arXiv subjects

James Robinson

Publications and source records attributed to James Robinson.

17 recordsLinked to original sources

arXiv2025

Technical overview and architecture of the FastNet Machine Learning weather prediction model, version 1.0

We present FastNet version 1.0, a data-driven medium range numerical weather prediction (NWP) model based on a Graph Neural Network architecture, developed jointly between the Alan Turing Institute and the Met Office. FastNet uses an encode-process-decode structure to produce deterministic global weather predictions out to 10 days. The architecture is independent of spatial resolution and we have trained models at 1$^{\circ}$ and 0.25$^{\circ}$ resolution, with a six hour time step. FastNet uses a multi-level mesh in the processor, which is able to capture both short-range and long-range patterns in the spatial structure of the atmosphere. The model is pre-trained on ECMWF's ERA5 reanalysis data and then fine-tuned on additional autoregressive rollout steps, which improves accuracy over longer time horizons. We evaluate the model performance at 1.5$^{\circ}$ resolution using 2022 as a hold-out year and compare with the Met Office Global Model, finding that FastNet surpasses the skill of the current Met Office Global Model NWP system using a variety of evaluation metrics on a number of atmospheric variables. Our results show that both our 1$^{\circ}$ and 0.25$^{\circ}$ FastNet models outperform the current Global Model and produce results with predictive skill approaching those of other data-driven models trained on 0.25$^{\circ}$ ERA5.

physics.ao-ph↗

arXiv2025

FastNet: Improving the physical consistency of machine-learning weather prediction models through loss function design

Machine learning weather prediction (MLWP) models have demonstrated remarkable potential in delivering accurate forecasts at significantly reduced computational cost compared to traditional numerical weather prediction (NWP) systems. However, challenges remain in ensuring the physical consistency of MLWP outputs, particularly in deterministic settings. This study presents FastNet, a graph neural network (GNN)-based global prediction model, and investigates the impact of alternative loss function designs on improving the physical realism of its forecasts. We explore three key modifications to the standard mean squared error (MSE) loss: (1) a modified spherical harmonic (MSH) loss that penalises spectral amplitude errors to reduce blurring and enhance small-scale structure retention; (2) inclusion of horizontal gradient terms in the loss to suppress non-physical artefacts; and (3) an alternative wind representation that decouples speed and direction to better capture extreme wind events. Results show that while the MSH and gradient-based losses \textit{alone} may slightly degrade RMSE scores, when trained in combination the model exhibits very similar MSE performance to an MSE-trained model while at the same time significantly improving spectral fidelity and physical consistency. The alternative wind representation further improves wind speed accuracy and reduces directional bias. Collectively, these findings highlight the importance of loss function design as a mechanism for embedding domain knowledge into MLWP models and advancing their operational readiness.

physics.ao-ph↗

arXiv2023

Characterizing the nucleus of comet 162P/Siding Spring using ground-based photometry

Comet 162P/Siding Spring is a large Jupiter-family comet with extensive archival lightcurve data. We report new r-band nucleus lightcurves for this comet, acquired in 2018, 2021 and 2022. With the addition of these lightcurves, the phase angles at which the nucleus has been observed range from $0.39^\circ$ to $16.33^\circ$. We absolutely-calibrate the comet lightcurves to r-band Pan-STARRS 1 magnitudes, and use these lightcurves to create a convex shape model of the nucleus by convex lightcurve inversion. The best-fitting shape model for 162P has axis ratios $a/b = 1.56$ and $b/c = 2.33$, sidereal period $P = 32.864\pm0.001$ h, and a rotation pole oriented towards ecliptic longitude $\lambda_E = 118^\circ \pm 26^\circ$ and latitude $\beta_E=-50^\circ\pm21^\circ$. We constrain the possible nucleus elongation to lie within $1.4 < a/b < 2.0$ and discuss tentative evidence that 162P may have a bilobed structure. Using the shape model to correct the lightcurves for rotational effects, we derive a linear phase function with slope $\beta=0.051\pm0.002$ mag deg$^{-1}$ and intercept $H_r(1,1,0) = 13.86 \pm 0.02$ for 162P. We find no evidence that the nucleus exhibited an opposition surge at phase angles down to 0.39$^\circ$. The challenges associated with modelling the shapes of comet nuclei from lightcurves are highlighted, and we comment on the extent to which we anticipate that LSST will alleviate these challenges in the coming decade.

astro-ph.EP↗

arXiv2021

Improved Regret Bounds for Tracking Experts with Memory

We address the problem of sequential prediction with expert advice in a non-stationary environment with long-term memory guarantees in the sense of Bousquet and Warmuth [4]. We give a linear-time algorithm that improves on the best known regret bounds [26]. This algorithm incorporates a relative entropy projection step. This projection is advantageous over previous weight-sharing approaches in that weight updates may come with implicit costs as in for example portfolio optimization. We give an algorithm to compute this projection step in linear time, which may be of independent interest.

