arXiv subjects

Sambatra Andrianomena

Publications and source records attributed to Sambatra Andrianomena.

At least 19 recordsLinked to original sources

arXiv2025

Towards Mitigating Systematics in Large-Scale Surveys via Few-Shot Optimal Transport-Based Feature Alignment

Systematics contaminate observables, leading to distribution shifts relative to theoretically simulated signals-posing a major challenge for using pre-trained models to label such observables. Since systematics are often poorly understood and difficult to model, removing them directly and entirely may not be feasible. To address this challenge, we propose a novel method that aligns learned features between in-distribution (ID) and out-of-distribution (OOD) samples by optimizing a feature-alignment loss on the representations extracted from a pre-trained ID model. We first experimentally validate the method on the MNIST dataset using possible alignment losses, including mean squared error and optimal transport, and subsequently apply it to large-scale maps of neutral hydrogen. Our results show that optimal transport is particularly effective at aligning OOD features when parity between ID and OOD samples is unknown, even with limited data-mimicking real-world conditions in extracting information from large-scale surveys. Our code is available at https://github.com/sultan-hassan/feature-alignment-for-OOD-generalization.

astro-ph.IM↗

arXiv2025

Probabilistic cosmological inference on HI tomographic data

We explore the possibility of retrieving cosmological information from 21-cm tomographic data at intermediate redshift. The first step in our approach consists of training an encoder, composed of several three dimensional convolutional layers, to cast the neutral hydrogen 3D data into a lower dimension latent space. Once pre-trained, the featurizer is able to generate 3D grid representations which, in turn, will be mapped onto cosmology ($\Omega_{\rm m}$, $\sigma_{8}$) via likelihood-free inference. For the latter, which is framed as a density estimation problem, we consider a Bayesian approximation method which exploits the capacity of Masked Autoregressive Flow to estimate the posterior. It is found that the representations learned by the deep encoder are separable in latent space. Results show that the neural density estimator, trained on the latent codes, is able to constrain cosmology with a precision of $R^2 \ge 0.91$ on all parameters and that most of the ground truth of the instances in the test set fall within $1\sigma$ uncertainty. It is established that the posterior uncertainty from the density estimator is reasonably calibrated. We also investigate the robustness of the feature extractor by using it to compress out-of-distribution dataset, that is either from a different simulation or from the same simulation but at different redshift. We find that, while trained on the latent codes corresponding to different types of out-of-distribution dataset, the probabilistic model is still reasonably capable of constraining cosmology, with $R^2 \ge 0.80$ in general. This highlights both the predictive power of the density estimator considered in this work and the meaningfulness of the latent codes retrieved by the encoder. We believe that the approach prescribed in this proof of concept will be of great use when analyzing 21-cm data from various surveys in the near future.

astro-ph.IM↗

arXiv2024

Towards cosmological inference on unlabeled out-of-distribution HI observational data

We present an approach that can be utilized in order to account for the covariate shift between two datasets of the same observable with different distributions. This helps improve the generalizability of a neural network model trained on in-distribution samples (IDs) when inferring cosmology at the field level on out-of-distribution samples (OODs) of {\it unknown labels}. We make use of HI maps from the two simulation suites in CAMELS, IllustrisTNG and SIMBA. We consider two different techniques, namely adversarial approach and optimal transport, to adapt a target network whose initial weights are those of a source network pre-trained on a labeled dataset. Results show that after adaptation, salient features that are extracted by source and target encoders are well aligned in the embedding space. This indicates that the target encoder has learned the representations of the target domain via the adversarial training and optimal transport. Furthermore, in all scenarios considered in our analyses, the target encoder, which does not have access to any labels ($\Omega_{\rm m}$) during adaptation phase, is able to retrieve the underlying $\Omega_{\rm m}$ from out-of-distribution maps to a great accuracy of $R^{2}$ score $\ge$ 0.9, comparable to the performance of the source encoder trained in a supervised learning setup. We further test the viability of the techniques when only a few out-of-distribution instances are available for training and find that the target encoder still reasonably recovers the matter density. Our approach is critical in extracting information from upcoming large scale surveys.

