arXiv subjects

Wesley Armour

Publications and source records attributed to Wesley Armour.

At least 19 recordsLinked to original sources

arXiv2026

Real-Time RFI Mitigation in SPOTLIGHT: A Two-Stage Approach for Transient Searches

Radio Frequency Interference (RFI) remains one of the primary challenges limiting the sensitivity and reliability of modern radio transient surveys, particularly for real-time searches of fast radio transients. The SPOTLIGHT system is a commensal real-time transient search backend operating at the upgraded Giant Metrewave Radio Telescope (uGMRT), where robust and computationally efficient RFI mitigation is essential for sustained operations. We present the real-time two-stage RFI mitigation framework developed for SPOTLIGHT, comprising an antenna-level voltage-filtering module (VOLT) operating prior to correlation beamforming and the SPOTLIGHT Time-domain RFI Processing Engine (STRIPE), a statistical RFI-filtering framework applied to beamformed data. Together, these complementary techniques mitigate a broad spectrum of RFI, ranging from broadband impulsive interference mitigated by VOLT to narrowband spectrally confined spurious signals mitigated by STRIPE, while remaining computationally efficient enough to satisfy the stringent requirements of real-time processing. The framework is evaluated using routine commensal GMRT observations, controlled 76 pulsar observations, and benchmarking against PRESTO's rfifind. The deployed system reduced the false detection rate by 98%. The recovered astrophysical pulses exhibit a 2.7x improvement in S/N after the two-stage filtering compared with the unfiltered data. These improvements enhance SPOTLIGHT's detection efficiency, sensitivity, and operational reliability, strengthening its capability to discover radio transients with the uGMRT.

astro-ph.IM↗

arXiv2026

MARS: A Lightweight Morphology-Aware RFI Segmentation Network for Mask-Guided Mitigation in Radio Astronomy

Next-generation radio telescopes generate filterbank data at rates that make storing all observations for later offline mitigation impractical. Mitigation must therefore operate in real or near-real time within the search pipeline while preserving dispersed astrophysical signals. CPU tools fit GPU-centred search pipelines poorly, while neural alternatives can be computationally heavy. We present MARS, a GPU-based RFI mitigation pipeline centred on a lightweight Morphology-Aware RFI Segmentation Network. The model is a reduced-width, full-resolution U-Net with a bottleneck containing local, horizontal, and vertical filters to capture compact and elongated RFI structures in the frequency-time plane. Normalisation, patch construction, mask reconstruction, replacement, baseline removal, and output rescaling are also implemented on GPU. Training includes an astronomical-signal preservation loss that discourages false flagging of dispersed pulses. In controlled patch-level tests, MARS achieves an RFI-mask F1 score of $0.978$ and a precision of $0.995$. It retains $97.6\%$ of the injected dispersed-signal fluence in clean patches and $96.4\%$ of the non-overlapping signal fluence in patches containing mixed injected RFI. Ablation experiments show that the astronomical-signal preservation loss particularly improves the protection of compact, low-DM, high-S/N pulses. At filterbank level, period-matched PRESTO candidates recovered after MARS mitigation have median significance ratios of $0.90$--$0.99$ relative to filtool. Both methods also recover the known pulsars in two real GMRT observations. On an NVIDIA GH200 GPU, MARS achieves a compute-only speedup of $6.2\times$--$7.0\times$ over the fastest tested multi-threaded filtool configurations on an AMD EPYC 9825 CPU.

astro-ph.IM↗

arXiv2026

A Geometric Lens on Physics-Aligned Data Compression

In AI for Science, physics-informed losses are increasingly used to train learned compressors for scientific data, but their rate-distortion implications remain poorly understood. At fixed bitrate, these objectives often improve preservation of a target physical observable while degrading standard reconstruction fidelity. We develop a local geometric theory showing that this tradeoff is governed by the interaction of latent-space sensitivities induced by the entropy model, the physical observable, and the distortion metric. At each operating point, these induce preferred directions along which compression noise should be suppressed, yielding an anisotropic error-allocation mechanism. When these directions are misaligned, improving the observable at fixed rate necessarily worsens standard distortion, establishing a fundamental limit on simultaneous preservation. We formalise this through a local tangent-space rate-distortion law and introduce a practical alignment diagnostic based on dominant eigenspace overlap. Experiments across scientific domains test the theory and validate that the alignment diagnostic correlates with observed data- and physics-space trade-offs.

