Search

Sequential Monte Carlo (SMC) methods are powerful tools for Bayesian inference but suffer from requiring many particles for accurate estimates, leading to high computational costs. We introduce persistent sampling (PS), an extension of SMC that mitigates this issue by allowing particles from previous iterations to persist. This generates a growing, weighted ensemble of particles distributed across iterations. In each iteration, PS utilizes multiple importance sampling and resampling from the mixture of all previous distributions to produce the next generation of particles. This addresses particle impoverishment and mode collapse, resulting in more accurate posterior approximations. Furthermore, this approach provides lower-variance marginal likelihood estimates for model comparison. Additionally, the persistent particles improve transition kernel adaptation for efficient exploration. Experiments on complex distributions show that PS consistently outperforms standard methods, achieving lower squared bias in posterior moment estimation and significantly reduced marginal likelihood errors, all at a lower computational cost. PS offers a robust, efficient, and scalable framework for Bayesian inference., Comment: 30 pages, 9 figures, 4 tables. Submitted to Statistics & Computing

Ensemble Kalman Inversion (EKI) has been proposed as an efficient method for the approximate solution of Bayesian inverse problems with expensive forward models. However, when applied to the Bayesian inverse problem EKI is only exact in the regime of Gaussian target measures and linear forward models. In this work we propose embedding EKI and Flow Annealed Kalman Inversion (FAKI), its normalizing flow (NF) preconditioned variant, within a Bayesian annealing scheme as part of an adaptive implementation of the $t$-preconditioned Crank-Nicolson (tpCN) sampler. The tpCN sampler differs from standard pCN in that its proposal is reversible with respect to the multivariate $t$-distribution. The more flexible tail behaviour allows for better adaptation to sampling from non-Gaussian targets. Within our Sequential Kalman Tuning (SKT) adaptation scheme, EKI is used to initialize and precondition the tpCN sampler for each annealed target. The subsequent tpCN iterations ensure particles are correctly distributed according to each annealed target, avoiding the accumulation of errors that would otherwise impact EKI. We demonstrate the performance of SKT for tpCN on three challenging numerical benchmarks, showing significant improvements in the rate of convergence compared to adaptation within standard SMC with importance weighted resampling at each temperature level, and compared to similar adaptive implementations of standard pCN. The SKT scheme applied to tpCN offers an efficient, practical solution for solving the Bayesian inverse problem when gradients of the forward model are not available. Code implementing the SKT schemes for tpCN is available at \url{https://github.com/RichardGrumitt/KalmanMC}., Comment: 58 pages, 22 figures, accepted in Inverse Problems

For many scientific inverse problems we are required to evaluate an expensive forward model. Moreover, the model is often given in such a form that it is unrealistic to access its gradients. In such a scenario, standard Markov Chain Monte Carlo algorithms quickly become impractical, requiring a large number of serial model evaluations to converge on the target distribution. In this paper we introduce Flow Annealed Kalman Inversion (FAKI). This is a generalization of Ensemble Kalman Inversion (EKI), where we embed the Kalman filter updates in a temperature annealing scheme, and use normalizing flows (NF) to map the intermediate measures corresponding to each temperature level to the standard Gaussian. In doing so, we relax the Gaussian ansatz for the intermediate measures used in standard EKI, allowing us to achieve higher fidelity approximations to non-Gaussian targets. We demonstrate the performance of FAKI on two numerical benchmarks, showing dramatic improvements over standard EKI in terms of accuracy whilst accelerating its already rapid convergence properties (typically in $\mathcal{O}(10)$ steps)., Comment: 9 pages, 2 figues. Presented at MaxEnt 2023. Modified version to appear in MaxEnt 2023 proceedings

The goal of this thesis is twofold; introduce the fundamentals of Bayesian inference and computation focusing on astronomical and cosmological applications, and present recent advances in probabilistic computational methods developed by the author that aim to facilitate Bayesian data analysis for the next generation of astronomical observations and theoretical models. The first part of this thesis familiarises the reader with the notion of probability and its relevance for science through the prism of Bayesian reasoning, by introducing the key constituents of the theory and discussing its best practices. The second part includes a pedagogical introduction to the principles of Bayesian computation motivated by the geometric characteristics of probability distributions and followed by a detailed exposition of various methods including Markov chain Monte Carlo (MCMC), Sequential Monte Carlo (SMC), and Nested Sampling (NS). Finally, the third part presents two novel computational methods (Ensemble Slice Sampling and Preconditioned Monte Carlo) and their respective software implementations (zeus and pocoMC). [abridged], Comment: PhD Thesis, 280 pages

We present jax-cosmo, a library for automatically differentiable cosmological theory calculations. It uses the JAX library, which has created a new coding ecosystem, especially in probabilistic programming. As well as batch acceleration, just-in-time compilation, and automatic optimization of code for different hardware modalities (CPU, GPU, TPU), JAX exposes an automatic differentiation (autodiff) mechanism. Thanks to autodiff, jax-cosmo gives access to the derivatives of cosmological likelihoods with respect to any of their parameters, and thus enables a range of powerful Bayesian inference algorithms, otherwise impractical in cosmology, such as Hamiltonian Monte Carlo and Variational Inference. In its initial release, jax-cosmo implements background evolution, linear and non-linear power spectra (using halofit or the Eisenstein and Hu transfer function), as well as angular power spectra with the Limber approximation for galaxy and weak lensing probes, all differentiable with respect to the cosmological parameters and their other inputs. We illustrate how autodiff can be a game-changer for common tasks involving Fisher matrix computations, or full posterior inference with gradient-based techniques. In particular, we show how Fisher matrices are now fast, exact, no longer require any fine tuning, and are themselves differentiable. Finally, using a Dark Energy Survey Year 1 3x2pt analysis as a benchmark, we demonstrate how jax-cosmo can be combined with Probabilistic Programming Languages to perform posterior inference with state-of-the-art algorithms including a No U-Turn Sampler, Automatic Differentiation Variational Inference,and Neural Transport HMC. We further demonstrate that Normalizing Flows using Neural Transport are a promising methodology for model validation in the early stages of analysis.