astro-ph.IM↗

arXiv2024

Cosmological multifield emulator

We demonstrate the use of deep network to learn the distribution of data from state-of-the-art hydrodynamic simulations of the CAMELS project. To this end, we train a generative adversarial network to generate images composed of three different channels that represent gas density (Mgas), neutral hydrogen density (HI), and magnetic field amplitudes (B). We consider an unconstrained model and another scenario where the model is conditioned on the matter density $\Omega_{\rm m}$ and the amplitude of density fluctuations $\sigma_{8}$. We find that the generated images exhibit great quality which is on a par with that of data, visually. Quantitatively, we find that our model generates maps whose statistical properties, quantified by probability distribution function of pixel values and auto-power spectra, agree reasonably well with those of the real maps. Moreover, the cross-correlations between fields in all maps produced by the emulator are in good agreement with those of the real images, which indicates that our model generates instances whose maps in all three channels describe the same physical region. Furthermore, a CNN regressor, which has been trained to extract $\Omega_{\rm m}$ and $\sigma_{8}$ from CAMELS multifield dataset, recovers the cosmology from the maps generated by our conditional model, achieving $R^{2}$ = 0.96 and 0.83 corresponding to $\Omega_{\rm m}$ and $\sigma_{8}$ respectively. This further demonstrates the great capability of the model to mimic CAMELS data. Our model can be useful for generating data that are required to analyze the information from upcoming multi-wavelength cosmological surveys.

astro-ph.CO↗

arXiv2023

Towards out-of-distribution generalization in large-scale astronomical surveys: robust networks learn similar representations

The generalization of machine learning (ML) models to out-of-distribution (OOD) examples remains a key challenge in extracting information from upcoming astronomical surveys. Interpretability approaches are a natural way to gain insights into the OOD generalization problem. We use Centered Kernel Alignment (CKA), a similarity measure metric of neural network representations, to examine the relationship between representation similarity and performance of pre-trained Convolutional Neural Networks (CNNs) on the CAMELS Multifield Dataset. We find that when models are robust to a distribution shift, they produce substantially different representations across their layers on OOD data. However, when they fail to generalize, these representations change less from layer to layer on OOD data. We discuss the potential application of similarity representation in guiding model design, training strategy, and mitigating the OOD problem by incorporating CKA as an inductive bias during training.

astro-ph.IM↗

arXiv2023

Radio Galaxy Zoo: Leveraging latent space representations from variational autoencoder

We propose to learn latent space representations of radio galaxies, and train a very deep variational autoencoder (\protect\Verb+VDVAE+) on RGZ DR1, an unlabeled dataset, to this end. We show that the encoded features can be leveraged for downstream tasks such as classifying galaxies in labeled datasets, and similarity search. Results show that the model is able to reconstruct its given inputs, capturing the salient features of the latter. We use the latent codes of galaxy images, from MiraBest Confident and FR-DEEP NVSS datasets, to train various non-neural network classifiers. It is found that the latter can differentiate FRI from FRII galaxies achieving \textit{accuracy} $\ge 76\%$, \textit{roc-auc} $\ge 0.86$, \textit{specificity} $\ge 0.73$ and \textit{recall} $\ge 0.78$ on MiraBest Confident dataset, comparable to results obtained in previous studies. The performance of simple classifiers trained on FR-DEEP NVSS data representations is on par with that of a deep learning classifier (CNN based) trained on images in previous work, highlighting how powerful the compressed information is. We successfully exploit the learned representations to search for galaxies in a dataset that are semantically similar to a query image belonging to a different dataset. Although generating new galaxy images (e.g. for data augmentation) is not our primary objective, we find that the \protect\Verb+VDVAE+ model is a relatively good emulator. Finally, as a step toward detecting anomaly/novelty, a density estimator -- Masked Autoregressive Flow (\protect\Verb+MAF+) -- is trained on the latent codes, such that the log-likelihood of data can be estimated. The downstream tasks conducted in this work demonstrate the meaningfulness of the latent codes.

astro-ph.GA↗

arXiv2023

Latent space representations of cosmological fields

We investigate the possibility of learning the representations of cosmological multifield dataset from the CAMELS project. We train a very deep variational encoder on images which comprise three channels, namely gas density (Mgas), neutral hydrogen density (HI), and magnetic field amplitudes (B). The clustering of the images in feature space with respect to some cosmological/astrophysical parameters (e.g. $Ω_{\rm m}$) suggests that the generative model has learned latent space representations of the high dimensional inputs. We assess the quality of the latent codes by conducting a linear test on the extracted features, and find that a single dense layer is capable of recovering some of the parameters to a promising level of accuracy, especially the matter density whose prediction corresponds to a coefficient of determination $R^{2}$ = 0.93. Furthermore, results show that the generative model is able to produce images that exhibit statistical properties which are consistent with those of the training data, down to scales of $k\sim 4h/{\rm Mpc}.$

astro-ph.CO↗

arXiv2023

HIDM: Emulating Large Scale HI Maps using Score-based Diffusion Models

Efficiently analyzing maps from upcoming large-scale surveys requires gaining direct access to a high-dimensional likelihood and generating large-scale fields with high fidelity, which both represent major challenges. Using CAMELS simulations, we employ the state-of-the-art score-based diffusion models to simultaneously achieve both tasks. We show that our model, HIDM, is able to efficiently generate high fidelity large scale HI maps that are in a good agreement with the CAMELS's power spectrum, probability distribution, and likelihood up to second moments. HIDM represents a step forward towards maximizing the scientific return of future large scale surveys.

