arXiv subjects

Priya Goyal

Publications and source records attributed to Priya Goyal.

16 recordsLinked to original sources

arXiv2026

Morphological Signatures of Gravitational Evolution, Redshift-Space Distortions, and Massive Neutrinos in Large-Scale Structure

We investigate the morphological properties of large-scale structure in the Universe and the physical processes that modify the excursion-set morphology of the three-dimensional matter density field. Using the Quijote N-body simulation suite, we study how an initially Gaussian random matter density field is altered by non-linear gravitational evolution, redshift-space distortions, and massive neutrino free-streaming. To quantify these effects, we employ a comprehensive set of morphological descriptors, including Minkowski Functionals, Betti numbers, Minkowski Tensors, and local measures of the size and shape of connected components and cavities. We find that gravitational evolution, on quasi-linear scales $R_G \sim 10 h^{-1} \mathrm{Mpc}$, strongly skews the one-point distribution and slightly smooths the field via the merging of critical points, with a more pronounced effect for minima and wall saddle points than for peaks. Redshift-space distortions produce the strongest morphological signal, generating pronounced anisotropies that are robustly captured by Minkowski Tensors and local shape measures, arising from both coherent large-scale flows and non-linear Finger-of-God effects. In contrast, massive neutrinos induce an approximately isotropic suppression of small-scale structure, slightly reducing the amplitudes of the Minkowski Functionals while leaving individual shape measures largely unchanged. We further explore the sensitivity of these statistics to variations in cosmological parameters $\Omega_m$, $n_s$, and $\sigma_8$, finding that they probe strongly degenerate combinations of $\Omega_m$ and $n_s$, while also exhibiting sensitivity to $\sigma_8$ through the non-Gaussianity of the evolved density field.

astro-ph.CO↗

arXiv2025

Local patch analysis of ACT DR6 convergence map using morphological statistics

We carry out a comprehensive hierarchical multi-scale morphological analysis to search for anomalous behaviour in the large scale matter distribution using convergence map provided by the Atacama Cosmology Telescope (ACT) Data Release 6. We use a suite of morphological statistics consisting of Minkowski functionals, contour Minkowski tensor and Betti numbers for the analysis, and compute their deviations from the ensemble expectations and median values obtained from isotropic $\Lambda$CDM simulations provided by ACT. To assess the statistical significance of these deviations, we devise a general methodology based on the persistence of the deviations across threshold ranges and spatial resolutions, while taking into account correlations among the statistics. From the analysis of the full dataset, and hemispherical regions, we find consistency with isotropic $\Lambda$CDM simulations provided by ACT. Since deviations in smaller sky regions tend to get washed out when averaged over larger regions, we further analyze smaller sky patches. This localized analysis reveals some patches that exhibit statistically significant deviations which we refer to as 'anomalous'. We find that near the CMB cold spot, both the positive and negative density fluctuations are anomalous, at 99% CL and 95% CL respectively. This region also encompasses an anomalous southern spot previously identified in Planck CMB temperature data. We also carry out a comparison of anomalous patches identified here for ACT data with a previous analysis of the convergence map from Planck. We do not find common patches between the two datasets, which suggest that the anomalous behavior of the Planck data arises from noise in the map. Further investigation of the atypical patches using large scale structure surveys is warranted to determine their physical origin.

astro-ph.CO↗

arXiv2024

Investigating Cosmic Homogeneity Using Multi-fractal Analysis of the SDSS-IV eBOSS DR16 Quasar Catalog

