A5009541437- Topic Modeling
- Radiomics and Machine Learning in Medical Imaging
- Natural Language Processing Techniques
- Generative Adversarial Networks and Image Synthesis
- Machine Learning in Healthcare
- Statistical Methods and Inference
- Robotics and Sensor-Based Localization
- Artificial Intelligence in Healthcare
- Robotic Path Planning Algorithms
- Remote Sensing and LiDAR Applications
- AI in cancer detection
- Social Robot Interaction and HRI
- Health, Environment, Cognitive Aging
- Robotics and Automated Systems
- Heart Rate Variability and Autonomic Control
- Face recognition and analysis
- Advanced Causal Inference Techniques
- Statistical Methods in Clinical Trials
- Prosthetics and Rehabilitation Robotics
- Image and Signal Denoising Methods
- Distributed Sensor Networks and Detection Algorithms
- Populism, Right-Wing Movements
- Multi-Agent Systems and Negotiation
- Soft Robotics and Applications
- COVID-19 diagnosis using AI
University of Oxford
2021-2024
The Alan Turing Institute
2024
Dyson (United Kingdom)
2019-2020
Imperial College London
2018-2020
Carnegie Mellon University
2019-2020
University College London
2019
University of Michigan
1992
BackgroundUpdatable estimates of COVID-19 onset, progression, and trajectories underpin pandemic mitigation efforts. To identify characterise disease trajectories, we aimed to define validate ten phenotypes from nationwide linked electronic health records (EHR) using an extensible framework.MethodsIn this cohort study, used eight National Health Service (NHS) datasets for people in England alive on Jan 23, 2020. Data testing, vaccination, primary secondary care records, death registrations...
State-space models (SSMs) and transformers dominate the language modeling landscape. However, they are constrained to a lower computational complexity than classical recurrent neural networks (RNNs), limiting their expressivity. In contrast, RNNs lack parallelization during training, raising fundamental questions about trade off between We propose implicit SSMs, which iterate transformation until convergence fixed point. Theoretically, we show that SSMs implement non-linear state-transitions...
U-Nets are a go-to, state-of-the-art neural architecture across numerous tasks for continuous signals on square such as images and Partial Differential Equations (PDE), however their design is understudied. In this paper, we provide framework designing analysing general U-Net architectures. We present theoretical results which characterise the role of encoder decoder in U-Net, high-resolution scaling limits conjugacy to ResNets via preconditioning. propose Multi-ResNets, with simplified,...
The proximity between newspapers and political parties is strongly subjective difficult to measure. Yet, tendencies of can have a significant impact on voters’ opinion‐forming ought be known by the public in transparent timely manner. This article introduces Sentiment Political Compass ( SPC ), data‐driven framework for analyzing bias toward parties. Using SPC, are embedded two‐dimensional space (left‐leaning vs. right‐leaning, libertarian autocratic). To assess informative value our...
Radiology reporting is a complex task that requires detailed image understanding, integration of multiple inputs, including comparison with prior imaging, and precise language generation. This makes it ideal for the development use generative multimodal models. Here, we extend report generation to include localisation individual findings on - call grounded Prior work indicates grounding important clarifying understanding interpreting AI-generated text. Therefore, stands improve utility...
Work in deep clustering focuses on finding a single partition of data. However, high-dimensional data, such as images, typically feature multiple interesting characteristics one could cluster over. For example, images objects against background be clustered over the shape object and separately by colour background. In this paper, we introduce Multi-Facet Clustering Variational Autoencoders (MFCVAE), novel class variational autoencoders with hierarchy latent variables, each...
We introduce Robot DE NIRO, an autonomous, collaborative, humanoid robot for mobile manipulation. built NIRO to perform a wide variety of manipulation behaviors, with focus on pick-and-place tasks. is designed be used in domestic environment, especially support caregivers working the elderly. Given this design focus, can interact naturally, reliably, and safely humans, autonomously navigate through environments command, intelligently retrieve or move target objects, avoid collisions...
