arXiv subjects

Daniel Wilson

Publications and source records attributed to Daniel Wilson.

18 recordsLinked to original sources

arXiv2026

Commissioning and on-sky performance verification of iLocater

iLocater is a high-resolution near-infrared extreme precision radial velocity (EPRV) spectrograph that was deployed to the Large Binocular Telescope (LBT) in June 2026. iLocater operates over $\lambda=966-1312$ nm with a median resolving power of $R=205,000$ as measured in the laboratory. We present the commissioning and initial on-sky verification program using solar and night-time observations at the LBT. First light was achieved on 27 June 2026, and nearly 150 on-sky spectra have now been recorded. Observations include single stars ranging in spectral type from B5 to M6. iLocater uses the LBT AO system, and we have demonstrated its ability to obtain spatially resolved spectra of close ($\theta < 1''$) binary stars.

astro-ph.IM↗

arXiv2026

Power-Flexible AI Data Centers: A New Paradigm for Grid-Responsive Compute

The rapid expansion of artificial intelligence (AI) infrastructure is driving unprecedented growth in electricity demand from data centers. Traditional power-system planning treats large computing facilities as inflexible peak loads, leading to costly infrastructure upgrades and long delays in grid interconnection. Recent work has shown that AI clusters can reduce electricity consumption during peak demand through software-based workload orchestration. This article explores how modern GPU-based AI data centers can operate as grid-interactive assets that respond dynamically to power system conditions. We describe an architecture integrating grid signals, workload scheduling, and power telemetry for fine-grained cluster power control. Experimental results from a real-world deployment on a 130 kW GPU cluster demonstrate multiple forms of flexibility, including rapid load reduction, sustained curtailment, and carbon-aware operation while preserving service levels for priority jobs. We further demonstrate performance-aware load shifting across geographically distributed clusters, enabling workloads to migrate toward regions with lower grid stress. Together, these capabilities transform AI infrastructure from static electricity consumers into flexible resources that support grid reliability, accelerate interconnection, and improve computing sustainability.

cs.DC↗

arXiv2025

Testing the Limits of Machine Translation from One Book

Current state-of-the-art models demonstrate capacity to leverage in-context learning to translate into previously unseen language contexts. Tanzer et al. [2024] utilize language materials (e.g. a grammar) to improve translation quality for Kalamang using large language models (LLMs). We focus on Kanuri, a language that, despite having substantial speaker population, has minimal digital resources. We design two datasets for evaluation: one focused on health and humanitarian terms, and another containing generalized terminology, investigating how domain-specific tasks impact LLM translation quality. By providing different combinations of language resources (grammar, dictionary, and parallel sentences), we measure LLM translation effectiveness, comparing results to native speaker translations and human linguist performance. We evaluate using both automatic metrics and native speaker assessments of fluency and accuracy. Results demonstrate that parallel sentences remain the most effective data source, outperforming other methods in human evaluations and automatic metrics. While incorporating grammar improves over zero-shot translation, it fails as an effective standalone data source. Human evaluations reveal that LLMs achieve accuracy (meaning) more effectively than fluency (grammaticality). These findings suggest LLM translation evaluation benefits from multidimensional assessment beyond simple accuracy metrics, and that grammar alone, without parallel sentences, does not provide sufficient context for effective domain-specific translation.

cs.CL↗

arXiv2023

M\=aori algorithmic sovereignty: idea, principles, and use

Due to the emergence of data-driven technologies in Aotearoa New Zealand that use M\=aori data, there is a need for values-based frameworks to guide thinking around balancing the tension between the opportunities these create, and the inherent risks that these technologies can impose. Algorithms can be framed as a particular use of data, therefore data frameworks that currently exist can be extended to include algorithms. M\=aori data sovereignty principles are well-known and are used by researchers and government agencies to guide the culturally appropriate use of M\=aori data. Extending these principles to fit the context of algorithms, and re-working the underlying sub-principles to address issues related to responsible algorithms from a M\=aori perspective leads to the M\=aori algorithmic sovereignty principles. We define this idea, present the updated principles and subprinciples, and highlight how these can be used to decolonise algorithms currently in use, and argue that these ideas could potentially be used to developed Indigenised algorithms.