We analyze the volume-limited subsamples extracted from the sixteenth data release of the SDSS-IV eBOSS quasar survey spanning a redshift interval of $0.8 < z < 2.2$, to estimate the scale of transition to homogeneity in the Universe. The multi-fractal analysis used for this purpose considers the scaling behavior of different moments of quasar distribution in different density environments. This analysis gives the spectrum of generalized dimension $D_q$, where positive values of $q$ characterize the scaling behavior in over-dense regions and the negative ones in under-dense regions. We expect fractal correlation dimension $D_q(r) = 3$, for a homogeneous, random point distribution in 3-Dimensions. The fractal correlation dimension $D_q(r)$, corresponding to $q=2$ obtained in our study stabilizes in the range (2.8-2.9) for scales $r>80$ $h^{-1}$ Mpc. The observed quasar distribution shows consistency with the simulated mock data and the random distribution of quasars within one sigma. Further, the generalized dimension spectrum $D_q(r)$ also reveals transition to homogeneity beyond $>110$ $h^{-1}$ Mpc, and the dominance of clustering at small scales $r<80$ $h^{-1}$ Mpc. Consequently, our study provides strong evidence for the homogeneity in SDSS quasar distribution, offering insights into large-scale structure properties and, thus can play a pivotal role in scrutinizing the clustering properties of quasars and its evolution in various upcoming surveys such as Dark Energy Spectroscopic Instrument (DESI) and Extremely Large Telescope (ELT).

astro-ph.CO↗

arXiv2024

CLIP the Bias: How Useful is Balancing Data in Multimodal Learning?

We study the effectiveness of data-balancing for mitigating biases in contrastive language-image pretraining (CLIP), identifying areas of strength and limitation. First, we reaffirm prior conclusions that CLIP models can inadvertently absorb societal stereotypes. To counter this, we present a novel algorithm, called Multi-Modal Moment Matching (M4), designed to reduce both representation and association biases (i.e. in first- and second-order statistics) in multimodal data. We use M4 to conduct an in-depth analysis taking into account various factors, such as the model, representation, and data size. Our study also explores the dynamic nature of how CLIP learns and unlearns biases. In particular, we find that fine-tuning is effective in countering representation biases, though its impact diminishes for association biases. Also, data balancing has a mixed impact on quality: it tends to improve classification but can hurt retrieval. Interestingly, data and architectural improvements seem to mitigate the negative impact of data balancing on performance; e.g. applying M4 to SigLIP-B/16 with data quality filters improves COCO image-to-text retrieval @5 from 86% (without data balancing) to 87% and ImageNet 0-shot classification from 77% to 77.5%! Finally, we conclude with recommendations for improving the efficacy of data balancing in multimodal systems.

cs.LG↗

arXiv2022

A Self-Supervised Descriptor for Image Copy Detection

Image copy detection is an important task for content moderation. We introduce SSCD, a model that builds on a recent self-supervised contrastive training objective. We adapt this method to the copy detection task by changing the architecture and training objective, including a pooling operator from the instance matching literature, and adapting contrastive learning to augmentations that combine images. Our approach relies on an entropy regularization term, promoting consistent separation between descriptor vectors, and we demonstrate that this significantly improves copy detection accuracy. Our method produces a compact descriptor vector, suitable for real-world web scale applications. Statistical information from a background image distribution can be incorporated into the descriptor. On the recent DISC2021 benchmark, SSCD is shown to outperform both baseline copy detection models and self-supervised architectures designed for image classification by huge margins, in all settings. For example, SSCD out-performs SimCLR descriptors by 48% absolute. Code is available at https://github.com/facebookresearch/sscd-copy-detection.

cs.CV↗

arXiv2022

Vision Models Are More Robust And Fair When Pretrained On Uncurated Images Without Supervision

Discriminative self-supervised learning allows training models on any random group of internet images, and possibly recover salient information that helps differentiate between the images. Applied to ImageNet, this leads to object centric features that perform on par with supervised features on most object-centric downstream tasks. In this work, we question if using this ability, we can learn any salient and more representative information present in diverse unbounded set of images from across the globe. To do so, we train models on billions of random images without any data pre-processing or prior assumptions about what we want the model to learn. We scale our model size to dense 10 billion parameters to avoid underfitting on a large data size. We extensively study and validate our model performance on over 50 benchmarks including fairness, robustness to distribution shift, geographical diversity, fine grained recognition, image copy detection and many image classification datasets. The resulting model, not only captures well semantic information, it also captures information about artistic style and learns salient information such as geolocations and multilingual word embeddings based on visual content only. More importantly, we discover that such model is more robust, more fair, less harmful and less biased than supervised models or models trained on object centric datasets such as ImageNet.