cs.LG↗

arXiv2019

Design choices for productive, secure, data-intensive research at scale in the cloud

We present a policy and process framework for secure environments for productive data science research projects at scale, by combining prevailing data security threat and risk profiles into five sensitivity tiers, and, at each tier, specifying recommended policies for data classification, data ingress, software ingress, data egress, user access, user device control, and analysis environments. By presenting design patterns for security choices for each tier, and using software defined infrastructure so that a different, independent, secure research environment can be instantiated for each project appropriate to its classification, we hope to maximise researcher productivity and minimise risk, allowing research organisations to operate with confidence.

cs.CR↗

arXiv2019

Online Prediction of Switching Graph Labelings with Cluster Specialists

We address the problem of predicting the labeling of a graph in an online setting when the labeling is changing over time. We present an algorithm based on a specialist approach; we develop the machinery of cluster specialists which probabilistically exploits the cluster structure in the graph. Our algorithm has two variants, one of which surprisingly only requires $\mathcal{O}(\log n)$ time on any trial $t$ on an $n$-vertex graph, an exponential speed up over existing methods. We prove switching mistake-bound guarantees for both variants of our algorithm. Furthermore these mistake bounds smoothly vary with the magnitude of the change between successive labelings. We perform experiments on Chicago Divvy Bicycle Sharing data and show that our algorithms significantly outperform an existing algorithm (a kernelized Perceptron) as well as several natural benchmarks.

cs.LG↗

arXiv2017

Concurrent Geometric Multicasting

We present MCFR, a multicasting concurrent face routing algorithm that uses geometric routing to deliver a message from source to multiple targets. We describe the algorithm's operation, prove it correct, estimate its performance bounds and evaluate its performance using simulation. Our estimate shows that MCFR is the first geometric multicast routing algorithm whose message delivery latency is independent of network size and only proportional to the distance between the source and the targets. Our simulation indicates that MCFR has significantly better reliability than existing algorithms.

cs.DC↗

arXiv2000

Constraining Primordial Non-Gaussianity With the Abundance of High Redshift Clusters

We show how observations of the evolution of the galaxy cluster number abundance can be used to constrain primordial non-Gaussianity in the universe. We carry out a maximum likelihood analysis incorporating a number of current datasets and accounting for a wide range of sources of systematic error. Under the assumption of Gaussianity, the current data prefer a universe with matter density $Ω_m\simeq 0.3$ and are inconsistent with $Ω_m=1$ at the $2σ$ level. If we assume $Ω_m=1$, the predicted degree of cluster evolution is consistent with the data for non-Gaussian models where the primordial fluctuations have at least two times as many peaks of height $3σ$ or more as a Gaussian distribution does. These results are robust to almost all sources of systematic error considered: in particular, the $Ω_m=1$ Gaussian case can only be reconciled with the data if a number of systematic effects conspire to modify the analysis in the right direction. Given an independent measurement of $Ω_m$, the techniques described here represent a powerful tool with which to constrain non-Gaussianity in the primordial universe, independent of specific details of the non-Gaussian physics. We discuss the prospects and strategies for improving the constraints with future observations.

astro-ph↗

arXiv2000

Cosmological constraints from the correlation function of galaxy clusters

I compare various semi-analytic models for the bias of dark matter halos with halo clustering properties observed in recent numerical simulations. The best fitting model is one based on the collapse of ellipsoidal perturbations proposed by Sheth, Mo & Tormen (1999), which fits the halo correlation length to better than 8 per cent accuracy. Using this model, I confirm that the correlation length of clusters of a given separation depends primarily on the shape and amplitude of mass fluctuations in the universe, and is almost independent of other cosmological parameters. Current observational uncertainties make it difficult to draw robust conclusions, but for illustrative purposes I discuss the constraints on the mass power spectrum which are implied by recent analyses of the APM cluster sample. I also discuss the prospects for improving these constraints using future surveys such as the Sloan Digital Sky Survey. Finally, I show how these constraints can be combined with observations of the cluster number abundance to place strong limits on the matter density of the universe.

astro-ph↗

arXiv1999

Evolution of the cluster abundance in non-Gaussian models

We carry out N-body simulations of several non-Gaussian structure formation models, including Peebles' isocurvature cold dark matter model, cosmic string models, and a model with primordial voids. We compare the evolution of the cluster mass function in these simulations with that predicted by a modified version of the Press-Schechter formalism. We find that the Press-Schechter formula can accurately fit the cluster evolution over a wide range of redshifts for all of the models considered, with typical errors in the mass function of less than 25%, considerably smaller than the amount by which predictions for different models may differ. This work demonstrates that the Press-Schechter formalism can be used to place strong model independent constraints on non-Gaussianity in the universe.