cs.CY↗

arXiv2022

Deep Riemannian Networks for End-to-End EEG Decoding

State-of-the-art performance in electroencephalography (EEG) decoding tasks is currently often achieved with either Deep-Learning (DL) or Riemannian-Geometry-based decoders (RBDs). Recently, there is growing interest in Deep Riemannian Networks (DRNs) possibly combining the advantages of both previous classes of methods. However, there are still a range of topics where additional insight is needed to pave the way for a more widespread application of DRNs in EEG. These include architecture design questions such as network size and end-to-end ability. How these factors affect model performance has not been explored. Additionally, it is not clear how the data within these networks is transformed, and whether this would correlate with traditional EEG decoding. Our study aims to lay the groundwork in the area of these topics through the analysis of DRNs for EEG with a wide range of hyperparameters. Networks were tested on five public EEG datasets and compared with state-of-the-art ConvNets. Here we propose EE(G)-SPDNet, and we show that this wide, end-to-end DRN can outperform the ConvNets, and in doing so use physiologically plausible frequency regions. We also show that the end-to-end approach learns more complex filters than traditional band-pass filters targeting the classical alpha, beta, and gamma frequency bands of the EEG, and that performance can benefit from channel specific filtering approaches. Additionally, architectural analysis revealed areas for further improvement due to the possible under utilisation of Riemannian specific information throughout the network. Our study thus shows how to design and train DRNs to infer task-related information from the raw EEG without the need of handcrafted filterbanks and highlights the potential of end-to-end DRNs such as EE(G)-SPDNet for high-performance EEG decoding.

cs.LG↗

arXiv2021

Visual and Object Geo-localization: A Comprehensive Survey

The concept of geo-localization refers to the process of determining where on earth some `entity' is located, typically using Global Positioning System (GPS) coordinates. The entity of interest may be an image, sequence of images, a video, satellite image, or even objects visible within the image. As massive datasets of GPS tagged media have rapidly become available due to smartphones and the internet, and deep learning has risen to enhance the performance capabilities of machine learning models, the fields of visual and object geo-localization have emerged due to its significant impact on a wide range of applications such as augmented reality, robotics, self-driving vehicles, road maintenance, and 3D reconstruction. This paper provides a comprehensive survey of geo-localization involving images, which involves either determining from where an image has been captured (Image geo-localization) or geo-locating objects within an image (Object geo-localization). We will provide an in-depth study, including a summary of popular algorithms, a description of proposed datasets, and an analysis of performance results to illustrate the current state of each field.

cs.CV↗

arXiv2021

Object Tracking and Geo-localization from Street Images

Geo-localizing static objects from street images is challenging but also very important for road asset mapping and autonomous driving. In this paper we present a two-stage framework that detects and geolocalizes traffic signs from low frame rate street videos. Our proposed system uses a modified version of RetinaNet (GPS-RetinaNet), which predicts a positional offset for each sign relative to the camera, in addition to performing the standard classification and bounding box regression. Candidate sign detections from GPS-RetinaNet are condensed into geolocalized signs by our custom tracker, which consists of a learned metric network and a variant of the Hungarian Algorithm. Our metric network estimates the similarity between pairs of detections, then the Hungarian Algorithm matches detections across images using the similarity scores provided by the metric network. Our models were trained using an updated version of the ARTS dataset, which contains 25,544 images and 47.589 sign annotations ~\cite{arts}. The proposed dataset covers a diverse set of environments gathered from a broad selection of roads. Each annotaiton contains a sign class label, its geospatial location, an assembly label, a side of road indicator, and unique identifiers that aid in the evaluation. This dataset will support future progress in the field, and the proposed system demonstrates how to take advantage of some of the unique characteristics of a realistic geolocalization dataset.

cs.CV↗

arXiv2021

Points2Polygons: Context-Based Segmentation from Weak Labels Using Adversarial Networks

In applied image segmentation tasks, the ability to provide numerous and precise labels for training is paramount to the accuracy of the model at inference time. However, this overhead is often neglected, and recently proposed segmentation architectures rely heavily on the availability and fidelity of ground truth labels to achieve state-of-the-art accuracies. Failure to acknowledge the difficulty in creating adequate ground truths can lead to an over-reliance on pre-trained models or a lack of adoption in real-world applications. We introduce Points2Polygons (P2P), a model which makes use of contextual metric learning techniques that directly addresses this problem. Points2Polygons performs well against existing fully-supervised segmentation baselines with limited training data, despite using lightweight segmentation models (U-Net with a ResNet18 backbone) and having access to only weak labels in the form of object centroids and no pre-training. We demonstrate this on several different small but non-trivial datasets. We show that metric learning using contextual data provides key insights for self-supervised tasks in general, and allow segmentation models to easily generalize across traditionally label-intensive domains in computer vision.

