University of Oxford

In early 2020, adult volunteers were invited to participate in a first-in-human trial of the COVID-19 vaccine, ChAdOx1 nCoV-19, in the United Kingdom (UK) at the height of the global pandemic when there was uncertainty regarding vaccine efficacy and side-effects. We conducted a retrospective survey of these uniquely situated individuals to gain insight into their views about the risks, motivations, and expectations of the trial and potential vaccine deployment. Our data from 349 respondents show that these volunteers were educated to a high-level with a clear understanding of the seriousness of the COVID-19 pandemic, as well as an appreciation of the role of science and research in developing a vaccine to address this global problem. Individuals were primarily motivated with altruistic intent and expressed a desire to contribute to the scientific effort. Respondents appreciated that their participation was associated with risk but appeared comfortable that this risk was low. Through our analysis, we highlight these individuals as a group with strong levels of trust in science and a sense of societal responsibility, and therefore are a potential valuable resource to improve confidence in novel vaccines. Vaccine trial participants could offer a credible collective voice to support positive messaging around vaccination.

Peterson et al. (2023) present a range of ethical issues that arise when considering the use of psychedelic substances within medicine. But psychedelics are, by their nature, boundary-dissolving, and we suggest that progress in the Ethics of Psychedelic Medicine is best made within a broader-ranging Psychedelic Bioethics, which encompasses not just medicine, but wider society, including the breadth of cultural containers and settings in which these compounds are used.

Psychedelics such as psilocybin reliably produce significantly altered states of consciousness with a variety of subjectively experienced effects. These include certain changes to perception, cognition, and affect, which we refer to here as the acute subjective effects of psychedelics. In recent years, psychedelics such as psilocybin have also shown considerable promise as therapeutic agents when combined with talk therapy: for example, in the treatment of major depression or substance use disorder. However, it is currently unclear whether the aforementioned acute subjective effects are necessary to bring about the observed therapeutic effects of psilocybin and other psychedelics. This uncertainty has sparked a lively – though still largely-hypothetical – debate on whether psychedelics without subjective effects (“nonsubjective psychedelics” or “non-hallucinogenic psychedelics”) could still have the same therapeutic impact, or whether the acute subjective effects are in fact necessary for this impact to be fully realized.

The concept of medical necessity is often used to explain or justify certain decisions—for example, which treatments should be allowed under certain conditions—as though it had an obvious, agreed-upon meaning as well as an inherent normative force. In introducing this special issue of Clinical Ethics on medical necessity, we argue that the term, as used in various discourses, generally lacks a definition that is clear, non-circular, conceptually plausible, and fit-for-purpose. We propose that future work on this concept should address three main questions: what medical necessity is (i.e., what makes something medically necessary, as opposed to something else); what the concept does (what ‘work’ is it doing when invoked in different settings); and what should follow, normatively, from the fact that something is indeed medically necessary (on some plausible conception)?

In this article, we explore the potential of enhancing academic prose and idea generation by fine-tuning a large language model (here, GPT-3) on one’s own previously published writings: AUTOGEN (‘AI Unique Tailored Output GENerator’). We develop, test, and describe three distinct AUTOGEN models trained on the prior scholarly output of three of the current authors (SBM, BDE, JS), with a fourth model trained on the combined works of all three. Our AUTOGEN models demonstrate greater variance in quality than the base GPT-3 model, with many outputs outperforming the base model in format, style, overall quality, and novel idea generation. As proof of principle, we present and discuss examples of AUTOGEN-written sections of existing and hypothetical research papers. We further discuss ethical opportunities, concerns, and open questions associated with personalized academic prose and idea generators. Ethical opportunities of personalized LLMs such as AUTOGEN include increased productivity, preservation of writing styles and cultural traditions, and aiding consensus building. However, ethical concerns arise due to the potential for personalized LLMs to reduce output diversity, violate privacy and intellectual property rights, and facilitate plagiarism or fraud. The use of co-authored or multiple-source trained models further complicates issues surrounding ownership and attribution. Open questions concern a potential credit-blame asymmetry for LLM outputs, the legitimacy of licensing agreements in authorship ascription, and the ethical implications of co-authorship attribution for data contributors. Ensuring the output is sufficiently distinct from the source material is crucial to maintaining ethical standards in academic writing. These opportunities, risks, and open issues highlight the intricate ethical landscape surrounding the use of personalized LLMs in academia. We also discuss open technical questions concerning the integration of AUTOGEN-style personalized LLMs with other LLMs, such as GPT-4, for iterative refinement and improvement of generated text. In conclusion, we argue that AUTOGEN-style personalized LLMs offer significant potential benefits in terms of both prose generation and, to a lesser extent, idea generation. If associated ethical issues are appropriately addressed, AUTOGEN alone or in combination with other LLMs can be seen as a potent form of academic enhancement. As a note to readers, this abstract was generated by AUTOGEN and edited for accuracy by the authors. The rest of the text was written manually.

