arXiv subjects

Graham Murray

Publications and source records attributed to Graham Murray.

12 recordsLinked to original sources

arXiv2024

The Explanation Necessity for Healthcare AI

Explainability is a critical factor in enhancing the trustworthiness and acceptance of artificial intelligence (AI) in healthcare, where decisions directly impact patient outcomes. Despite advancements in AI interpretability, clear guidelines on when and to what extent explanations are required in medical applications remain lacking. We propose a novel categorization system comprising four classes of explanation necessity (self-explainable, semi-explainable, non-explainable, and new-patterns discovery), guiding the required level of explanation; whether local (patient or sample level), global (cohort or dataset level), or both. To support this system, we introduce a mathematical formulation that incorporates three key factors: (i) robustness of the evaluation protocol, (ii) variability of expert observations, and (iii) representation dimensionality of the application. This framework provides a practical tool for researchers to determine the appropriate depth of explainability needed, addressing the critical question: When does an AI medical application need to be explained, and at what level of detail?

cs.AI↗

arXiv2024

Contrastive-Adversarial and Diffusion: Exploring pre-training and fine-tuning strategies for sulcal identification

In the last decade, computer vision has witnessed the establishment of various training and learning approaches. Techniques like adversarial learning, contrastive learning, diffusion denoising learning, and ordinary reconstruction learning have become standard, representing state-of-the-art methods extensively employed for fully training or pre-training networks across various vision tasks. The exploration of fine-tuning approaches has emerged as a current focal point, addressing the need for efficient model tuning with reduced GPU memory usage and time costs while enhancing overall performance, as exemplified by methodologies like low-rank adaptation (LoRA). Key questions arise: which pre-training technique yields optimal results - adversarial, contrastive, reconstruction, or diffusion denoising? How does the performance of these approaches vary as the complexity of fine-tuning is adjusted? This study aims to elucidate the advantages of pre-training techniques and fine-tuning strategies to enhance the learning process of neural networks in independent identical distribution (IID) cohorts. We underscore the significance of fine-tuning by examining various cases, including full tuning, decoder tuning, top-level tuning, and fine-tuning of linear parameters using LoRA. Systematic summaries of model performance and efficiency are presented, leveraging metrics such as accuracy, time cost, and memory efficiency. To empirically demonstrate our findings, we focus on a multi-task segmentation-classification challenge involving the paracingulate sulcus (PCS) using different 3D Convolutional Neural Network (CNN) architectures by using the TOP-OSLO cohort comprising 596 subjects.

eess.IV↗

arXiv2024

Solving the enigma: Enhancing faithfulness and comprehensibility in explanations of deep networks

The accelerated progress of artificial intelligence (AI) has popularized deep learning models across various domains, yet their inherent opacity poses challenges, particularly in critical fields like healthcare, medicine, and the geosciences. Explainable AI (XAI) has emerged to shed light on these 'black box' models, aiding in deciphering their decision-making processes. However, different XAI methods often produce significantly different explanations, leading to high inter-method variability that increases uncertainty and undermines trust in deep networks' predictions. In this study, we address this challenge by introducing a novel framework designed to enhance the explainability of deep networks through a dual focus on maximizing both accuracy and comprehensibility in the explanations. Our framework integrates outputs from multiple established XAI methods and leverages a non-linear neural network model, termed the 'explanation optimizer,' to construct a unified, optimal explanation. The optimizer evaluates explanations using two key metrics: faithfulness (accuracy in reflecting the network's decisions) and complexity (comprehensibility). By balancing these, it provides accurate and accessible explanations, addressing a key XAI limitation. Experiments on multi-class and binary classification in 2D object and 3D neuroscience imaging confirm its efficacy. Our optimizer achieved faithfulness scores 155% and 63% higher than the best XAI methods in 3D and 2D tasks, respectively, while also reducing complexity for better understanding. These results demonstrate that optimal explanations based on specific quality criteria are achievable, offering a solution to the issue of inter-method variability in the current XAI literature and supporting more trustworthy deep network predictions

cs.CV↗

arXiv2023

An explainable three dimension framework to uncover learning patterns: A unified look in variable sulci recognition

