arXiv subjects

Mark Harris

Publications and source records attributed to Mark Harris.

3 recordsLinked to original sources

arXiv2026

A Scalable PyTorch Abstraction for Multi-GPU Gaussian Splatting

Gaussian splatting methods have become increasingly popular for neural reconstruction of the real world. However, they are often limited in scale and resolution due to compute and memory constraints. We present a multi-GPU Gaussian splatting approach that scales reconstruction to higher resolutions and larger scenes while abstracting away the code complexity typically associated with distributing a model. To accomplish this, we propose a PyTorch backend that distributes the Gaussian parameters and splatting operators across GPUs via CUDA unified memory and NVLink. Because distribution occurs at the operator level, the model code requires no explicit cross-device communication. More broadly, the backend exposes multiple GPUs as an aggregate PyTorch device and supports other PyTorch operators. We demonstrate city-scale reconstructions with street-level detail consisting of over 1 billion Gaussian splats, more than 25 times as many as the current state of the art.

cs.CV↗

arXiv2025

Using LLMs to support assessment of student work in higher education: a viva voce simulator

One of the emergent challenges of student work submitted for assessment is the widespread use of large language models (LLMs) to support and even produce written work. This particularly affects subjects where long-form written work is a key part of assessment. We propose a novel approach to addressing this challenge, using LLMs themselves to support the assessment process. We have developed a proof-of-concept viva voce examination simulator, which accepts the student's written submission as input, generates an interactive series of questions from the LLM and answers from the student. The viva voce simulator is an interactive tool which asks questions which a human examiner might plausibly ask, and uses the student's answers to form a judgment about whether the submitted piece of work is likely to be the student's own work. The interaction transcript is provided to the human examiner to support their final judgment. We suggest theoretical and practical points which are critical to real-world deployment of such a tool.

cs.CY↗

arXiv2023

Automated Speckle Interferometry of Known Binaries

Astronomers have been measuring the separations and position angles between the two components of binary stars since William Herschel began his observations in 1781. In 1970, Anton Labeyrie pioneered a method, speckle interferometry, that overcomes the usual resolution limits induced by atmospheric turbulence by taking hundreds or thousands of short exposures and reducing them in Fourier space. Our 2022 automation of speckle interferometry allowed us to use a fully robotic 1.0-meter PlaneWave Instruments telescope, located at the El Sauce Observatory in the Atacama Desert of Chile, to obtain observations of many known binaries with established orbits. The long-term objective of these observations is to establish the precision, accuracy, and limitations of this telescope's automated speckle interferometry measurements. This paper provides an early overview of the Known Binaries Project and provide example results on a small-separation (0.27") binary, WDS 12274-2843 B 228.

astro-ph.IM↗