cs.CV↗

arXiv2022

Fairness Indicators for Systematic Assessments of Visual Feature Extractors

Does everyone equally benefit from computer vision systems? Answers to this question become more and more important as computer vision systems are deployed at large scale, and can spark major concerns when they exhibit vast performance discrepancies between people from various demographic and social backgrounds. Systematic diagnosis of fairness, harms, and biases of computer vision systems is an important step towards building socially responsible systems. To initiate an effort towards standardized fairness audits, we propose three fairness indicators, which aim at quantifying harms and biases of visual systems. Our indicators use existing publicly available datasets collected for fairness evaluations, and focus on three main types of harms and bias identified in the literature, namely harmful label associations, disparity in learned representations of social and demographic traits, and biased performance on geographically diverse images from across the world.We define precise experimental protocols applicable to a wide range of computer vision models. These indicators are part of an ever-evolving suite of fairness probes and are not intended to be a substitute for a thorough analysis of the broader impact of the new computer vision technologies. Yet, we believe it is a necessary first step towards (1) facilitating the widespread adoption and mandate of the fairness assessments in computer vision research, and (2) tracking progress towards building socially responsible models. To study the practical effectiveness and broad applicability of our proposed indicators to any visual system, we apply them to off-the-shelf models built using widely adopted model training paradigms which vary in their ability to whether they can predict labels on a given image or only produce the embeddings. We also systematically study the effect of data domain and model size.

cs.CV↗

arXiv2021

The geometrical meaning of statistical isotropy of smooth random fields in two dimensions

We revisit the geometrical meaning of statistical isotropy that is manifest in excursion sets of smooth random fields in two dimensions. Using the contour Minkowski tensor, $\W_1$, as our basic tool we first examine geometrical properties of single structures. For simple closed curves in two dimensions we show that $\W_1$ is proportional to the identity matrix if the curve has $m$-fold symmetry, with $m\ge 3$. Then we elaborate on how $\W_1$ maps any arbitrary shaped simple closed curve to an ellipse that is unique up to translations of its centroid. We also carry out a comparison of the shape parameters, $\alpha$ and $\beta$, defined using $\W_1$, with the filamentarity parameter defined using two scalar Minkowski functionals - area and contour length. We show that they contain complementary shape information, with $\W_1$ containing additional information of orientation of structures. Next, we apply our method to boundaries of excursion sets of random fields and examine what statistical isotropy means for the geometry of the excursion sets. Focusing on Gaussian isotropic fields, and using a semi-numerical approach we quantify the effect of finite sampling of the field on the geometry of the excursion sets. In doing so we obtain an analytic expression for $\alpha$ which takes into account the effect of finite sampling. Finally we derive an analytic expression for the ensemble expectation of $\W_1$ for Gaussian anisotropic random fields. Our results provide insights that are useful for designing tests of statistical isotropy using cosmological data.

astro-ph.CO↗

arXiv2021

Local patch analysis for testing statistical isotropy of the Planck convergence map

The small but measurable effect of weak gravitational lensing on the cosmic microwave background radiation provide information about the large-scale distribution of matter in the universe. We use the all sky distribution of matter, as represented by the {\em convergence map} that is inferred from CMB lensing measurement by Planck survey, to test the fundamental assumption of Statistical Isotropy (SI) of the universe. For the analysis we use the $\alpha$ statistic that is devised from the contour Minkowski tensor, a tensorial generalization of the scalar Minkowski functional, the contour length. In essence, the $\alpha$ statistic captures the ellipticity of isofield contours at any chosen threshold value of a smooth random field and provides a measure of anisotropy. The SI of the observed convergence map is tested against the suite of realistic simulations of the convergence map provided by the Planck collaboration. We first carry out a global analysis using the full sky data after applying the galactic and point sources mask. We find that the observed data is consistent with SI. Further we carry out a local search for departure from SI in small patches of the sky using $\alpha$. This analysis reveals several sky patches which exhibit deviations from simulations with statistical significance higher than 95\% confidence level (CL). Our analysis indicates that the source of the anomalous behaviour of most of the outlier patches is inaccurate estimation of noise. We identify two outlier patches which exhibit anomalous behaviour originating from departure from SI at higher than 95\% CL. Most of the anomalous patches are found to be located roughly along the ecliptic plane or in proximity to the ecliptic poles.

