warm/hot giants, while high-quality JWST time-series spectroscopy typically reaches $\sim 10$--$50$ ppm (occasionally $\sim 100$--$200$ ppm in the presence of stellar variability or stronger systematics), making correlated noise across temporal and spectral dimensions a key limitation. With JWST delivering an increasing volume of high-precision transmission spectra, and Ariel set to extend this to a homogeneous survey of $\sim 10^3$ exoplanet atmospheres, robust benchmarking resources with known ground truth are essential to develop and validate data-driven (including ML-based) detrending approaches. As a major step towards this goal, we use ExoSim2 and TauREx to generate one of the most comprehensive public datasets based on the current payload design of the ESA Ariel mission, specifically intended to benchmark detrending algorithms. We also provide a deep neural network baseline for time-series reduction, and use it to highlight the limitations of ML based detrendng methods, i.e. the risks posed by dataset shift when observed distributions diverge from those of the training set, a scenario likely to arise in real observations. This dataset is featured in the Ariel Data Challenge 2024 on Kaggle and has been field-tested for robustness and simulation fidelity. By making these resources publicly available, we aim to support the community in developing, comparing, and stress-testing scalable and reliable methods for exoplanet transmission spectroscopy.

astro-ph.EP↗

arXiv2026

Adaptive Online Emulation for Accelerating Complex Physical Simulations

Complex physical simulations often require trade-offs between model fidelity and computational feasibility. We introduce Adaptive Online Emulation (AOE), which dynamically learns neural network surrogates during simulation execution to accelerate expensive components. Unlike existing methods requiring extensive offline training, AOE uses Online Sequential Extreme Learning Machines (OS-ELMs) to continuously adapt emulators along the actual simulation trajectory. We employ a numerically stable variant of the OS-ELM using cumulative sufficient statistics to avoid matrix inversion instabilities. AOE integrates with time-stepping frameworks through a three-phase strategy balancing data collection, updates, and surrogate usage, while requiring orders of magnitude less training data than conventional surrogate approaches. Demonstrated on a 1D atmospheric model of exoplanet GJ1214b, AOE achieves 11.1 times speedup (91% time reduction) across 200,000 timesteps while maintaining accuracy, potentially making previously intractable high-fidelity time-stepping simulations computationally feasible.

physics.comp-ph↗

arXiv2025

Investigating the Influence of Asymmetric Errors on Retrievals of Exoplanet Transmission Spectra

In studies of exoplanet atmospheres using transmission spectroscopy, Bayesian retrievals are the most popular form of analysis. In these procedures it is common to adopt a Gaussian likelihood. However, this implicitly assumes that the upper and lower error bars on the spectral points are equal. With recent observations from the James Webb Space Telescope (JWST) offering higher quality of data, it is worth revisiting this assumption to understand the impact that an asymmetry between the error bars may have on retrieved parameters. In this study, we challenge the approximation by comparing retrievals using a symmetric, Gaussian likelihood, and an asymmetric, split normal likelihood. We find that the influence of this assumption is minimal at the scales of asymmetry observed in JWST observations of WASP-39 b (with a maximum asymmetry of 77%) but we show that it would become critical with greater levels of asymmetry (e.g. an average asymmetry of 80%). Furthermore, we stress the importance of the shape of the asymmetric distribution and the difficulty in fitting this distribution from three summary statistics (the median and an upper and lower bound on the transit depth). An asymmetric likelihood sampler will incorrectly predict parameters if the shape of the likelihood does not match that of the underlying noise distribution even when the levels of asymmetry are equal in both. Overall, we find that it is safe to use the Gaussian likelihood assumption for current datasets but it is worth considering the potential bias if greater asymmetries are observed.

astro-ph.EP↗

arXiv2025

Extreme Learning Machines for Exoplanet Simulations: A Faster, Lightweight Alternative to Deep Learning

