arXiv subjects

James Bishop

Publications and source records attributed to James Bishop.

10 recordsLinked to original sources

arXiv2026

The Role of Ambiguity in Error Prediction via Uncertainty Quantification

The task of Error Prediction, namely predicting whether a model output is correct, is commonly tackled with Uncertainty Quantification (UQ). However, while uncertainty metrics capture when models lack knowledge or capacity to make a prediction, they also reflect aleatoric uncertainty, which is inherent in the model input and context. This paper presents a method for improving error prediction for Large Language Models (LLMs), by disentangling input ambiguity from UQ signal. We conduct experiments on the task of Question Answering (QA) with six UQ metrics and show that UQ metrics are more predictive of errors on unambiguous instances than on questions with multiple plausible answers. We use Gated Experts and Selective Prediction to incorporate gold and predicted ambiguity labels into the error prediction pipeline. We find that ambiguity information improves error prediction scores across model families, training and evaluation paradigms, datasets (including allegedly unambiguous ones), and sources of aleatoric uncertainty, yielding improvements of over 10 points of PRR for individual UQ metrics on standard datasets.

cs.CL↗

arXiv2025

Correcting Prompt Dependence in LLM Benchmarks: A Bayesian Hierarchical Model with Embedding-Space Clustering

LLM benchmarking metrics often misstate performance and uncertainty as they rely on two assumptions that frequently do not hold in practice: (i) a sufficient number of evaluations are available for classical inference, and (ii) test prompts are independent. We propose a corrective Bayesian hierarchical model with embedding-space clustering that provides robust performance metrics in limited-data settings while correcting for prompt dependence. We apply the approach to adversarial robustness benchmarks, showing consistent recovery of clustering structure, resulting in more reliable performance metrics, with 4-73% improvements to mean absolute errors and 40-450 unit improvements to expected log posterior densities.

cs.CR↗

arXiv2025

On Corrigibility and Alignment in Multi Agent Games

Corrigibility of autonomous agents is an under explored part of system design, with previous work focusing on single agent systems. It has been suggested that uncertainty over the human preferences acts to keep the agents corrigible, even in the face of human irrationality. We present a general framework for modelling corrigibility in a multi-agent setting as a 2 player game in which the agents always have a move in which they can ask the human for supervision. This is formulated as a Bayesian game for the purpose of introducing uncertainty over the human beliefs. We further analyse two specific cases. First, a two player corrigibility game, in which we want corrigibility displayed in both agents for both common payoff (monotone) games and harmonic games. Then we investigate an adversary setting, in which one agent is considered to be a `defending' agent and the other an `adversary'. A general result is provided for what belief over the games and human rationality the defending agent is required to have to induce corrigibility.

cs.GT↗

arXiv2024

Biosignatures from pre-oxygen photosynthesising life on TRAPPIST-1e

In order to assess observational evidence for potential atmospheric biosignatures on exoplanets, it will be essential to test whether spectral fingerprints from multiple gases can be explained by abiotic or biotic-only processes. Here, we develop and apply a coupled 1D atmosphere-ocean-ecosystem model to understand how primitive biospheres, which exploit abiotic sources of H2, CO and O2, could influence the atmospheric composition of rocky terrestrial exoplanets. We apply this to the Earth at 3.8 Ga and to TRAPPIST-1e. We focus on metabolisms that evolved before the evolution of oxygenic photosynthesis, which consume H2 and CO and produce potentially detectable levels of CH4. O2-consuming metabolisms are also considered for TRAPPIST-1e, as abiotic O2 production is predicted on M-dwarf orbiting planets. We show that these biospheres can lead to high levels of surface O2 (approximately 1-5 %) as a result of \ch{CO} consumption, which could allow high O2 scenarios, by removing the main loss mechanisms of atomic oxygen. Increasing stratospheric temperatures, which increases atmospheric OH can reduce the likelihood of such a state forming. O2-consuming metabolisms could also lower O2 levels to around 10 ppm and support a productive biosphere at low reductant inputs. Using predicted transmission spectral features from CH4, CO, O2/O3 and CO2 across the hypothesis space for tectonic reductant input, we show that biotically-produced CH4 may only be detectable at high reductant inputs. CO is also likely to be a dominant feature in transmission spectra for planets orbiting M-dwarfs, which could reduce the confidence in any potential biosignature observations linked to these biospheres.

astro-ph.EP↗

arXiv2024

Hydrogen plasma inhibits ion beam restructuring of materials

Focused ion beam (FIB) techniques are employed widely for nanofabrication, and processing of materials and devices. However, ion irradiation often gives rise to severe damage due to atomic displacements that cause defect formation, migration and clustering within the ion-solid interaction volume. The resulting restructuring degrades the functionality of materials, and limits the utility FIB ablation and nanofabrication techniques. Here we show that such restructuring can be inhibited by performing FIB irradiation in a hydrogen plasma environment via chemical pathways that modify defect binding energies and transport kinetics, as well as material ablation rates. The method is minimally-invasive and has the potential to greatly expand the utility of FIB nanofabrication techniques in processing of functional materials and devices.