astro-ph.CO↗

arXiv2023

Invertible mapping between fields in CAMELS

We build a bijective mapping between different physical fields from hydrodynamic CAMELS simulations. We train a CycleGAN on three different setups: translating dark matter to neutral hydrogen (Mcdm-HI), mapping between dark matter and magnetic fields magnitude (Mcdm-B), and finally predicting magnetic fields magnitude from neutral hydrogen (HI-B). We assess the performance of the models using various summary statistics, such as the probability distribution function (PDF) of the pixel values and 2D power spectrum ($P(k)$). Results suggest that in all setups, the model is capable of predicting the target field from the source field and vice versa, and the predicted maps exhibit statistical properties which are consistent with those of the target maps. This is indicated by the fact that the mean and standard deviation of the PDF of maps from the test set is in good agreement with those of the generated maps. The mean and variance of $P(k)$ of the real maps agree well with those of generated ones. The consistency tests on the model suggest that the source field can be recovered reasonably well by a forward mapping (source to target) followed by a backward mapping (target to source). This is demonstrated by the agreement between the statistical properties of the source images and those of the recovered ones.

astro-ph.CO↗

arXiv2022

Emulating cosmological multifields with generative adversarial networks

We explore the possibility of using deep learning to generate multifield images from state-of-the-art hydrodynamic simulations of the CAMELS project. We use a generative adversarial network to generate images with three different channels that represent gas density (Mgas), neutral hydrogen density (HI), and magnetic field amplitudes (B). The quality of each map in each example generated by the model looks very promising. The GAN considered in this study is able to generate maps whose mean and standard deviation of the probability density distribution of the pixels are consistent with those of the maps from the training data. The mean and standard deviation of the auto power spectra of the generated maps of each field agree well with those computed from the maps of IllustrisTNG. Moreover, the cross-correlations between fields in all instances produced by the emulator are in good agreement with those of the dataset. This implies that all three maps in each output of the generator encode the same underlying cosmology and astrophysics.

astro-ph.CO↗

arXiv2022

Predictive uncertainty on improved astrophysics recovery from multifield cosmology

We investigate how the constraints on cosmological and astrophysical parameters ($Ω_{\rm m}$, $σ_{8}$, $A_{\rm SN1}$, $A_{\rm SN2}$) vary when exploiting information from multiple fields in cosmology. We make use of a convolutional neural network to retrieve the salient features from different combinations of field maps from IllustrisTNG in the CAMELS project. The fields considered are neutral hydrogen (HI), gas density (Mgas), magnetic fields (B) and gas metallicity (Z). We estimate the predictive uncertainty on the predictions of our model by using Monte Carlo dropout, a Bayesian approximation. Results show that overall, the performance of the model improves on all parameters as the number of channels of its input is increased. As compared to previous works, our model is able to predict the astrophysical parameters with up to $5\%$ higher in accuracy. In the best setup which includes all fields (four channel input, Mgas-HI-B-Z) the model achieves $R^{2} > 0.96$ on all parameters. Similarly, we find that the total uncertainty, which is dominated by the aleatoric uncertainty, decreases as more fields are used to train the model in general. The uncertainties obtained by dropout variational inference are overestimated on all parameters in our case, in that the predictive uncertainty is much larger than the actual squared error. After calibration, which consists of a simple $σ$ scaling method, the average deviation of the total uncertainty from the actual error goes down to $25\%$ at most (on $A_{\rm SN1}$).

astro-ph.CO↗

arXiv2022

HIFlow: Generating Diverse HI Maps and Inferring Cosmology while Marginalizing over Astrophysics using Normalizing Flows