Increasing resolution and coverage of astrophysical and climate data necessitates increasingly sophisticated models, often pushing the limits of computational feasibility. While emulation methods can reduce calculation costs, the neural architectures typically used--optimised via gradient descent--are themselves computationally expensive to train, particularly in terms of data generation requirements. This paper investigates the utility of the Extreme Learning Machine (ELM) as a lightweight, non-gradient-based machine learning algorithm for accelerating complex physical models. We evaluate ELM surrogate models in two test cases with different data structures: (i) sequentially-structured data, and (ii) image-structured data. For test case (i), where the number of samples $N$ >> the dimensionality of input data $d$, ELMs achieve remarkable efficiency, offering a 100,000$\times$ faster training time and a 40$\times$ faster prediction speed compared to a Bi-Directional Recurrent Neural Network (BIRNN), whilst improving upon BIRNN test performance. For test case (ii), characterised by $d >> N$ and image-based inputs, a single ELM was insufficient, but an ensemble of 50 individual ELM predictors achieves comparable accuracy to a benchmark Convolutional Neural Network (CNN), with a 16.4$\times$ reduction in training time, though costing a 6.9$\times$ increase in prediction time. We find different sample efficiency characteristics between the test cases: in test case (i) individual ELMs demonstrate superior sample efficiency, requiring only 0.28% of the training dataset compared to the benchmark BIRNN, while in test case (ii) the ensemble approach requires 78% of the data used by the CNN to achieve comparable results--representing a trade-off between sample efficiency and model complexity.

astro-ph.EP↗

arXiv2025

The effect of spectroscopic binning on atmospheric retrievals

With the JWST offering higher resolution data in space-based transmission spectroscopy, understanding the capabilities of our current atmospheric retrieval pipelines is essential. These new data cover wider wavelength ranges and at much higher spectral resolution than previous instruments have been able to offer. Therefore, it is often appealing to bin spectra to fewer points, better constrained in their transit depth, before using them as inputs for atmospheric retrievals. As such, we produce a simulation replicating the observations of WASP-39b by the Near Infrared Spectrograph (NIRSpec) instrument on board JWST using the PRISM dispersion element. Then, we assess the accuracy and consistency of retrievals while varying both the resolution and the average photometric error of this simulated spectrum. We repeat this analysis on three different simulation setups where each includes an opaque cloud layer at a different height in the atmosphere. In agreement with previous studies, we find that a much greater resolution is needed in the case of a high cloud deck since features are already heavily muted by the presence of the clouds. In the other two cases, there are large 'safe zones' in the parameter space where accurate estimations are made. If these maps can be generalized, they could be used to inform future observations on how long to observe a given target in order to achieve the most accurate retrieval results. We also find that the resolution required to fully resolve the degeneracies between the parameters contributing to the spectra is much greater than that needed to constrain the marginalized posterior distributions for each parameter individually.

astro-ph.EP↗

arXiv2024

Enhancing 3D Planetary Atmosphere Simulations with a Surrogate Radiative Transfer Model

This work introduces an approach to enhancing the computational efficiency of 3D atmospheric simulations by integrating a machine-learned surrogate model into the OASIS global circulation model (GCM). Traditional GCMs, which are based on repeatedly numerically integrating physical equations governing atmospheric processes across a series of time-steps, are time-intensive, leading to compromises in spatial and temporal resolution of simulations. This research improves upon this limitation, enabling higher resolution simulations within practical timeframes. Speeding up 3D simulations holds significant implications in multiple domains. Firstly, it facilitates the integration of 3D models into exoplanet inference pipelines, allowing for robust characterisation of exoplanets from a previously unseen wealth of data anticipated from JWST and post-JWST instruments. Secondly, acceleration of 3D models will enable higher resolution atmospheric simulations of Earth and Solar System planets, enabling more detailed insights into their atmospheric physics and chemistry. Our method replaces the radiative transfer module in OASIS with a recurrent neural network-based model trained on simulation inputs and outputs. Radiative transfer is typically one of the slowest components of a GCM, thus providing the largest scope for overall model speed-up. The surrogate model was trained and tested on the specific test case of the Venusian atmosphere, to benchmark the utility of this approach in the case of non-terrestrial atmospheres. This approach yields promising results, with the surrogate-integrated GCM demonstrating above 99.0% accuracy and 147 factor GPU speed-up of the entire simulation compared to using the matched original GCM under Venus-like conditions.