cs.LG↗

arXiv2025

Beyond the Mean: Fisher-Orthogonal Projection for Natural Gradient Descent in Large Batch Training

Modern GPUs are equipped with large amounts of high-bandwidth memory, enabling them to support mini-batch sizes of up to tens of thousands of training samples. However, most existing optimizers struggle to perform effectively at such a large batch size. As batch size increases, gradient noise decreases due to averaging over many samples, limiting the ability of first-order methods to escape sharp or suboptimal minima and reach the global minimum. Meanwhile, second-order methods like the natural gradient with Kronecker-Factored Approximate Curvature (KFAC) often require excessively high damping to remain stable at large batch sizes. This high damping effectively washes out the curvature information that gives these methods their advantage, reducing their performance to that of simple gradient descent. In this paper, we introduce Fisher-Orthogonal Projection (FOP), a novel technique that restores the effectiveness of the second-order method at very large batch sizes, enabling scalable training with improved generalization and faster convergence. FOP constructs a variance-aware update direction by leveraging gradients from two sub-batches, enhancing the average gradient with a component of the gradient difference that is orthogonal to the average under the Fisher-metric.

cs.LG↗

arXiv2025

Adaptive Illumination-Invariant Synergistic Feature Integration in a Stratified Granular Framework for Visible-Infrared Re-Identification

Visible-Infrared Person Re-Identification (VI-ReID) plays a crucial role in applications such as search and rescue, infrastructure protection, and nighttime surveillance. However, it faces significant challenges due to modality discrepancies, varying illumination, and frequent occlusions. To overcome these obstacles, we propose \textbf{AMINet}, an Adaptive Modality Interaction Network. AMINet employs multi-granularity feature extraction to capture comprehensive identity attributes from both full-body and upper-body images, improving robustness against occlusions and background clutter. The model integrates an interactive feature fusion strategy for deep intra-modal and cross-modal alignment, enhancing generalization and effectively bridging the RGB-IR modality gap. Furthermore, AMINet utilizes phase congruency for robust, illumination-invariant feature extraction and incorporates an adaptive multi-scale kernel MMD to align feature distributions across varying scales. Extensive experiments on benchmark datasets demonstrate the effectiveness of our approach, achieving a Rank-1 accuracy of $74.75\%$ on SYSU-MM01, surpassing the baseline by $7.93\%$ and outperforming the current state-of-the-art by $3.95\%$.

cs.CV↗

arXiv2025

CRIUgpu: Transparent Checkpointing of GPU-Accelerated Workloads

Deep learning training at scale is resource-intensive and time-consuming, often running across hundreds or thousands of GPUs for weeks or months. Efficient checkpointing is crucial for running these workloads, especially in multi-tenant environments where compute resources are shared, and job preemptions or interruptions are common. However, transparent and unified GPU snapshots are particularly challenging because of the hardware architecture differences between CPU and GPU, including memory subsystems, dynamic parallelism, and thread synchronization. State-of-the-art GPU checkpointing techniques typically leverage mechanisms that intercept, log, and replay device API calls. However, this approach adds performance overhead and requires hardware-specific implementation that is difficult to test, maintain, and integrate with existing container platforms. In this paper, we present CRIUgpu - a novel approach for transparent checkpointing of GPU-accelerated workloads that builds on recently introduced driver capabilities, enabling support for CUDA and ROCm applications. Our evaluation results show that CRIUgpu works with a variety of deep learning and high-performance computing workloads running across multiple GPUs, completely eliminating steady-state performance overheads, and significantly reducing recovery times compared to state-of-the-art transparent GPU checkpointing mechanisms.

cs.DC↗

arXiv2024

Distribution of the H\"ogbom CLEAN Algorithm Using Tiled Images with Feedback

Data sizes for next generation radio telescopes, such as the Square Kilometre Array (SKA), are far above that of their predecessors. The CLEAN algorithm was originally developed by H\"ogbom [1974], long before such data sizes were thought possible and is still the most popular tool used for deconvolution in interferometric imaging. In order to facilitate these new large data sizes and reduce computation time a distributed approach to the algorithm has been investigated. The serial nature of the CLEAN algorithm, due to its matching pursuit design, makes this challenging. Splitting the image into a number of tiles which can be individually deconvolved has been investigated, but this creates discontinuities in the deconvolved image and makes it difficult to deconvolve faint sources in the presence of a point spread function associated with bright sources in other tiles. A method of feedback between each of the tiles has been developed to deal with these problems. This new approach has been tested on a simulated dataset containing multiple point sources of known intensity. When compared to a standard H\"ogbom deconvolution the tiled feedback version produced a reconstructed image, containing sources up to 2.1 Jy, which agreed to between -0.1 Jy and +0.04 Jy of the standard method across the whole deconvolved image at a speed up to 10.66 times faster.