The significant features identified in a representative subset of the dataset during the learning process of an artificial intelligence model are referred to as a 'global' explanation. 3D global explanations are crucial in neuroimaging, where a complex representational space demands more than basic 2D interpretations. However, current studies in the literature often lack the accuracy, comprehensibility, and 3D global explanations needed in neuroimaging and beyond. To address this gap, we developed an explainable artificial intelligence (XAI) 3D-Framework capable of providing accurate, low-complexity global explanations. We evaluated the framework using various 3D deep learning models trained on a well-annotated cohort of 596 structural MRIs. The binary classification task focused on detecting the presence or absence of the paracingulate sulcus, a highly variable brain structure associated with psychosis. Our framework integrates statistical features (Shape) and XAI methods (GradCam and SHAP) with dimensionality reduction, ensuring that explanations reflect both model learning and cohort-specific variability. By combining Shape, GradCam, and SHAP, our framework reduces inter-method variability, enhancing the faithfulness and reliability of global explanations. These robust explanations facilitated the identification of critical sub-regions, including the posterior temporal and internal parietal regions, as well as the cingulate region and thalamus, suggesting potential genetic or developmental influences. Our XAI 3D-Framework leverages global explanations to uncover the broader developmental context of specific cortical features. This approach advances the fields of deep learning and neuroscience by offering insights into normative brain development and atypical trajectories linked to mental illness, paving the way for more reliable and interpretable AI applications in neuroimaging.

cs.CV↗

arXiv2018

Design and production of the DESI fibre cables

The Dark Energy Spectroscopic Instrument (DESI) is under construction to measure the expansion history of the Universe using the Baryonic Acoustic Oscillation technique. The spectra of 35 million galaxies and quasars over 14000 sq deg will be measured during the life of the experiment. A new prime focus corrector for the KPNO Mayall telescope will deliver light to 5000 fibre optic positioners. The fibres in turn feed 10 broad-band spectrographs. We will describe the design and production progress on the fibre cables, strain relief system and preparation of the slit end. In contrast to former projects, the larger scale of production required for DESI requires teaming up with industry to find a solution to reduce the time scale of production as well as to minimise the stress on the optical fibres.

astro-ph.IM↗

arXiv2016

The Current Status of Prime Focus Instrument of Subaru Prime Focus Spectrograph

The Prime Focus Spectrograph (PFS) is a new optical/near-infrared multi-fiber spectrograph design for the prime focus of the 8.2m Subaru telescope. PFS will cover 1.3 degree diameter field with 2394 fibers to complement the imaging capability of Hyper SuprimeCam (HSC). The prime focus unit of PFS called Prime Focus Instrument (PFI) provides the interface with the top structure of Subaru telescope and also accommodates the optical bench in which Cobra fiber positioners are located. In addition, the acquisition and guiding cameras (AGCs), the optical fiber positioner system, the cable wrapper, the fiducial fibers, illuminator, and viewer, the field element, and the telemetry system are located inside the PFI. The mechanical structure of the PFI was designed with special care such that its deflections sufficiently match those of the HSC's Wide Field Corrector (WFC) so the fibers will stay on targets over the course of the observations within the required accuracy. In this report, the latest status of PFI development will be given including the performance of PFI components, the setup and performance of the integration and testing equipment.

astro-ph.IM↗

arXiv2014

Studying focal ratio degradation of optical fibers for Subaru Prime Focus Spectrograph

Focal Ration Degradation (FRD) is a change in light angular distribution caused by fiber optics. FRD is important to fiber-fed, spectroscopic astronomical systems because it can cause loss of signal, degradation in spectral resolution, or increased complexity in spectrograph design. Laborat\'orio Nacional de Astrof\'isica (LNA) has developed a system that can accurately and precisely measures FRD, using an absolute method that can also measure fiber throughput. This paper describes the metrology system and shows measurements of Polymicro fiber FBP129168190, FBP127165190 and Fujikura fiber 128170190. Although the FRD of the two fibers are low and similar to one another, it is very important to know the exact characteristics of these fibers since both will be used in the construction of FOCCoS (Fiber Optical Cable and Connectors System) for PFS (Prime Focus Spectrograph) to be installed at the Subaru telescope.

astro-ph.IM↗

arXiv2014

Fiber Optical Cable and Connector System (FOCCoS) for PFS/Subaru

FOCCoS, Fiber Optical Cable and Connector System, has the main function of capturing the direct light from the focal plane of Subaru Telescope using optical fibers, each one with a microlens in its tip, and conducting this light through a route containing connectors to a set of four spectrographs. The optical fiber cable is divided in 3 different segments called Cable A, Cable B and Cable C. Multi-fibers connectors assure precise connection among all optical fibers of the segments, providing flexibility for instrument changes. To assure strong and accurate connection, these sets are arranged inside two types of assemblies: the Tower Connector, for connection between Cable C and Cable B; and the Gang Connector, for connection between Cable B and Cable A. Throughput tests were made to evaluate the efficiency of the connections. A lifetime test connection is in progress. Cable C is installed inside the PFI, Prime Focus Instrument, where each fiber tip with a microlens is bonded to the end of the shaft of a 2-stage piezo-electric rotatory motor positioner; this assembly allows each fiber to be placed anywhere within its patrol region, which is 9.5mm diameter.. Each positioner uses a fiber arm to support the ferrule, the microlens, and the optical fiber. 2400 of these assemblies are arranged on a motor bench plate in a hexagonal-closed-packed disposition.