astro-ph.EP↗

arXiv2024

A Population Analysis of 20 Exoplanets Observed from the Optical to the Near-infrared Wavelengths with HST: Evidence for Widespread Stellar Contamination

We present a population study of 20 exoplanets, ranging from Neptune-like to inflated hot-Jupiter planets, observed during transit with the STIS and WFC3 instruments aboard the Hubble Space Telescope. To obtain spectral information from the near-UV to the near-infrared, we reanalysed sixteen WFC3 and over fifty STIS archival data sets with our dedicated HST pipeline. We also include twenty-four WFC3 data sets previously reduced with the same software. Across our target sample we observe significant divergence among multiple observations conducted with the same STIS grating at various epochs, whilst we do not detect variations in the WFC3 data sets. These results are suggestive of stellar contamination, which we have investigated further using known Bayesian tools and other tailored metrics, facilitating a more objective assessment of stellar activity intensity within each system. Our findings reveal that stellar activity contaminates up to half of the studied exoplanet atmospheres, albeit at varying extents. Accounting for stellar activity can significantly alter planetary atmospheric parameters like molecular abundances (up to 6 orders of magnitude) and temperature (up to 145 %), contrasting with the results of analyses that neglect activity. Our results emphasise the importance of considering the effects of stellar contamination in exoplanet transit studies; this issue is particularly true for data sets obtained with facilities that do not cover the optical and/or UV spectral range where the activity is expected to be more impactful but also more easily detectable. Our results also provide a catalogue of potentially active stars for further investigation and monitoring.

astro-ph.EP↗

arXiv2024

Operational range bounding of spectroscopy models with anomaly detection

Safe operation of machine learning models requires architectures that explicitly delimit their operational ranges. We evaluate the ability of anomaly detection algorithms to provide indicators correlated with degraded model performance. By placing acceptance thresholds over such indicators, hard boundaries are formed that define the model's coverage. As a use case, we consider the extraction of exoplanetary spectra from transit light curves, specifically within the context of ESA's upcoming Ariel mission. Isolation Forests are shown to effectively identify contexts where prediction models are likely to fail. Coverage/error trade-offs are evaluated under conditions of data and concept drift. The best performance is seen when Isolation Forests model projections of the prediction model's explainability SHAP values.

cs.LG↗

arXiv2024

Towards atmospheric retrievals of panchromatic light-curves: ExPLOR-ing generalized inversion techniques for transiting exoplanets with JWST and Ariel

Conventional atmospheric retrieval codes are designed to extract information, such as chemical abundances, thermal structures and cloud properties, from fully "reduced" spectra obtained during transit or eclipse. Reduced spectra, however, are assembled by fitting a series of simplified light-curves to time series observations, wavelength-by-wavelength. Thus, spectra are a post-processed summary statistics of the original data, which by definition does not encode all the available information (i.e., astrophysical signal, model covariance, instrumental noise). Here, we explore an alternative inversion strategy where the atmospheric retrieval is performed on the light-curve directly -- i.e., closer to the data. This method is implemented in ExPLOR (EXoplanet Panchromatic Light-curve Observation and Retrieval), a novel atmospheric retrieval code inheriting from the TauREx project. By explicitly considering time in the model, ExPLOR naturally handles transit, eclipse, phase-curve and other complex geometries for transiting exoplanets. In this paper, we have validated this new technique by inverting simulated panchromatic light-curves. The model was tested on realistic simulations of a WASP-43b like exoplanet as observed with the James Webb Space Telescope (JWST) and Ariel telescope. By comparing our panchromatic light-curve approach against conventional spectral retrievals on mock scenarios, we have identified key breaking points in information and noise propagation when employing past literature techniques. Throughout the paper, we discuss the importance of developing "closer-to-data" approaches such as the method presented in this work, and highlight the inevitable increase in model complexity and computing requirements associated with the recent JWST revolution.

astro-ph.EP↗

arXiv2024

Is the atmosphere of the ultra-hot Jupiter WASP-121b variable?