A wealth of cosmological and astrophysical information is expected from many ongoing and upcoming large-scale surveys. It is crucial to prepare for these surveys now and develop tools that can efficiently extract most information. We present HIFlow: a fast generative model of the neutral hydrogen (HI) maps that is conditioned only on cosmology ($Ω_{m}$ and $σ_{8}$) and designed using a class of normalizing flow models, the Masked Autoregressive Flow (MAF). HIFlow is trained on the state-of-the-art simulations from the Cosmology and Astrophysics with MachinE Learning Simulations (CAMELS) project. HIFlow has the ability to generate realistic diverse maps without explicitly incorporating the expected 2D maps structure into the flow as an inductive bias. We find that HIFlow is able to reproduce the CAMELS average and standard deviation HI power spectrum (Pk) within a factor of $\lesssim$ 2, scoring a very high $R^{2} > 90\%$. By inverting the flow, HIFlow provides a tractable high-dimensional likelihood for efficient parameter inference. We show that the conditional HIFlow on cosmology is successfully able to marginalize over astrophysics at the field level, regardless of the stellar and AGN feedback strengths. This new tool represents a first step toward a more powerful parameter inference, maximizing the scientific return of future HI surveys, and opening a new avenue to minimize the loss of complex information due to data compression down to summary statistics.

astro-ph.CO↗

arXiv2022

Probabilistic learning for pulsar classification

In this work, we explore the possibility of using probabilistic learning to identify pulsar candidates. We make use of Deep Gaussian Process (DGP) and Deep Kernel Learning (DKL). Trained on a balanced training set in order to avoid the effect of class imbalance, the performance of the models, achieving relatively high probability of differentiating the positive class from the negative one ($roc$-$auc \sim 0.98$), is very promising overall. We estimate the predictive entropy of each model predictions and find that DKL is more confident than DGP in its predictions and provides better uncertainty calibration. Upon investigating the effect of training with imbalanced dataset on the models, results show that each model performance decreases with an increasing number of the majority class in the training set. Interestingly, with a number of negative class $10\times$ that of positive class, the models still provide reasonably well calibrated uncertainty, i.e. an expected Uncertainty Calibration Error (UCE) less than $6\%$. We also show in this study how, in the case of relatively small amount of training dataset, a convolutional neural network based classifier trained via Bayesian Active Learning by Disagreement (BALD) performs. We find that, with an optimized number of training examples, the model -- being the most confident in its predictions -- generalizes relatively well and produces the best uncertainty calibration which corresponds to UCE = $3.118\%$.

astro-ph.IM↗

arXiv2020

Constraining the astrophysics and cosmology from 21cm tomography using deep learning with the SKA

Future Square Kilometre Array (SKA) surveys are expected to generate huge datasets of 21cm maps on cosmological scales from the Epoch of Reionization (EoR). We assess the viability of exploiting machine learning techniques, namely, convolutional neural networks (CNN), to simultaneously estimate the astrophysical and cosmological parameters from 21cm maps from semi-numerical simulations. We further convert the simulated 21cm maps into SKA-like mock maps using the detailed SKA antennae distribution, thermal noise and a recipe for foreground cleaning. We successfully design two CNN architectures (VGGNet-like and ResNet-like) that are both efficiently able to extract simultaneously three astrophysical parameters, namely the photon escape fraction (f$_{\rm esc}$), the ionizing emissivity power dependence on halo mass ($C_{\rm ion}$) and the ionizing emissivity redshift evolution index ($D_{\rm ion}$), and three cosmological parameters, namely the matter density parameter ($Ω_{m}$), the dimensionless Hubble constant ($h$), and the matter fluctuation amplitude ($σ_{8}$), from 21cm maps at several redshifts. With the presence of noise from SKA, our designed CNNs are still able to recover these astrophysical and cosmological parameters with great accuracy ($R^{2} > 92\%$), improving to $R^{2} > 99\%$ towards low redshift and low neutral fraction values. Our results show that future 21cm observations can play a key role to break degeneracy between models and tightly constrain the astrophysical and cosmological parameters, using only few frequency channels.

astro-ph.CO↗

arXiv2019

Testing General Relativity with the Doppler magnification effect

The apparent sizes and brightnesses of galaxies are correlated in a dipolar pattern around matter overdensities in redshift space, appearing larger on their near side and smaller on their far side. The opposite effect occurs for galaxies around an underdense region. These patterns of apparent magnification induce dipole and higher multipole terms in the cross-correlation of galaxy number density fluctuations with galaxy size/brightness (which is sensitive to the convergence field). This provides a means of directly measuring peculiar velocity statistics at low and intermediate redshift, with several advantages for performing cosmological tests of GR. In particular, it does not depend on empirically-calibrated scaling relations like the Tully-Fisher and Fundamental Plane methods. We show that the next generation of spectroscopic galaxy redshift surveys will be able to measure the Doppler magnification effect with sufficient signal-to-noise to test GR on large scales. We illustrate this with forecasts for the constraints that can be achieved on parametrised deviations from GR for forthcoming low-redshift galaxy surveys with DESI and SKA2. Although the cross-correlation statistic considered has a lower signal to noise than RSD, it will be a useful probe of GR since it is sensitive to different systematics.