In this letter, we present a non-mechanical 2D beam-steering system suitable for optical wireless communication. Steering is achieved using polarization gratings combined with nematic liquid crystal cells operating as voltage-controlled polarization shifters. These steering elements are combined with a holographic diffuser matched to the discrete steering angles of the polarization grating, enabling continuous angular coverage. Beam-steering is also used at the receiver, allowing a large collection area receiver with a relatively narrow field of view to be used. The approach presented here could, in principle, be applied to a broad range of wavelengths (including visible light and near-infrared wavelengths). Furthermore, the technique does not inherently limit the transmission data rate. Besides, it improves the link margin and offers the potential for a bidirectional steered link using the same beam-steering elements. Details of the approach are set out in the paper, followed by experimental results from a 50 Mbit/s optical link operating over one meter. Future directions are then discussed.

Previous work on fluorescent antennas in visible light communications (VLC) has primarily focused on downlink receivers, which have to be compact. In contrast, uplink receivers can potentially occupy much larger areas. A convenient and inexpensive approach to increasing the antenna’s area is to use a large array of fluorescent optical fibers. The challenge is then to couple this array to a much smaller photodiode. To address this challenge an antenna that consists of two stages is demonstrated. The first of these stages is a large array of fluorescent fibers that can absorb light from the uplink transmitter. The fluorescence from this array is then coupled into the second stage, which is a single fluorescent fiber that can absorb the fluorescence from the first stage. The diameter of this fiber means that it can be effectively coupled to a photodiode. Results are presented which demonstrate this concept by measuring the system’s signal gain, field of view (FoV), bandwidth, and data transmission performance.

It is often suggested in the popular press that food chains deliberately introduce enticing product aromas into (and in the immediate vicinity of) their premises in order to attract customers. However, despite the widespread use of odours in the field of sensory marketing, laboratory research suggests that their effectiveness in modulating people's food behaviours depends on a range of contextual factors. Given the evidence that has been published to date, only under a subset of conditions is there likely a measurable effect of the presence of ambient odours on people's food attitudes and choices. This narrative historical review summarizes the various ways in which food odours appear to bias people's food preferences (appetite) and food choices (food consumption and purchase). Emphasis is placed on those experimental studies that have been designed to investigate how the characteristics of the olfactory stimuli (e.g., the congruency between the olfactory cues and the foods, intensity and duration of exposure to odours, and taste properties of odours) modulate the effects of olfactory cues on food behaviour. The review also explores the moderating roles of individual differences, such as dietary restraint, Body Mass Index (BMI), genetic and cultural differences in odour sensitivity and perception. Ultimately, following a review of empirical studies on food-related olfaction, current approaches in scent marketing are discussed and a research agenda is proposed to help encourage further studies on the effective application of scents in promoting healthy foods.

Post-translational modifications (PTMs) of proteins are central to epigenetic regulation and cellular signalling, playing an important role in the pathogenesis and progression of numerous diseases. Growing evidence indicates that protein arginine citrullination, catalysed by peptidylarginine deiminases (PADs), is involved in many aspects of molecular and cell biology and is emerging as a potential druggable target in multiple diseases including cancer. However, we are only just beginning to understand the molecular activities of PADs, and their underlying mechanistic details in vivo under both physiological and pathological conditions. Many questions still remain regarding the dynamic cellular functions of citrullination and its interplay with other types of PTMs. This review, therefore, discusses the known functions of PADs with a focus on cancer biology, highlighting the cross-talk between citrullination and other types of PTMs, and how this interplay regulates downstream biological events. This article is part of the Theo Murphy meeting issue ‘The virtues and vices of protein citrullination’.

Alström syndrome (AS) is an ultra-rare disorder characterised by early-onset multi-organ dysfunction, such as insulin resistance, obesity, dyslipidaemia, and renal and cardiovascular disease. The objective is to explore whether AS is a disease of accelerated ageing and whether changes over time on echocardiography could reflect accelerated cardiac ageing. Cross-sectional measurement of Phenoage and retrospective analysis of serial echocardiography were performed between March 2012 and November 2022. The setting is a single national tertiary service jointly run by health service and patient charity. Forty-five adult patients aged over 16 years were included, 64% were male and 67% of White ethnicity. The median Phenoage was 48 years (interquartile range [IQR]: 35–72) in the 34 patients for whom this was calculable, which was significantly higher than the median chronological age of 29 years (IQR: 22–39, p<0.001). Phenoage was higher than chronological age in 85% (N=29) of patients, with a median difference of +18 years (IQR: +4, +34). On echocardiography, significant decreases were observed over time in left ventricular (LV) size at end-diastole (average of 0.046 cm per year, p<0.001) and end-systole (1.1% per year, p=0.025), with significant increase in posterior wall thickness at end-diastole (0.009 cm per year, p=0.008). LV systolic function measured by global longitudinal strain reduced (0.34 percentage points per year, p=0.020) and E/e’lat increased (2.5% per year, p=0.019). Most AS patients display a higher Phenoage compared to chronological age. Cardiac changes in AS patients were also reflective of accelerated ageing, with a reduction in LV size and increased wall thickening. AS may be a paradigm disease for premature ageing.