We present a comprehensive analysis of the Hubble Space Telescope observations of the atmosphere of WASP-121 b, a ultra-hot Jupiter. After reducing the transit, eclipse, and phase-curve observations with a uniform methodology and addressing the biases from instrument systematics, sophisticated atmospheric retrievals are used to extract robust constraints on the thermal structure, chemistry, and cloud properties of the atmosphere. Our analysis shows that the observations are consistent with a strong thermal inversion beginning at ~0.1 bar on the dayside, solar to subsolar metallicity Z (i.e., -0.77 < log(Z) < 0.05), and super-solar C/O ratio (i.e., 0.59 < C/O < 0.87). More importantly, utilizing the high signal-to-noise ratio and repeated observations of the planet, we identify the following unambiguous time-varying signals in the data: i) a shift of the putative hotspot offset between the two phase-curves and ii) varying spectral signatures in the transits and eclipses. By simulating the global dynamics of WASP-121 b atmosphere at high-resolution, we show that the identified signals are consistent with quasi-periodic weather patterns, hence atmospheric variability, with signatures at the level probed by the observations (~5% to ~10%) that change on a timescale of ~5 planet days; in the simulations, the weather patterns arise from the formation and movement of storms and fronts, causing hot (as well as cold) patches of atmosphere to deform, separate, and mix in time.

astro-ph.EP↗

arXiv2023

To Sample or Not To Sample: Retrieving Exoplanetary Spectra with Variational Inference and Normalising Flows

Current endeavours in exoplanet characterisation rely on atmospheric retrieval to quantify crucial physical properties of remote exoplanets from observations. However, the scalability and efficiency of the technique are under strain with increasing spectroscopic resolution and forward model complexity. The situation becomes more acute with the recent launch of the James Webb Space Telescope and other upcoming missions. Recent advances in Machine Learning provide optimisation-based Variational Inference as an alternative approach to perform approximate Bayesian Posterior Inference. In this investigation we combined Normalising Flow-based neural network with our newly developed differentiable forward model, Diff-Tau, to perform Bayesian Inference in the context of atmospheric retrieval. Using examples from real and simulated spectroscopic data, we demonstrated the superiority of our proposed framework: 1) Training Our neural network only requires a single observation; 2) It produces high-fidelity posterior distributions similar to sampling-based retrieval and; 3) It requires 75% less forward model computation to converge. 4.) We performed, for the first time, Bayesian model selection on our trained neural network. Our proposed framework contribute towards the latest development of a neural-powered atmospheric retrieval. Its flexibility and speed hold the potential to complement sampling-based approaches in large and complex data sets in the future.

astro-ph.EP↗

arXiv2023

Characterising a World Within the Hot Neptune Desert: Transit Observations of LTT 9779 b with HST WFC3

We present an atmospheric analysis of LTT 9779 b, a rare planet situated in the hot Neptune desert, that has been observed with HST WFC3 G102 and G141. The combined transmission spectrum, which covers 0.8 - 1.6 $μ$m, shows a gradual increase in transit depth with wavelength. Our preferred atmospheric model shows evidence for H$_{\rm 2}$O, CO$_{\rm 2}$ and FeH with a significance of 3.1 $σ$, 2.4 $σ$ and 2.1 $σ$, respectively. In an attempt to constrain the rate of atmospheric escape for this planet, we search for the 1.083 $μ$m Helium line in the G102 data but find no evidence of excess absorption that would indicate an escaping atmosphere using this tracer. We refine the orbital ephemerides of LTT 9779 b using our HST data and observations from TESS, searching for evidence of orbital decay or apsidal precession, which is not found. The phase-curve observation of LTT 9779 b with JWST NIRISS should provide deeper insights into the atmosphere of this planet and the expected atmospheric escape might be detected with further observations concentrated on other tracers such as Lyman $α$.