astro-ph↗

arXiv1999

WOMBAT & FORECAST: Making Realistic Maps of the Microwave Sky

The Wavelength-Oriented Microwave Background Analysis Team (WOMBAT) is constructing microwave maps which will be more realistic than previous simulations. Our foreground models represent a considerable improvement: where spatial templates are available for a given foreground, we predict the flux and spectral index of that component at each place on the sky and estimate uncertainties. We will produce maps containing simulated CMB anisotropy combined with expected foregrounds. The simulated maps will be provided to the community as the WOMBAT Challenge, so such maps can be analyzed to extract cosmological parameters by scientists who are unaware of their input values. This will test the efficacy of foreground subtraction, power spectrum analysis, and parameter estimation techniques and help identify the areas most in need of progress. These maps are also part of the FORECAST project, which allows web-based access to the known foreground maps for the planning of CMB missions.

astro-ph↗

arXiv1998

The WOMBAT Challenge: A "Hounds and Hares" Exercise for Cosmology

The Wavelength-Oriented Microwave Background Analysis Team (WOMBAT) is constructing microwave skymaps which will be more realistic than previous simulations. Our foreground models represent a considerable improvement: where spatial templates are available for a given foreground, we predict the flux and spectral index of that component at each place on the sky and estimate the uncertainties in these quantities. We will produce maps containing simulated Cosmic Microwave Background anisotropies combined with all major expected foreground components. The simulated maps will be provided to the cosmology community as the WOMBAT Challenge, a "hounds and hares" exercise where such maps can be analyzed to extract cosmological parameters by scientists who are unaware of their input values. This exercise will test the efficacy of current foreground subtraction, power spectrum analysis, and parameter estimation techniques and will help identify the areas most in need of progress.

astro-ph↗

arXiv1998

A Simultaneous Constraint on the Amplitude and Gaussianity of Mass Fluctuations in the Universe

We consider constraints on the amplitude of mass fluctuations in the universe, sigma_8, derived from two simple observations: the present number density of clusters and the amplitude of their correlation function. Allowing for the possibility that the primordial fluctuations are non-gaussian introduces a degeneracy in the value of sigma_8 preferred by each of these constraints. However, when the constraints are taken together this degeneracy is broken, yielding a precise determination of sigma_8 and the degree of non-gaussianity for a given background cosmology. For a flat, Omega_m=1 universe with a power spectrum parameterized by a CDM shape parameter Gamma=0.2, we find that the perturbations are consistent with a gaussian distribution with sigma_8=0.49(+0.08-0.07) (95% limits). For some popular choices of background model, including the favored low matter density models, the hypothesis that the primordial fluctuations are gaussian is ruled out with a high degree of confidence.

astro-ph↗

arXiv1997

The case against scaling defect models of cosmic structure formation

We calculate predictions from defect models of structure formation for both the matter and Cosmic Microwave Background (CMB) over all observable scales. Our results point to a serious problem reconciling the observed large-scale galaxy distribution with the COBE normalization, a result which is robust for a wide range of defect parameters. We conclude that standard scaling defect models are in conflict with the data, and show how attempts to resolve the problem by considering non-scaling defects would require radical departures from the standard scaling picture.

astro-ph↗

arXiv1996

Cosmic string formation and the power spectrum of field configurations

We examine the statistical properties of defects formed by the breaking of a U(1) symmetry when the Higgs field has a power spectrum $P(k) \propto k^n$. We find a marked dependence of the amount of infinite string on the spectral index $n$ and empirically identify an analytic form for this quantity. We also confirm that this result is robust to changes in the definition of infinite string. It is possible that this result could account for the apparent absence of infinite string in recent lattice-free simulations.

astro-ph↗

arXiv1996

A Statistic for identifying cosmic string wakes and other sheet-like structure

We describe an implementation of the structure functions of Babul \& Starkman, in order to quantify the ``sheet-like'' nature of a distribution of matter. We test this statistic on a toy model describing cosmic string wakes, and show that it does a better job than other statistics which have been proposed for distinguishing non-Gaussianity in the form of sheets. We conclude that the most favoured cosmic string model is unlikely to produce a significant increase in the sheet-like nature of the matter distribution beyond that which occurs in Gaussian models (with the same power spectrum) due to the formation of Zeldovich pancakes. Although the statistic was developed in the context of cosmic string wake formation, we expect it to be useful for comparing the observed galaxy distribution with a wide range of theoretical models with different power spectra.

astro-ph↗

arXiv1995

Causality and the Power Spectrum

We find constraints on the generation of super-causal-horizon energy perturbations from a smooth initial state, under a simple physical scheme. We quantify these constraints by placing the upper limit $λ_c = 3.0 d_H$ on the wavelength at which the power spectrum turns over to $k^4$ behavior. This means that sub-horizon processes can generate significant power on scales further outside the horizon than one might naively expect. The existence of this limit may have important implications for the interpretation of the small scale power spectrum of the Cosmic Microwave Background.

astro-ph↗