A5026151059- Generative Adversarial Networks and Image Synthesis
- Adversarial Robustness in Machine Learning
- Topic Modeling
- Gaussian Processes and Bayesian Inference
- Advanced Graph Neural Networks
- Anomaly Detection Techniques and Applications
- Neural Networks and Applications
- Model Reduction and Neural Networks
- Natural Language Processing Techniques
- Domain Adaptation and Few-Shot Learning
- Stochastic Gradient Optimization Techniques
- Digital Media Forensic Detection
- Music and Audio Processing
- Explainable Artificial Intelligence (XAI)
- Data Mining Algorithms and Applications
- Health Systems, Economic Evaluations, Quality of Life
- Markov Chains and Monte Carlo Methods
- Advanced Memory and Neural Computing
- Data Quality and Management
- Speech and Audio Processing
- Neural dynamics and brain function
- Bayesian Modeling and Causal Inference
- Machine Learning and Data Classification
- Face recognition and analysis
- Game Theory and Voting Systems
Ruhr University Bochum
2013-2024
ETH Zurich
2024
Universität Hamburg
2024
University of Bonn
1996-2021
Office Of Health Economics
2019
Université de Montréal
2014-2016
University of Copenhagen
2013-2015
National Institute for Health and Care Excellence
2008-2014
University of Manchester
2013
St George's, University of London
1999-2011
We examine the role of memorization in deep learning, drawing connections to capacity, generalization, and adversarial robustness. While networks are capable memorizing noise data, our results suggest that they tend prioritize learning simple patterns first. In experiments, we expose qualitative differences gradient-based optimization neural (DNNs) on vs. real data. also demonstrate for appropriately tuned explicit regularization (e.g., dropout) can degrade DNN training performance datasets...
Question Answering (QA) systems over Knowledge Graphs (KG) automatically answer natural language questions using facts contained in a knowledge graph. Simple questions, which can be answered by the extraction of single fact, constitute large part asked on web but still pose challenges to QA systems, especially when against resource. Existing usually rely various components each specialised solving different sub-tasks problem (such as segmentation, entity recognition, disambiguation, and...
We investigate the dynamical and convergent properties of stochastic gradient descent (SGD) applied to Deep Neural Networks (DNNs). Characterizing relation between learning rate, batch size final minima, such as width or generalization, remains an open question. In order tackle this problem we previously proposed approximation SGD by a differential equation (SDE). theoretically argue that three factors - covariance influence minima found SGD. particular find ratio rate is key determinant...
Deep neural networks can generate images that are astonishingly realistic, so much it is often hard for humans to distinguish them from actual photos. These achievements have been largely made possible by Generative Adversarial Networks (GANs). While deep fake thoroughly investigated in the image domain - a classical approach area of forensics an analysis frequency has missing far. In this paper, we address shortcoming and our results reveal space, GAN-generated exhibit severe artifacts be...
Abstract The transformation of the chemical industry to renewable energy and feedstock supply requires new paradigms for design flexible plants, (bio‐)catalysts, functional materials. Recent breakthroughs in machine learning (ML) provide unique opportunities, but only joint interdisciplinary research between ML engineering (CE) communities will unfold full potential. We identify six challenges that open methods CE formulate types problems ML: (1) optimal decision making, (2) introducing...
The need to make best use of limited resources in the English National Health Service is now greater than ever. This paper contributes this endeavour by synthesizing data from cost-effectiveness evidence produced support development public health guidance at Institute and Clinical Excellence (NICE). No comprehensive list estimates for interventions has previously been published England. Cost-effectiveness using cost were collected analysed 21 (of 26) economic analyses underpinning NICE...
Since their invention, generative adversarial networks (GANs) have become a popular approach for learning to model distribution of real (unlabeled) data. Convergence problems during training are overcome by Wasserstein GANs which minimize the distance between and empirical in terms different metric, but thereby introduce Lipschitz constraint into optimization problem. A simple way enforce on class functions, can be modeled neural network, is weight clipping. It was proposed that improved...
We show that Langevin Markov chain Monte Carlo inference in an energy-based model with latent variables has the property early steps of inference, starting from a stationary point, correspond to propagating error gradients into internal layers, similar backpropagation. The backpropagated is respect output units have received outside driving force pushing them away point. Backpropagated temporal derivatives activation hidden units. These lead weight update proportional product presynaptic...
The heterogeneity in recently published knowledge graph embedding models' implementations, training, and evaluation has made fair thorough comparisons difficult. To assess the reproducibility of previously results, we re-implemented evaluated 21 models PyKEEN software package. In this paper, outline which results could be reproduced with their reported hyper-parameters, only alternate not at all, as well provide insight to why might case. We then performed a large-scale benchmarking on four...
Almost three million people in the United Kingdom have diabetes and 850 000 may be undiagnosed. It has been estimated that five will by 2025. About 90% of them type 2 diabetes.1 15% (one seven) adults impaired glucose regulation,2 an 5-12% these develop each year.1 People with regulation are 5-15 times more likely to than those normal values.3 Successful prevention requires population based action for whole community,4 together interventions targeted at greatest risk. This article summarises...
This paper outlines the National Institute for Health and Clinical Excellence's (NICE) emerging conceptual framework public health. is based on experience of first 3 years producing health guidance at NICE (2005-2008). The has been used to shape revisions NICE's process methods manuals use post 2009, will inform which produce from April 2009. precept that both individual population patterns disease have causal mechanisms. These are analytically separate. Explanations diseases involve...
Question answering has emerged as an intuitive way of querying structured data sources, and attracted significant advancements over the years. In this article, we provide overview these recent advancements, focusing on neural network based question systems knowledge graphs. We introduce readers to challenges in tasks, current paradigms approaches, discuss notable outline emerging trends field. Through aim newcomers field with a suitable entry point, ease their process making informed...