cs.CV↗

arXiv2020

Machine-Learning-Based Diagnostics of EEG Pathology

Machine learning (ML) methods have the potential to automate clinical EEG analysis. They can be categorized into feature-based (with handcrafted features), and end-to-end approaches (with learned features). Previous studies on EEG pathology decoding have typically analyzed a limited number of features, decoders, or both. For a I) more elaborate feature-based EEG analysis, and II) in-depth comparisons of both approaches, here we first develop a comprehensive feature-based framework, and then compare this framework to state-of-the-art end-to-end methods. To this aim, we apply the proposed feature-based framework and deep neural networks including an EEG-optimized temporal convolutional network (TCN) to the task of pathological versus non-pathological EEG classification. For a robust comparison, we chose the Temple University Hospital (TUH) Abnormal EEG Corpus (v2.0.0), which contains approximately 3000 EEG recordings. The results demonstrate that the proposed feature-based decoding framework can achieve accuracies on the same level as state-of-the-art deep neural networks. We find accuracies across both approaches in an astonishingly narrow range from 81--86\%. Moreover, visualizations and analyses indicated that both approaches used similar aspects of the data, e.g., delta and theta band power at temporal electrode locations. We argue that the accuracies of current binary EEG pathology decoders could saturate near 90\% due to the imperfect inter-rater agreement of the clinical labels, and that such decoders are already clinically useful, such as in areas where clinical EEG experts are rare. We make the proposed feature-based framework available open source and thus offer a new tool for EEG machine learning research.

eess.IV↗

arXiv2019

Artificial Intelligence Approaches

Artificial Intelligence (AI) has received tremendous attention from academia, industry, and the general public in recent years. The integration of geography and AI, or GeoAI, provides novel approaches for addressing a variety of problems in the natural environment and our human society. This entry briefly reviews the recent development of AI with a focus on machine learning and deep learning approaches. We discuss the integration of AI with geography and particularly geographic information science, and present a number of GeoAI applications and possible future directions.

cs.AI↗

arXiv2019

Change of Variables with Local Time on Surfaces for Jump Processes

The `local time on curves' formula of Peskir provides a stochastic change of variables formula for a function whose derivatives may be discontinuous over a time-dependent curve, a setting which occurs often in applications in optimal control and beyond. This formula was further extended to higher dimensions and to include processes with jumps under conditions which may be hard to verify in practice. We build upon the work of Du Toit in weakening the required conditions by allowing semimartingales with jumps. In addition, under vanishing of the sectional first derivative (the so-called `smooth fit' condition), we show that the classical It\^o formula still holds under general conditions.

math.PR↗

arXiv2018

The Local Time-Space Integral and Stochastic Differential Equations

Processes which arise as solutions to stochastic differential equations involving the local time (SDELTs), such as skew Brownian motion, are frequent sources of inspiration in theory and applications. Existence and uniqueness results for such equations rely heavily on the It\^o-Tanaka formula. Recent interest in time-inhomogeneous SDELTs indicates the need for comprehensive existence and uniqueness results in the time-dependent case, however, the absence of a suitable time-dependent It\^o-Tanaka formula forms a major barrier. Rigorously developing a two-parameter integral with respect to local time, known as the local time-space integral, we connect together and extend many known formulae from the literature and establish a general time-dependent It\^o-Tanaka formula. Then, we prove the existence of a unique strong solution for a large class of time-inhomogeneous SDELTs.

math.PR↗

arXiv2018

Multi-resolution dimer models in heat baths with short-range and long-range interactions

This work investigates multi-resolution methodologies for simulating dimer models. The solvent particles which make up the heat bath interact with the monomers of the dimer either through direct collisions (short-range) or through harmonic springs (long-range). Two types of multi-resolution methodologies are considered in detail: (a) describing parts of the solvent far away from the dimer by a coarser approach; (b) describing each monomer of the dimer by using a model with different level of resolution. These methodologies are then utilised to investigate the effect of a shared heat bath versus two uncoupled heat baths, one for each monomer. Furthermore the validity of the multi-resolution methods is discussed by comparison to dynamics of macroscopic Langevin equations.

physics.comp-ph↗

arXiv2018

Solid-Immersion Metalenses for Infrared Focal Plane Arrays

Novel optical components based on metasurfaces (metalenses) offer a new methodology for microlens arrays. In particular, metalens arrays have the potential of being monolithically integrated with infrared focal plane arrays (IR FPAs) to increase the operating temperature and sensitivity of the latter. In this work, we demonstrate a new type of transmissive metalens that focuses the incident light ({\lambda} = 3-5 {\mu}m) on the detector plane after propagating through the substrate, i.e. solid-immersion type of focusing. The metalens is fabricated by etching the backside of the detector substrate material (GaSb here) making this approach compatible with the architecture of back-illuminated FPAs. In addition, our designs work for all incident polarizations. We fabricate a 10x10 metalens array that proves the scalability of this approach for FPAs. In the future, these solid-immersion metalenses arrays will be monolithically integrated with IR FPAs.