This work concerns continuous-time, continuous-space stochastic dynamical systems described by stochastic differential equations (SDE). It presents a new approach to compute probabilistic safety regions, namely sets of initial conditions of the SDE associated to trajectories that are safe with a probability larger than a given threshold. The approach introduces a functional that is minimised at the border of the probabilistic safety region, then solves an optimisation problem using techniques from Malliavin Calculus, which computes such region. Unlike existing results in the literature, the new approach allows one to compute probabilistic safety regions without gridding the state space of the SDE.

We propose and assess a new experimental technique to measure the fracture toughness of engineering materials and its sensitivity to strain rate. The proposed method is based on a ring expansion technique and it overcomes the limitations of current dynamic fracture tests, as it is not affected by transient stress wave propagation during loading and it results in spatially uniform remote stress and strain fields prior to fracture; the method is also suitable to achieve remote strain rates well in excess of 1000 s−1. We demonstrate the technique by measuring the plane-stress Mode I fracture toughness of PMMA specimens at remote strain rates ranging from 10−3 s−1 to 102 s−1. The experiments show an increase of the toughness of the material with increasing strain rate.

Holistic 3D human-scene reconstruction is a crucial and emerging research area in robot perception. A key challenge in holistic 3D human-scene reconstruction is to generate a physically plausible 3D scene from a single monocular RGB image. The existing research mainly proposes optimization-based approaches for reconstructing the scene from a sequence of RGB frames with explicitly defined physical laws and constraints between different scene elements (humans and objects). However, it is hard to explicitly define and model every physical law in every scenario. This letter proposes using an implicit feature representation of the scene elements to distinguish a physically plausible alignment of humans and objects from an implausible one. We propose using a graph-based holistic representation with an encoded physical representation of the scene to analyze the human-object and object-object interactions within the scene. Using this graphical representation, we adversarially train our model to learn the feasible alignments of the scene elements from the training data itself without explicitly defining the laws and constraints between them. Unlike the existing inference-time optimization-based approaches, we use this adversarially trained model to produce a per-frame 3D reconstruction of the scene that abides by the physical laws and constraints. Our learning-based method achieves comparable 3D reconstruction quality to existing optimization-based holistic human-scene reconstruction methods and does not need inference time optimization. This makes it better suited compared to existing methods, for potential use in robotic applications, such as robot navigation, $\underline{etc}$ .

In this work, we present a learning method for both lateral and longitudinal motion control of an ego-vehicle for the task of vehicle pursuit. The car being controlled does not have a pre-defined route, rather it reactively adapts to follow a target vehicle while maintaining a safety distance. To train our model, we do not rely on steering labels recorded from an expert driver, but effectively leverage a classical controller as an offline label generation tool. In addition, we account for the errors in the predicted control values, which can lead to a loss of tracking and catastrophic crashes of the controlled vehicle. To this end, we propose an effective data augmentation approach, which allows to train a network that is capable of handling different views of the target vehicle. During the pursuit, the target vehicle is firstly localized using a Convolutional Neural Network. The network takes a single RGB image along with cars' velocities and estimates target vehicle's pose with respect to the ego-vehicle. This information is then fed to a Multi-Layer Perceptron, which regresses the control commands for the ego-vehicle, namely throttle and steering angle. We extensively validate our approach using the CARLA simulator on a wide range of terrains. Our method demonstrates real-time performance, robustness to different scenarios including unseen trajectories and high route completion. Project page containing code and multimedia can be publicly accessed here: https://changyaozhou.github.io/Autonomous-Vehicle-Pursuit/ .