We introduce Ivy, a templated Deep Learning (DL) framework which abstracts existing DL frameworks. Ivy unifies the core functions of these frameworks to exhibit consistent call signatures, syntax and input-output behaviour. New high-level framework-agnostic classes, are usable alongside framework-specific code, can then be implemented as compositions unified low-level functions. currently supports TensorFlow, PyTorch, MXNet, Jax NumPy. also release four pure-Ivy libraries for mechanics, 3D...
Traditional methods for matching in causal inference are impractical high-dimensional datasets. They suffer from the curse of dimensionality: exact and coarsened find exponentially fewer matches as input dimension grows, propensity score may match highly unrelated units together. To overcome this problem, we develop theoretical results which motivate use neural networks to obtain non-trivial, multivariate balancing scores a chosen level coarseness, contrast classical, scalar score. We...
Social assistance robots in health and elderly care have the potential to support ease human lives. Given macrosocial trends of aging long-lived populations, robotics-based research mainly focused on helping live independently. In this paper, we introduce Robot DE NIRO, a platform that aims supporter (the caregiver) also offers direct human-robot interaction for recipient. Augmented by several sensors, NIRO is capable complex manipulation tasks. It reliably interacts with humans can...
Likelihood-based deep generative models such as score-based diffusion and variational autoencoders are state-of-the-art machine learning approximating high-dimensional distributions of data images, text, or audio. One many downstream tasks they can be naturally applied to is out-of-distribution (OOD) detection. However, seminal work by Nalisnick et al. which we reproduce showed that consistently infer higher log-likelihoods for OOD than were trained on, marking an open problem. In this work,...
Clinical trials involve the collection of a wealth data, comprising multiple diverse measurements performed at baseline and follow-up visits over course trial. The most common primary analysis is restricted to single, potentially composite endpoint one time point. While such an analytical focus promotes simple replicable conclusions, it does not necessarily fully capture multi-faceted effects drug in complex disease setting. Therefore, complement existing approaches, we set out here design...
In-context learning (ICL) has emerged as a particularly remarkable characteristic of Large Language Models (LLM): given pretrained LLM and an observed dataset, LLMs can make predictions for new data points from the same distribution without fine-tuning. Numerous works have postulated ICL approximately Bayesian inference, rendering this natural hypothesis. In work, we analyse hypothesis angle through martingale property, fundamental requirement system exchangeable data. We show that property...
Identifying patient subgroups with different treatment responses is an important task to inform medical recommendations, guidelines, and the design of future clinical trials. Existing approaches for subgroup analysis primarily rely on Randomised Controlled Trials (RCTs), in which assignment randomised. RCTs' cohorts are often constrained by cost, rendering them not representative heterogeneity patients likely receive real-world practice. When applied observational studies, suffer from...
There is growing interest in applying AI to radiology report generation, particularly for chest X-rays (CXRs). This paper investigates whether incorporating pixel-level information through segmentation masks can improve fine-grained image interpretation of multimodal large language models (MLLMs) generation. We introduce MAIRA-Seg, a segmentation-aware MLLM framework designed utilize semantic alongside CXRs generating reports. train expert obtain mask pseudolabels radiology-specific...
U-Net architectures are ubiquitous in state-of-the-art deep learning, however their regularisation properties and relationship to wavelets understudied. In this paper, we formulate a multi-resolution framework which identifies U-Nets as finite-dimensional truncations of models on an infinite-dimensional function space. We provide theoretical results prove that average pooling corresponds projection within the space square-integrable functions show with implicitly learn Haar wavelet basis...
Generally capable Spatial AI systems must build persistent scene representations where geometric models are combined with meaningful semantic labels. The many approaches to labelling scenes can be divided into two clear groups: view-based which estimate labels from the input view-wise data and then incrementally fuse them model as it is built; map-based label generated model. However, there has so far been no attempt quantitatively compare labelling. Here, we present an experimental...
Monitoring physiological responses to hemodynamic stress can help in determining appropriate treatment and ensuring good patient outcomes. Physicians' intuition suggests that the human body has a number of response patterns hemorrhage which escalate as blood loss continues, however exact etiology phenotypes such are not well known or understood only at coarse level. Although previous research shown machine learning models perform detection survival prediction, it is unclear whether could...