astro-ph.EP↗

arXiv2023

ESA-Ariel Data Challenge NeurIPS 2022: Introduction to exo-atmospheric studies and presentation of the Atmospheric Big Challenge (ABC) Database

This is an exciting era for exo-planetary exploration. The recently launched JWST, and other upcoming space missions such as Ariel, Twinkle and ELTs are set to bring fresh insights to the convoluted processes of planetary formation and evolution and its connections to atmospheric compositions. However, with new opportunities come new challenges. The field of exoplanet atmospheres is already struggling with the incoming volume and quality of data, and machine learning (ML) techniques lands itself as a promising alternative. Developing techniques of this kind is an inter-disciplinary task, one that requires domain knowledge of the field, access to relevant tools and expert insights on the capability and limitations of current ML models. These stringent requirements have so far limited the developments of ML in the field to a few isolated initiatives. In this paper, We present the Atmospheric Big Challenge Database (ABC Database), a carefully designed, organised and publicly available database dedicated to the study of the inverse problem in the context of exoplanetary studies. We have generated 105,887 forward models and 26,109 complementary posterior distributions generated with Nested Sampling algorithm. Alongside with the database, this paper provides a jargon-free introduction to non-field experts interested to dive into the intricacy of atmospheric studies. This database forms the basis for a multitude of research directions, including, but not limited to, developing rapid inference techniques, benchmarking model performance and mitigating data drifts. A successful application of this database is demonstrated in the NeurIPS Ariel ML Data Challenge 2022.

astro-ph.EP↗

arXiv2022

Exploring the Ability of HST WFC3 G141 to Uncover Trends in Populations of Exoplanet Atmospheres Through a Homogeneous Transmission Survey of 70 Gaseous Planets

We present the analysis of the atmospheres of 70 gaseous extrasolar planets via transit spectroscopy with Hubble's Wide Field Camera 3 (WFC3). For over half of these, we statistically detect spectral modulation which our retrievals attribute to molecular species. Among these, we use Bayesian Hierarchical Modelling to search for chemical trends with bulk parameters. We use the extracted water abundance to infer the atmospheric metallicity and compare it to the planet's mass. We also run chemical equilibrium retrievals, fitting for the atmospheric metallicity directly. However, although previous studies have found evidence of a mass-metallicity trend, we find no such relation within our data. For the hotter planets within our sample, we find evidence for thermal dissociation of dihydrogen and water via the H$^-$ opacity. We suggest that the general lack of trends seen across this population study could be due to i) the insufficient spectral coverage offered by HST WFC3 G141, ii) the lack of a simple trend across the whole population, iii) the essentially random nature of the target selection for this study or iv) a combination of all the above. We set out how we can learn from this vast dataset going forward in an attempt to ensure comparative planetology can be undertaken in the future with facilities such as JWST, Twinkle and Ariel. We conclude that a wider simultaneous spectral coverage is required as well as a more structured approach to target selection.

astro-ph.EP↗

arXiv2022

Don't Pay Attention to the Noise: Learning Self-supervised Representations of Light Curves with a Denoising Time Series Transformer

Astrophysical light curves are particularly challenging data objects due to the intensity and variety of noise contaminating them. Yet, despite the astronomical volumes of light curves available, the majority of algorithms used to process them are still operating on a per-sample basis. To remedy this, we propose a simple Transformer model -- called Denoising Time Series Transformer (DTST) -- and show that it excels at removing the noise and outliers in datasets of time series when trained with a masked objective, even when no clean targets are available. Moreover, the use of self-attention enables rich and illustrative queries into the learned representations. We present experiments on real stellar light curves from the Transiting Exoplanet Space Satellite (TESS), showing advantages of our approach compared to traditional denoising techniques.

astro-ph.IM↗