astro-ph.CO↗

arXiv2021

Self-supervised Pretraining of Visual Features in the Wild

Recently, self-supervised learning methods like MoCo, SimCLR, BYOL and SwAV have reduced the gap with supervised methods. These results have been achieved in a control environment, that is the highly curated ImageNet dataset. However, the premise of self-supervised learning is that it can learn from any random image and from any unbounded dataset. In this work, we explore if self-supervision lives to its expectation by training large models on random, uncurated images with no supervision. Our final SElf-supERvised (SEER) model, a RegNetY with 1.3B parameters trained on 1B random images with 512 GPUs achieves 84.2% top-1 accuracy, surpassing the best self-supervised pretrained model by 1% and confirming that self-supervised learning works in a real world setting. Interestingly, we also observe that self-supervised models are good few-shot learners achieving 77.9% top-1 with access to only 10% of ImageNet. Code: https://github.com/facebookresearch/vissl

cs.CV↗

arXiv2020

Unsupervised Learning of Visual Features by Contrasting Cluster Assignments

Unsupervised image representations have significantly reduced the gap with supervised pretraining, notably with the recent achievements of contrastive learning methods. These contrastive methods typically work online and rely on a large number of explicit pairwise feature comparisons, which is computationally challenging. In this paper, we propose an online algorithm, SwAV, that takes advantage of contrastive methods without requiring to compute pairwise comparisons. Specifically, our method simultaneously clusters the data while enforcing consistency between cluster assignments produced for different augmentations (or views) of the same image, instead of comparing features directly as in contrastive learning. Simply put, we use a swapped prediction mechanism where we predict the cluster assignment of a view from the representation of another view. Our method can be trained with large and small batches and can scale to unlimited amounts of data. Compared to previous contrastive methods, our method is more memory efficient since it does not require a large memory bank or a special momentum network. In addition, we also propose a new data augmentation strategy, multi-crop, that uses a mix of views with different resolutions in place of two full-resolution views, without increasing the memory or compute requirements much. We validate our findings by achieving 75.3% top-1 accuracy on ImageNet with ResNet-50, as well as surpassing supervised pretraining on all the considered transfer tasks.

cs.CV↗

arXiv2019

Morphology of CMB fields -- effect of weak gravitational lensing

We study the morphology of the cosmic microwave background temperature and polarization fields using the shape and alignment parameters, $\beta$ and $\alpha$, that are constructed from the contour Minkowski tensor. The primary goal of our paper is to understand the effect of weak gravitational lensing on the morphology of the CMB fields. In order to isolate different physical effects that can be potentially confused with the effect of lensing, we first study the effect of varying the cosmology on $\alpha$ and $\beta$, and show that they are relatively insensitive to variation of cosmological parameters. Next we analyze the signatures of hemispherical anisotropy, and show that information of such anisotropy in $\alpha$ gets washed out at small angular scales and become pronounced only at large angular scales. For $\beta$ we find characteristic distortions which vary with the field threshold. We then study the effect of weak gravitational lensing using simulations of lensed temperature and $E$ and $B$ modes. We quantify the distortion induced in the fields across different angular scales. We find that lensing makes structures of all fields increasingly more anisotropic as we probe down to smaller scales. We find distinct behaviour of morphological distortions as a function of threshold for the different fields. The effect is small for temperature and $E$ mode, while it is significantly large for $B$ mode. Further, we find that lensing does not induce statistical anisotropy, as expected from the isotropic distribution of large scale structure of matter. We expect that the results obtained in this work will provide insights on the reconstruction of the lensing potential.

astro-ph.CO↗

arXiv2019

Scaling and Benchmarking Self-Supervised Visual Representation Learning

Self-supervised learning aims to learn representations from the data itself without explicit manual supervision. Existing efforts ignore a crucial aspect of self-supervised learning - the ability to scale to large amount of data because self-supervision requires no manual labels. In this work, we revisit this principle and scale two popular self-supervised approaches to 100 million images. We show that by scaling on various axes (including data size and problem 'hardness'), one can largely match or even exceed the performance of supervised pre-training on a variety of tasks such as object detection, surface normal estimation (3D) and visual navigation using reinforcement learning. Scaling these methods also provides many interesting insights into the limitations of current self-supervised techniques and evaluations. We conclude that current self-supervised methods are not 'hard' enough to take full advantage of large scale data and do not seem to learn effective high level semantic representations. We also introduce an extensive benchmark across 9 different datasets and tasks. We believe that such a benchmark along with comparable evaluation settings is necessary to make meaningful progress. Code is at: https://github.com/facebookresearch/fair_self_supervision_benchmark.