astro-ph.CO↗

arXiv2019

Classifying galaxies according to their HI content

We use machine learning to classify galaxies according to their HI content, based on both their optical photometry and environmental properties. The data used for our analyses are the outputs in the range $z = 0-1$ from MUFASA cosmological hydrodynamic simulation. In our previous paper, where we predicted the galaxy HI content using the same input features, HI rich galaxies were only selected for the training. In order for the predictions on real observation data to be more accurate, the classifiers built in this study will first establish if a galaxy is HI rich ($\rm{log(M_{HI}/M_{*})} > -2 $) before estimating its neutral hydrogen content using the regressors developed in the first paper. We resort to various machine learning algorithms and assess their performance with various metrics such as accuracy for instance. The performance of the classifiers gets better with increasing redshift and reaches their peak performance around $z = 1$. Random Forest method, the most robust among the classifiers when considering only the mock data for both training and test in this study, reaches an accuracy above $98.6 \%$ at $z = 0$ and above $99.0 \%$ at $z = 1$. We test our algorithms, trained with simulation data, on classification of the galaxies in RESOLVE, ALFALFA and GASS surveys. Interestingly, SVM algorithm, the best classifier for the tests, achieves a precision, the relevant metric for the tests, above $87.60\%$ and a specificity above $71.4\%$ with all the tests, indicating that the classifier is capable of learning from the simulated data to classify HI rich/HI poor galaxies from the real observation data. With the advent of large HI 21 cm surveys such as the SKA, this set of classifiers, together with the regressors developed in the first paper, will be part of a pipeline, a very useful tool, which is aimed at predicting HI content of galaxies.

astro-ph.GA↗

arXiv2018

Cosmology with Phase 1 of the Square Kilometre Array; Red Book 2018: Technical specifications and performance forecasts

We present a detailed overview of the cosmological surveys that will be carried out with Phase 1 of the Square Kilometre Array (SKA1), and the science that they will enable. We highlight three main surveys: a medium-deep continuum weak lensing and low-redshift spectroscopic HI galaxy survey over 5,000 sqdeg; a wide and deep continuum galaxy and HI intensity mapping survey over 20,000 sqdeg from z = 0.35 - 3; and a deep, high-redshift HI intensity mapping survey over 100 sqdeg from z = 3 - 6. Taken together, these surveys will achieve an array of important scientific goals: measuring the equation of state of dark energy out to z ~ 3 with percent-level precision measurements of the cosmic expansion rate; constraining possible deviations from General Relativity on cosmological scales by measuring the growth rate of structure through multiple independent methods; mapping the structure of the Universe on the largest accessible scales, thus constraining fundamental properties such as isotropy, homogeneity, and non-Gaussianity; and measuring the HI density and bias out to z = 6. These surveys will also provide highly complementary clustering and weak lensing measurements that have independent systematic uncertainties to those of optical surveys like LSST and Euclid, leading to a multitude of synergies that can improve constraints significantly beyond what optical or radio surveys can achieve on their own. This document, the 2018 Red Book, provides reference technical specifications, cosmological parameter forecasts, and an overview of relevant systematic effects for the three key surveys, and will be regularly updated by the Cosmology Science Working Group in the run up to start of operations and the Key Science Programme of SKA1.

astro-ph.CO↗

arXiv2018

Non-linear relativistic contributions to the cosmological weak-lensing convergence

Relativistic contributions to the dynamics of structure formation come in a variety of forms, and can potentially give corrections to the standard picture on typical scales of 100 Mpc. These corrections cannot be obtained by Newtonian numerical simulations, so it is important to accurately estimate the magnitude of these relativistic effects. Density fluctuations couple to produce a background of gravitational waves, which is larger than any primordial background. A similar interaction produces a much larger spectrum of vector modes which represent the frame-dragging rotation of spacetime. These can change the metric at the percent level in the concordance model at scales below the equality scale. Vector modes modify the lensing of background galaxies by large-scale structure. This gives in principle the exciting possibility of measuring relativistic frame dragging effects on cosmological scales. The effects of the non-linear tensor and vector modes on the cosmic convergence are computed and compared to first-order lensing contributions from density fluctuations, Doppler lensing, and smaller Sachs-Wolfe effects. The lensing from gravitational waves is negligible so we concentrate on the vector modes. We show the relative importance of this for future surveys such as Euclid and SKA. We find that these non-linear effects only marginally affect the overall weak lensing signal so they can safely be neglected in most analyses, though are still much larger than the linear Sachs-Wolfe terms. The second-order vector contribution can dominate the first-order Doppler lensing term at moderate redshifts and are actually more important for survey geometries like the SKA.

gr-qc↗