physics.optics↗

arXiv2017

Reactions, Diffusion and Volume Exclusion in a Heterogeneous System of Interacting Particles

Complex biological and physical transport processes are often described through systems of interacting particles. Excluded-volume effects on these transport processes are well studied, however the interplay between volume exclusion and reactions between heterogenous particles is less well known. In this paper we develop a novel framework for modeling reaction-diffusion processes which directly incorporates volume exclusion. From an off-lattice microscopic individual based model we use the Fokker--Planck equation and the method of matched asymptotic expansions to derive a low-dimensional macroscopic system of nonlinear partial differential equations describing the evolution of the particles. A biologically motivated, hybrid model of chemotaxis with volume exclusion is explored, where reactions occur at rates dependent upon the chemotactic environment. Further, we show that for reactions due to contact interactions the appropriate reaction term in the macroscopic model is of lower order in the asymptotic expansion than the nonlinear diffusion term. However, we find that the next reaction term in the expansion is needed to ensure good agreement with simulations of the microscopic model. Our macroscopic model allows for more direct parameterization to experimental data than the models available to date.

cond-mat.stat-mech↗

arXiv2016

Multi-level methods and approximating distribution functions

Biochemical reaction networks are often modelled using discrete-state, continuous-time Markov chains. System statistics of these Markov chains usually cannot be calculated analytically and therefore estimates must be generated via simulation techniques. There is a well documented class of simulation techniques known as exact stochastic simulation algorithms, an example of which is Gillespie's direct method. These algorithms often come with high computational costs, therefore approximate stochastic simulation algorithms such as the tau-leap method are used. However, in order to minimise the bias in the estimates generated using them, a relatively small value of tau is needed, rendering the computational costs comparable to Gillespie's direct method. The multi-level Monte Carlo method (Anderson and Higham, Multiscale Model. Simul. 10:146-179, 2012) provides a reduction in computational costs whilst minimising or even eliminating the bias in the estimates of system statistics. This is achieved by first crudely approximating required statistics with many sample paths of low accuracy. Then correction terms are added until a required level of accuracy is reached. Recent literature has primarily focussed on implementing the multi-level method efficiently to estimate a single system statistic. However, it is clearly also of interest to be able to approximate entire probability distributions of species counts. We present two novel methods that combine known techniques for distribution reconstruction with the multi-level method. We demonstrate the potential of our methods using a number of examples.

q-bio.QM↗

arXiv2015

New Routes to Phylogeography

Phylogeographic methods aim to infer migration trends and the history of sampled lineages from genetic data. Applications of phylogeography are broad, and in the context of pathogens include the reconstruction of transmission histories and the origin and emergence of outbreaks. Phylogeographic inference based on bottom-up population genetics models is computationally expensive, and as a result faster alternatives based on the evolution of discrete traits have become popular. In this paper, we show that inference of migration rates and root locations based on discrete trait models is extremely unreliable and sensitive to biased sampling. To address this problem, we introduce BASTA (BAyesian STructured coalescent Approximation), a new approach implemented in BEAST2 that combines the accuracy of methods based on the structured coalescent with the computational efficiency required to handle more than just few populations. We illustrate the potentially severe implications of poor model choice for phylogeographic analyses by investigating the zoonotic transmission of Ebola virus. Whereas the structured coalescent analysis correctly infers that successive human Ebola outbreaks have been seeded by a large unsampled non-human reservoir population, the discrete trait analysis implausibly concludes that undetected human-to-human transmission has allowed the virus to persist over the past four decades. As genomics takes on an increasingly prominent role informing the control and prevention of infectious diseases, it will be vital that phylogeographic inference provides robust insights into transmission history.

q-bio.PE↗

arXiv2003

The Mad Hatter´s Cocktail Party: A Social Mobile Audio Space Supporting Multiple Simultaneous Conversations

This paper presents a mobile audio space intended for use by gelled social groups. In face-to-face interactions in such social groups, conversational floors change frequently, e.g., two participants split off to form a new conversational floor, a participant moves from one conversational floor to another, etc. To date, audio spaces have provided little support for such dynamic regroupings of participants, either requiring that the participants explicitly specify with whom they wish to talk or simply presenting all participants as though they are in a single floor. By contrast, the audio space described here monitors participant behavior to identify conversational floors as they emerge. The system dynamically modifies the audio delivered to each participant to enhance the salience of the participants with whom they are currently conversing. We report a user study of the system, focusing on conversation analytic results.

cs.HC↗