astro-ph.IM↗

arXiv2024

Lowering In-Memory Footprint of Antenna Beams via Polynomial Approximation

With the emergence of new radio telescopes promising larger fields of view at lower observation frequencies (e.g., SKA), addressing direction-dependent effects (DDE) (e.g., direction-specific beam responses), polarisation leakage, and pointing errors has become all the more important. Be it through A-projection or otherwise, addressing said effects often requires reliable representations of antenna/station beams; yet, these require significant amounts of computational memory as they are baseline-, frequency-, time-, and polarisation-dependent. A novel prototype is reported here to approximate antenna beams suitable for SKA-MID using Zernike polynomials. It is shown that beam kernels can be well approximated, paving the way for future optimisations towards facilitating more efficient beam-dependent solutions and approaches to tackling the aforementioned challenges, all of which are essential for large-scale radio telescopes.

astro-ph.IM↗

arXiv2024

Double-Exponential Increases in Inference Energy: The Cost of the Race for Accuracy

Deep learning models in computer vision have achieved significant success but pose increasing concerns about energy consumption and sustainability. Despite these concerns, there is a lack of comprehensive understanding of their energy efficiency during inference. In this study, we conduct a comprehensive analysis of the inference energy consumption of 1,200 ImageNet classification models - the largest evaluation of its kind to date. Our findings reveal a steep diminishing return in accuracy gains relative to the increase in energy usage, highlighting sustainability concerns in the pursuit of marginal improvements. We identify key factors contributing to energy consumption and demonstrate methods to improve energy efficiency. To promote more sustainable AI practices, we introduce an energy efficiency scoring system and develop an interactive web application that allows users to compare models based on accuracy and energy consumption. By providing extensive empirical data and practical tools, we aim to facilitate informed decision-making and encourage collaborative efforts in developing energy-efficient AI technologies.

cs.CV↗

arXiv2024

Pulscan: Binary pulsar detection using unmatched filters on NVIDIA GPUs

The Fourier Domain Acceleration Search (FDAS) and Fourier Domain Jerk Search (FDJS) are proven matched filtering techniques for detecting binary pulsar signatures in time-domain radio astronomy datasets. Next generation radio telescopes such as the SPOTLIGHT project at the GMRT produce data at rates that mandate real-time processing, as storage of the entire captured dataset for subsequent offline processing is infeasible. The computational demands of FDAS and FDJS make them challenging to implement in real-time detection pipelines, requiring costly high performance computing facilities. To address this we propose Pulscan, an unmatched filtering approach which achieves order-of-magnitude improvements in runtime performance compared to FDAS whilst being able to detect both accelerated and some jerked binary pulsars. We profile the sensitivity of Pulscan using a distribution (N = 10,955) of synthetic binary pulsars and compare its performance with FDAS and FDJS. Our implementation of Pulscan includes an OpenMP version for multicore CPU acceleration, a version for heterogeneous CPU/GPU environments such as NVIDIA Grace Hopper, and a fully optimized NVIDIA GPU implementation for integration into an AstroAccelerate pipeline, which will be deployed in the SPOTLIGHT project at the GMRT. Our results demonstrate that unmatched filtering in Pulscan can serve as an efficient data reduction step, prioritizing datasets for further analysis and focusing human and subsequent computational resources on likely binary pulsar signatures.