pocoMC is a Python package for accelerated Bayesian inference in astronomy and cosmology. The code is designed to sample efficiently from posterior distributions with non-trivial geometry, including strong multimodality and non-linearity. To this end, pocoMC relies on the Preconditioned Monte Carlo algorithm which utilises a Normalising Flow in order to decorrelate the parameters of the posterior. It facilitates both tasks of parameter estimation and model comparison, focusing especially on computationally expensive applications. It allows fitting arbitrary models defined as a log-likelihood function and a log-prior probability density function in Python. Compared to popular alternatives (e.g. nested sampling) pocoMC can speed up the sampling procedure by orders of magnitude, cutting down the computational cost substantially. Finally, parallelisation to computing clusters manifests linear scaling., Comment: 6 pages, 1 figure. Submitted to JOSS. Code available at https://github.com/minaskar/pocomc

We introduce Preconditioned Monte Carlo (PMC), a novel Monte Carlo method for Bayesian inference that facilitates efficient sampling of probability distributions with non-trivial geometry. PMC utilises a Normalising Flow (NF) in order to decorrelate the parameters of the distribution and then proceeds by sampling from the preconditioned target distribution using an adaptive Sequential Monte Carlo (SMC) scheme. The results produced by PMC include samples from the posterior distribution and an estimate of the model evidence that can be used for parameter inference and model comparison respectively. The aforementioned framework has been thoroughly tested in a variety of challenging target distributions achieving state-of-the-art sampling performance. In the cases of primordial feature analysis and gravitational wave inference, PMC is approximately 50 and 25 times faster respectively than Nested Sampling (NS). We found that in higher dimensional applications the acceleration is even greater. Finally, PMC is directly parallelisable, manifesting linear scaling up to thousands of CPUs., Comment: 10 pages, 9 figures. Submitted to MNRAS. Code available at https://github.com/minaskar/pocomc

We present $\texttt{matryoshka}$, a suite of neural network based emulators and accompanying Python package that have been developed with the goal of producing fast and accurate predictions of the nonlinear galaxy power spectrum. The suite of emulators consists of four linear component emulators, from which fast linear predictions of the power spectrum can be made, allowing all nonlinearities to be included in predictions from a nonlinear boost component emulator. The linear component emulators includes an emulator for the matter transfer function that produces predictions in $\sim 0.0004 \ \mathrm{s}$, with an error of $<0.08\%$ (at $1\sigma$ level) on scales $10^{-4} \ h \ \mathrm{Mpc}^{-1}

We introduce hankl, a lightweight Python implementation of the FFTLog algorithm for Cosmology. The FFTLog algorithm is an extension of the Fast Fourier Transform (FFT) for logarithmically spaced periodic sequences. It can be used to efficiently compute Hankel transformations, which are paramount for many modern cosmological analyses that are based on the power spectrum or the 2-point correlation function multipoles. The code is well-tested, open source, and publicly available., Comment: 6 pages, 2 figures; Code available at https://github.com/minaskar/hankl

We introduce zeus, a well-tested Python implementation of the Ensemble Slice Sampling (ESS) method for Bayesian parameter inference. ESS is a novel Markov chain Monte Carlo (MCMC) algorithm specifically designed to tackle the computational challenges posed by modern astronomical and cosmological analyses. In particular, the method requires only minimal hand--tuning of 1-2 hyper-parameters that are often trivial to set; its performance is insensitive to linear correlations and it can scale up to 1000s of CPUs without any extra effort. Furthermore, its locally adaptive nature allows to sample efficiently even when strong non-linear correlations are present. Lastly, the method achieves a high performance even in strongly multimodal distributions in high dimensions. Compared to emcee, a popular MCMC sampler, zeus performs 9 and 29 times better in a cosmological and an exoplanet application respectively., Comment: 15 pages, 17 figures, 2 tables, published in MNRAS; Code available at https://github.com/minaskar/zeus

Slice Sampling has emerged as a powerful Markov Chain Monte Carlo algorithm that adapts to the characteristics of the target distribution with minimal hand-tuning. However, Slice Sampling's performance is highly sensitive to the user-specified initial length scale hyperparameter and the method generally struggles with poorly scaled or strongly correlated distributions. This paper introduces Ensemble Slice Sampling (ESS), a new class of algorithms that bypasses such difficulties by adaptively tuning the initial length scale and utilising an ensemble of parallel walkers in order to efficiently handle strong correlations between parameters. These affine-invariant algorithms are trivial to construct, require no hand-tuning, and can easily be implemented in parallel computing environments. Empirical tests show that Ensemble Slice Sampling can improve efficiency by more than an order of magnitude compared to conventional MCMC methods on a broad range of highly correlated target distributions. In cases of strongly multimodal target distributions, Ensemble Slice Sampling can sample efficiently even in high dimensions. We argue that the parallel, black-box and gradient-free nature of the method renders it ideal for use in scientific fields such as physics, astrophysics and cosmology which are dominated by a wide variety of computationally expensive and non-differentiable models., Comment: Published in Statistics & Computing; 18 pages, 11 figures, 2 tables; Code available https://github.com/minaskar/zeus