astro-ph.IM↗

arXiv2013

End effects in optical fibres

The performance of highly-multiplexed spectrographs is limited by focal degradation (FRD) in the optical fibres. It has already been shown that this is caused mainly by processes concentrated around the mounting points at the ends of the fibres. We use the thickness of rings produced in the farfield when a fibre is illuminated by a collimated beam, to estimate the size of the region where the FRD is generated. This requires the development of a new model, using features of existing ray-tracing and wave-based models, which fits existing data very well. The results suggest that the amount of FRD is primarily determined by the length of fibre bonded into the supporting ferrule. We point out the implications for the production of future fibre systems.

astro-ph.IM↗

arXiv2012

Simulation of complex phenomena in optical fibres

Optical fibres are essential for many types of highly-multiplexed and precision spectroscopy. The success of the new generation of multifibre instruments under construction to investigate fundamental problems in cosmology, such as the nature of dark energy, requires accurate modellisation of the fibre system to achieve their signal-to-noise goals. Despite their simple construction, fibres exhibit unexpected behaviour including non-conservation of Etendue (Focal Ratio Degradation; FRD) and modal noise. Furthermore, new fibre geometries (non-circular or tapered) have become available to improve the scrambling properties that, together with modal noise, limit the achievable SNR in precision spectroscopy. These issues have often been addressed by extensive tests on candidate fibres and their terminations but these are difficult and time-consuming. Modelling by ray-tracing and wave analysis is possible with commercial software packages but these do not address the more complex features, in particular FRD. We use a phase-tracking ray-tracing method to provide a practical description of FRD derived from our previous experimental work on circular fibres and apply it to non-standard fibres. This allows the relationship between scrambling and FRD to be quantified for the first time. We find that scrambling primarily affects the shape of the near-field pattern but has negligible effect on the barycentre. FRD helps to homogenise the nearfield pattern but does not make it completely uniform. Fibres with polygonal cross-section improve scrambling without amplifying the FRD. Elliptical fibres, in conjunction with tapering, may offer an efficient means of image slicing to improve the product of resolving power and throughput but the result is sensitive to the details of illumination.

astro-ph.IM↗

arXiv2010

The Fibre Multi-Object Spectrograph (FMOS) for Subaru Telescope

Fibre Multi-Object Spectrograph (FMOS) is the first near-infrared instrument with a wide field of view capable of acquiring spectra simultaneously from up to 400 objects. It has been developed as a common-use instrument for the F/2 prime-focus of the Subaru Telescope. The field coverage of 30' diameter is achieved using a new 3-element corrector optimized in the near-infrared (0.9-1.8um) wavelength range. Due to limited space at the prime-focus, we have had to develop a novel fibre positioner called "Echidna" together with two OH-airglow suppressed spectrographs. FMOS consists of three subsystems: the prime focus unit for IR, the fibre positioning system/connector units, and the two spectrographs. After full systems integration, FMOS was installed on the telescope in late 2007. Many aspects of performance were checked through various test and engineering observations. In this paper, we present the optical and mechanical components of FMOS and show the results of our on-sky engineering observations to date.

astro-ph.IM↗

arXiv2002

The GMOS integral field unit: first integral field spectroscopy with an 8m telescope

The Gemini Multiobject Spectrograph (GMOS) installed on the Gemini-North telescope has a facility for integral field spectroscopy over the wavelength range 0.4-1.0um. GMOS is converted to this mode by the remote insertion of an integral field unit (IFU) into the beam in place of the masks used for multiobject spectroscopy. With the IFU deployed, integral field spectroscopy is available over a fully-filled contiguous field of 5 x 7 arcsec with a sampling of 0.2 arcsec. A separate field of half the area, but otherwise identical, is also provided to aid background subtraction. The IFU contains 1500 lenslet-coupled fibres and is the largest-format fibre-based IFU yet tested on the sky and the first facility of any type for integral field spectroscopy employed on an 8/10m telescope. We describe the IFU and present results from commissioning.

astro-ph↗