astro-ph.IM↗

arXiv2024

CLEAN algorithm implementation comparisons between popular software packages

The CLEAN algorithm, first published by H\"{o}gbom and its later variants such as Multiscale CLEAN (msCLEAN) by Cornwell, has been the most popular tool for deconvolution in radio astronomy. Interferometric imaging used in aperture synthesis radio telescopes requires deconvolution for removal of the telescopes point spread function from the observed images. We have compared source fluxes produced by different implementations of H\"{o}gbom and msCLEAN (WSCLEAN, CASA) with a prototype implementation of H\"{o}gbom and msCLEAN for the Square Kilometer Array (SKA) on two datasets. First is a simulation of multiple point sources of known intensity using H\"{o}gbom, where none of the software packages detected all the simulated point sources to within 1.0% of the simulated values. The second is of supernova remnant G055.7+3.4 taken by the Karl G. Jansky Very Large Array (VLA) using msCLEAN, where each of the software packages produced different images for the same settings.

astro-ph.IM↗

arXiv2024

Toward using GANs in astrophysical Monte-Carlo simulations

Accurate modelling of spectra produced by X-ray sources requires the use of Monte-Carlo simulations. These simulations need to evaluate physical processes, such as those occurring in accretion processes around compact objects by sampling a number of different probability distributions. This is computationally time-consuming and could be sped up if replaced by neural networks. We demonstrate, on an example of the Maxwell-J\"uttner distribution that describes the speed of relativistic electrons, that the generative adversarial network (GAN) is capable of statistically replicating the distribution. The average value of the Kolmogorov-Smirnov test is 0.5 for samples generated by the neural network, showing that the generated distribution cannot be distinguished from the true distribution.

astro-ph.HE↗

arXiv2023

Part-time Power Measurements: nvidia-smi's Lack of Attention

The GPU has emerged as the go-to accelerator for high throughput and parallel workloads, spanning scientific simulations to AI, thanks to its performance and power efficiency. Given that 6 out of the top 10 fastest supercomputers in the world use NVIDIA GPUs and many AI companies each employ 10,000's of NVIDIA GPUs, an accurate understanding of GPU power consumption is essential for making progress to further improve its efficiency. Despite the limited documentation and the lack of understanding of its mechanisms, NVIDIA GPUs' built-in power sensor, providing easily accessible power readings via the nvidia-smi interface, is widely used in energy efficient computing research on GPUs. Our study seeks to elucidate the internal mechanisms of the power readings provided by nvidia-smi and assess the accuracy of the power and energy consumption data. We have developed a suite of micro-benchmarks to profile the behaviour of nvidia-smi power readings and have evaluated them on over 70 different GPUs from all architectural generations since power measurement was first introduced in the 'Fermi' generation. We have identified several unforeseen problems in terms of power/energy measurement using nvidia-smi, for example on the A100 and H100 GPUs only 25% of the runtime is sampled for power consumption, during the other 75% of the time, the GPU can be using drastically different power and nvidia-smi and results presented by it are unaware of this. This along with other findings can lead to a drastic under/overestimation of energy consumed, especially when considering data centres housing tens of thousands of GPUs. We proposed several good practices that help to mitigate these problems. By comparing our results to those measured from an external power-meter, we have reduced the error in the energy measurement by an average of 35% and in some cases by as much as 65% in the test cases we present.

cs.DC↗

arXiv2023

Accelerating Dedispersion using Many-Core Architectures

Astrophysical radio signals are excellent probes of extreme physical processes that emit them. However, to reach Earth, electromagnetic radiation passes through the ionised interstellar medium (ISM), introducing a frequency-dependent time delay (dispersion) to the emitted signal. Removing dispersion enables searches for transient signals like Fast Radio Bursts (FRB) or repeating signals from isolated pulsars or those in orbit around other compact objects. The sheer volume and high resolution of data that next generation radio telescopes will produce require High-Performance Computing (HPC) solutions and algorithms to be used in time-domain data processing pipelines to extract scientifically valuable results in real-time. This paper presents a state-of-the-art implementation of brute force incoherent dedispersion on NVIDIA GPUs, and on Intel and AMD CPUs. We show that our implementation is 4x faster (8-bit 8192 channels input) than other available solutions and demonstrate, using 11 existing telescopes, that our implementation is at least 20 faster than real-time. This work is part of the AstroAccelerate package.