physics.app-ph↗

arXiv2022

Suppression of surface roughening during ion bombardment of semiconductors

Ion beams are used routinely for processing of semiconductors, particularly sputtering, ion implantation and direct-write fabrication of nanostructures. However, the utility of ion beam techniques is limited by crystal damage and surface roughening. Damage can be reduced or eliminated by performing irradiation at elevated temperatures. However, at these conditions, surface roughening is highly problematic due to thermal mobility of adatoms and surface vacancies. Here we solve this problem using hydrogen gas, which we use to stabilize surface mass flow and suppress roughening during ion bombardment of elemental and compound semiconductors. We achieve smooth surfaces during ion-beam processing, and show that the method can be enhanced by radicalizing H2 gas using a remote plasma source. Our approach is broadly applicable, and expands the utility of ion beam techniques for the processing and fabrication of functional materials and nanostructures.

cond-mat.mtrl-sci↗

arXiv2021

Recoil Implantation Using Gas-Phase Precursor Molecules

Ion implantation underpins a vast range of devices and technologies that require precise control over the physical, chemical, electronic, magnetic and optical properties of materials. A variant termed recoil implantation - in which a precursor is deposited onto a substrate as a thin film and implanted via momentum transfer from incident energetic ions - has a number of compelling advantages, particularly when performed using an inert ion nano-beam [Fröch et al., Nat Commun 11, 5039 (2020)]. However, a major drawback of this approach is that the implant species are limited to the constituents of solid thin films. Here we overcome this limitation by demonstrating recoil implantation using gas-phase precursors. Specifically, we fabricate nitrogen-vacancy (NV) color centers in diamond using an Ar ion beam and the nitrogen-containing precursor gases N2, NH3 and NF3. Our work expands the applicability of recoil implantation to most of the periodic table, and to applications in which thin film deposition or removal is impractical.

physics.app-ph↗

arXiv2018

Bottom up engineering of near-identical quantum emitters in atomically thin materials

Quantum technologies require robust and photostable single photon emitters (SPEs) that can be reliably engineered. Hexagonal boron nitride (hBN) has recently emerged as a promising candidate host to bright and optically stable SPEs operating at room temperature. However, the emission wavelength of the fluorescent defects in hBN has, to date, been shown to be uncontrolled. The emitters usually display a large spread of zero phonon line (ZPL) energies spanning over a broad spectral range (hundreds of nanometers), which hinders the potential development of hBN-based devices and applications. We demonstrate bottom-up, chemical vapor deposition growth of large-area, few layer hBN that hosts large quantities of SPEs: 100 per 10x10 μm2. Remarkably, more than 85 percent of the emitters have a ZPL at (580{\pm}10)nm, a distribution which is over an order of magnitude narrower than previously reported. Exploiting the high density and uniformity of the emitters, we demonstrate electrical modulation and tuning of the ZPL emission wavelength by up to 15 nm. Our results constitute a definite advancement towards the practical deployment of hBN single photon emitters in scalable quantum photonic and hybrid optoelectronic devices based on 2D materials.

physics.app-ph↗

arXiv2018

Photonic Crystal Cavities from Hexagonal Boron Nitride

Development of scalable quantum photonic technologies requires on-chip integration of components such as photonic crystal cavities and waveguides with nonclassical light sources. Recently, hexagonal boron nitride (hBN) has emerged as a promising platform for nanophotonics, following reports of hyperbolic phonon-polaritons and optically stable, ultra-bright quantum emitters. However, exploitation of hBN in scalable, on-chip nanophotonic circuits, quantum information processing and cavity quantum electrodynamics (QED) experiments requires robust techniques for the fabrication of monolithic optical resonators. In this letter, we design and engineer high quality photonic crystal cavities from hBN. We employ two approaches based on a focused ion beam method and a minimally-invasive electron beam induced etching (EBIE) technique to fabricate suspended two dimensional (2D) and one dimensional (1D) cavities with quality (Q) factors in excess of 2,000. Subsequently, we show deterministic, iterative tuning of individual cavities by direct-write, single-step EBIE without significant degradation of the Q-factor. The demonstration of tunable, high Q cavities made from hBN is an unprecedented advance in nanophotonics based on van der Waals materials. Our results and hBN processing methods open up promising new avenues for solid-state systems with applications in integrated quantum photonics, polaritonics and cavity QED experiments.

physics.app-ph↗

arXiv2015

Emergent formation of dynamic topographic patterns in electron beam induced etching

Spontaneous formation of geometric patterns is a fascinating, ubiquitous process that provides fundamental insights into the roles of symmetry breaking, anisotropy and nonlinear interactions in emergent phenomena. Here we report dynamic, highly ordered topographic patterns on the surface of diamond that span multiple length scales and have a symmetry controlled by the chemical species of a precursor gas used in electron beam induced etching (EBIE). This behavior reveals an underlying etch rate anisotropy and an electron energy transfer pathway that has been overlooked by existing EBIE theory. We present an etch rate kinetics model that fully explains our results and is universally applicable to EBIE. Our findings can be exploited for controlled wetting, optical structuring and other emerging applications that require nano and micro-scale surface texturing.

cond-mat.mtrl-sci↗