cs.CV↗

arXiv2018

Tensor Comprehensions: Framework-Agnostic High-Performance Machine Learning Abstractions

Deep learning models with convolutional and recurrent networks are now ubiquitous and analyze massive amounts of audio, image, video, text and graph data, with applications in automatic translation, speech-to-text, scene understanding, ranking user preferences, ad placement, etc. Competing frameworks for building these networks such as TensorFlow, Chainer, CNTK, Torch/PyTorch, Caffe1/2, MXNet and Theano, explore different tradeoffs between usability and expressiveness, research or production orientation and supported hardware. They operate on a DAG of computational operators, wrapping high-performance libraries such as CUDNN (for NVIDIA GPUs) or NNPACK (for various CPUs), and automate memory allocation, synchronization, distribution. Custom operators are needed where the computation does not fit existing high-performance library calls, usually at a high engineering cost. This is frequently required when new operators are invented by researchers: such operators suffer a severe performance penalty, which limits the pace of innovation. Furthermore, even if there is an existing runtime call these frameworks can use, it often doesn't offer optimal performance for a user's particular network architecture and dataset, missing optimizations between operators as well as optimizations that can be done knowing the size and shape of data. Our contributions include (1) a language close to the mathematics of deep learning called Tensor Comprehensions, (2) a polyhedral Just-In-Time compiler to convert a mathematical description of a deep learning DAG into a CUDA kernel with delegated memory management and synchronization, also providing optimizations such as operator fusion and specialization for specific sizes, (3) a compilation cache populated by an autotuner. [Abstract cutoff]

cs.PL↗

arXiv2017

Focal Loss for Dense Object Detection

The highest accuracy object detectors to date are based on a two-stage approach popularized by R-CNN, where a classifier is applied to a sparse set of candidate object locations. In contrast, one-stage detectors that are applied over a regular, dense sampling of possible object locations have the potential to be faster and simpler, but have trailed the accuracy of two-stage detectors thus far. In this paper, we investigate why this is the case. We discover that the extreme foreground-background class imbalance encountered during training of dense detectors is the central cause. We propose to address this class imbalance by reshaping the standard cross entropy loss such that it down-weights the loss assigned to well-classified examples. Our novel Focal Loss focuses training on a sparse set of hard examples and prevents the vast number of easy negatives from overwhelming the detector during training. To evaluate the effectiveness of our loss, we design and train a simple dense detector we call RetinaNet. Our results show that when trained with the focal loss, RetinaNet is able to match the speed of previous one-stage detectors while surpassing the accuracy of all existing state-of-the-art two-stage detectors. Code is at: https://github.com/facebookresearch/Detectron.

cs.CV↗

arXiv2017

Accurate, Large Minibatch SGD: Training ImageNet in 1 Hour

Deep learning thrives with large neural networks and large datasets. However, larger networks and larger datasets result in longer training times that impede research and development progress. Distributed synchronous SGD offers a potential solution to this problem by dividing SGD minibatches over a pool of parallel workers. Yet to make this scheme efficient, the per-worker workload must be large, which implies nontrivial growth in the SGD minibatch size. In this paper, we empirically show that on the ImageNet dataset large minibatches cause optimization difficulties, but when these are addressed the trained networks exhibit good generalization. Specifically, we show no loss of accuracy when training with large minibatch sizes up to 8192 images. To achieve this result, we adopt a hyper-parameter-free linear scaling rule for adjusting learning rates as a function of minibatch size and develop a new warmup scheme that overcomes optimization challenges early in training. With these simple techniques, our Caffe2-based system trains ResNet-50 with a minibatch size of 8192 on 256 GPUs in one hour, while matching small minibatch accuracy. Using commodity hardware, our implementation achieves ~90% scaling efficiency when moving from 8 to 256 GPUs. Our findings enable training visual recognition models on internet-scale data with high efficiency.

cs.CV↗