Event-based cameras asynchronously capture individual visual changes in a scene. This makes them more robust than traditional frame-based cameras to highly dynamic motions and poor illumination. It also means that every measurement in a scene can occur at a unique time. Handling these different measurement times is a major challenge of using event-based cameras. It is often addressed in visual odometry (VO) pipelines by approximating temporally close measurements as occurring at one common time. This grouping simplifies the estimation problem but, absent additional sensors, sacrifices the inherent temporal resolution of event-based cameras. This paper instead presents a complete stereo VO pipeline that estimates directly with individual event-measurement times without requiring any grouping or approximation in the estimation state. It uses continuous-time trajectory estimation to maintain the temporal fidelity and asynchronous nature of event-based cameras through Gaussian process regression with a physically motivated prior. Its performance is evaluated on the MVSEC dataset, where it achieves $7.9\cdot 10^{-3}$ and $5.9\cdot 10^{-3}$ RMS relative error on two independent sequences, outperforming the existing publicly available event-based stereo VO pipeline by two and four times, respectively.

Existing methods for interpretability of model predictions are largely based on technical insights and are not linked to clinical context. We use the question of predicting response to radiotherapy in colorectal cancer patients as an exemplar for developing prediction models that do provide such contextual information and therefore can effectively support clinical decision making. There is a growing body of evidence that about 30% of colorectal cancer patients do not respond to radiotherapy and will need alternative treatment. The consensus molecular subtypes for colorectal cancer (CMS) provide one such approach to categorising patients based on their disease biology. Here we select the CMS4 subtype as a proxy for stromal infiltration. By jointly predicting a patient’s response to radiotherapy, the presence of CMS4, and the epithelial tissue map from morphological features extracted from standard H &E slides we provide a comprehensive clinically relevant assessment of a biopsy. A graph neural network is trained to achieve this joint prediction task, which subsequently provides novel interpretability maps to aid clinicians in their cancer treatment decision making process. Our model is trained and validated on two private rectal cancer datasets.

Automated segmentation of anatomical structures in fetal ultrasound video is challenging due to the highly diverse appearance of anatomies and image quality. In this paper, we propose an ultrasound video anatomy segmentation approach to iteratively memorise and segment incoming video frames, which is suitable for online segmentation. This is achieved by a spatio-temporal model that utilizes an adaptive memory bank to store the segmentation history of preceding frames to assist the current frame segmentation. The memory is updated adaptively using a skip gate mechanism based on segmentation confidence, preserving only high-confidence predictions for future use. We evaluate our approach and related state-of-the-art methods on a clinical dataset. The experimental results demonstrate that our method achieves superior performance with an F1 score of 84.83%. Visually, the use of adaptive temporal memory also aids in reducing error accumulation during video segmentation.

What is normativity in South Asian public culture? How does it shape subjectivity and how can it be mapped and contested? This chapter develops the theoretical framework, methodology, and the queer archive to critically engage with norms, their effect, and internalisation. It locates controversies in South Asian public culture as a means through which normativity can be pried open, and argues for the importance of a queer methodology and queer archive that traces the slippage through which normativity functions. It develops queering, in the verb form, as a means to navigate multiple contradictions that relegate queerness as Western and reify postcolonial subjectivities within nationalist discourses of tradition and culture. It queers normativity by studying its functioning, production, and disappearance in South Asian public culture.

The most challenging, yet practical, setting of semi-supervised federated learning (SSFL) is where a few clients have fully labeled data whereas the other clients have fully unlabeled data. This is particularly common in healthcare settings where collaborating partners (typically hospitals) may have images but not annotations. The bottleneck in this setting is the joint training of labeled and unlabeled clients as the objective function for each client varies based on the availability of labels. This paper investigates an alternative way for effective training with labeled and unlabeled clients in a federated setting. We propose a novel learning scheme specifically designed for SSFL which we call Isolated Federated Learning (IsoFed) that circumvents the problem by avoiding simple averaging of supervised and semi-supervised models together. In particular, our training approach consists of two parts - (a) isolated aggregation of labeled and unlabeled client models, and (b) local self-supervised pretraining of isolated global models in all clients. We evaluate our model performance on medical image datasets of four different modalities publicly available within the biomedical image classification benchmark MedMNIST. We further vary the proportion of labeled clients and the degree of heterogeneity to demonstrate the effectiveness of the proposed method under varied experimental settings.

In this chapter I examine the home as a fundamental and crucial space for the production and operation of normativity. I read contra-norm acts by women that run into nationalist discourses that circumscribe the woman, nation, and mother as home. I particularly examine the tripartite construction nation-goddess-mother as fundamental to the discourse of home in anti-colonial nationalisms within the sub-continent and its effects thereafter. Here I rely on the discursive productions prompted by the textual interlocutions between the book Mother India (1927) by American journalist Katherine Mayo, written prior to the partition of the sub-continent, and the 1957 Indian nationalist response in the film Mother India by Mehboob Khan. I examine the plethora of cultural production that is marked by the two ossified poles of colonial and nationalist discourses through which a normative gender, and womanhood, in South Asian public culture is coded.