astro-ph.IM↗

arXiv2022

Bits missing: Finding exotic pulsars using bfloat16 on NVIDIA GPUs

The Fourier Domain Acceleration Search (FDAS) is an effective technique for detecting faint binary pulsars in large radio astronomy datasets. This paper quantifies the sensitivity impact of reducing numerical precision in the GPU accelerated FDAS pipeline of the AstroAccelerate software package. The prior implementation used IEEE-754 single-precision in the entire binary pulsar detection pipeline, spending a large fraction of the runtime computing GPU accelerated FFTs. AstroAccelerate has been modified to use bfloat16 (and IEEE754 double-precision to provide a "gold standard" comparison) within the Fourier domain convolution section of the FDAS routine. Approximately 20,000 synthetic pulsar filterbank files representing binary pulsars were generated using SIGPROC with a range of physical parameters. They have been processed using bfloat16, single and double-precision convolutions. All bfloat16 peaks are within 3% of the predicted signal-to-noise ratio of their corresponding single-precision peaks. Of 14,971 "bright" single-precision fundamental peaks above a power of 44.982 (our experimentally measured highest noise value), 14,602 (97.53%) have a peak in the same acceleration and frequency bin in the bfloat16 output plane, whilst in the remaining 369 the nearest peak is located in the adjacent acceleration bin. There is no bin drift measured between the single and double-precision results. The bfloat16 version of FDAS achieves a speedup of approximately 1.6x compared to single-precision. A comparison between AstroAccelerate and the PRESTO software package is presented using observations collected with the GMRT of PSR J1544+4937, a 2.16ms black widow pulsar in a 2.8 hour compact orbit.

astro-ph.IM↗

arXiv2022

A Novel Greedy Approach To Harmonic Summing Using GPUs

Incoherent harmonic summing is a technique which is used to improve the sensitivity of Fourier domain search methods. A one dimensional harmonic sum is used in time-domain radio astronomy as part of the Fourier domain periodicity search, a type of search used to detect isolated single pulsars. The main problem faced when implementing the harmonic sum on many-core architectures, like GPUs, is the very unfavourable memory access pattern of the harmonic sum algorithm. The memory access pattern gets worse as the dimensionality of the harmonic sum increases. Here we present a set of algorithms for calculating the harmonic sum that are suited to many-core architectures such as GPUs. We present an evaluation of the sensitivity of these different approaches, and their performance. This work forms part of the AstroAccelerate project which is a GPU accelerated software package for processing time-domain radio astronomy data.

astro-ph.IM↗

arXiv2021

Efficiency Near the Edge: Increasing the Energy Efficiency of FFTs on GPUs for Real-time Edge Computing

The Square Kilometre Array (SKA) is an international initiative for developing the world's largest radio telescope with a total collecting area of over a million square meters. The scale of the operation, combined with the remote location of the telescope, requires the use of energy-efficient computational algorithms. This, along with the extreme data rates that will be produced by the SKA and the requirement for a real-time observing capability, necessitates in-situ data processing in an edge style computing solution. More generally, energy efficiency in the modern computing landscape is becoming of paramount concern. Whether it be the power budget that can limit some of the world's largest supercomputers, or the limited power available to the smallest Internet-of-Things devices. In this paper, we study the impact of hardware frequency scaling on the energy consumption and execution time of the Fast Fourier Transform (FFT) on NVIDIA GPUs using the cuFFT library. The FFT is used in many areas of science and it is one of the key algorithms used in radio astronomy data processing pipelines. Through the use of frequency scaling, we show that we can lower the power consumption of the NVIDIA V100 GPU when computing the FFT by up to 60% compared to the boost clock frequency, with less than a 10% increase in the execution time. Furthermore, using one common core clock frequency for all tested FFT lengths, we show on average a 50% reduction in power consumption compared to the boost core clock frequency with an increase in the execution time still below 10%. We demonstrate how these results can be used to lower the power consumption of existing data processing pipelines. These savings, when considered over years of operation, can yield significant financial savings, but can also lead to a significant reduction of greenhouse gas emissions.

cs.PF↗

arXiv2021

Implementation of 3D degridding algorithm on the NVIDIA GPUs using CUDA

Practical aperture synthesis imaging algorithms work by iterating between estimating the sky brightness distribution and a comparison of a prediction based on this estimate with the measured data ("visibilities"). Accuracy in the latter step is crucial but is made difficult by irregular and non-planar sampling of data by the telescope. In this work we present a GPU implementation of 3d de-gridding which accurately deals with these two difficulties and is designed for distributed operation. We address the load balancing issues caused by large variation in visibilities that need to be computed. Using CUDA and NVidia GPUs we measure performance up to 1.2 billion visibilities per second.

astro-ph.IM↗