A5053913176- Cutaneous Melanoma Detection and Management
- Reinforcement Learning in Robotics
- Robot Manipulation and Learning
- AI in cancer detection
- Data Stream Mining Techniques
- Advanced Neural Network Applications
- Machine Learning and Data Classification
- Adrenal and Paraganglionic Tumors
- Soft Robotics and Applications
- Advanced Memory and Neural Computing
- Statistical Methods and Inference
- Cell Image Analysis Techniques
- Multimodal Machine Learning Applications
- Congenital Diaphragmatic Hernia Studies
- Intestinal Malrotation and Obstruction Disorders
- Neuroblastoma Research and Treatments
- Modular Robots and Swarm Intelligence
- Adrenal Hormones and Disorders
- Genetic factors in colorectal cancer
- Adversarial Robustness in Machine Learning
- Nonmelanoma Skin Cancer Studies
- Hernia repair and management
- Cell Adhesion Molecules Research
- AI-based Problem Solving and Planning
- Robotic Mechanisms and Dynamics
Ghent University
2017-2023
Abstract Modern deep learning models achieve state-of-the-art results for many tasks in computer vision, such as image classification and segmentation. However, its adoption into high-risk applications, e.g. automated medical diagnosis systems, happens at a slow pace. One of the main reasons this is that regular neural networks do not capture uncertainty. To assess uncertainty classification, several techniques have been proposed casting network approaches Bayesian setting. Amongst these...
Abstract In dermatology, deep learning may be applied for skin lesion classification. However, a given input image, neural network only outputs label, obtained using the class probabilities, which do not model uncertainty. Our group developed novel method to quantify uncertainty in stochastic networks. this study, we aimed train such classification and evaluate its diagnostic performance uncertainty, compare results assessments by of dermatologists. By passing duplicates an image through...
Learning-based approaches for robotic grasping using visual sensors typically require collecting a large size dataset, either manually labeled or by many trial and errors of manipulator in the real simulated world. We propose simpler learning-from-demonstration approach that is able to detect object grasp from merely single demonstration convolutional neural network we call GraspNet. In order increase robustness decrease training time even further, leverage data previous demonstrations...
Deep neural networks have achieved state-of-the-art performance in image classification. Due to this success, deep learning is now also being applied other data modalities such as multispectral images, lidar and radar data. However, successfully training a network requires large reddataset. Therefore, transitioning new sensor modality (e.g., from regular camera images images) might result drop performance, due the limited availability of modality. This hinder adoption rate time market for...
Upon the advent of Industry 4.0, collaborative robotics and intelligent automation gain more traction for enterprises to improve their production processes. In order adapt this trend, new programming, learning techniques are investigated. Program-bydemonstration is one that aim reduce burden manually programming robots. However, often limited teaching grasp at a certain position, rather than grasping object. paper, we propose method learns an arbitrary object from visual input. While other...
Because of their state-of-the-art performance in computer vision, CNNs are becoming increasingly popular a variety fields, including medicine. However, as neural networks black box function approximators, it is difficult, if not impossible, for medical expert to reason about output. This could potentially result the distrusting network when he or she does agree with its In such case, explaining why CNN makes certain decision becomes valuable information. this paper, we try open by inspecting...
Reinforcement learning is a proven technique for an agent to learn task. However, when task using reinforcement learning, the cannot distinguish characteristics of environment from those This makes it harder transfer skills between tasks in same environment. Furthermore, this does not reduce risk training new In paper, we introduce approach decouple task-specific ones, allowing develop sense survival. We evaluate our where must sequence collection tasks, and show that decoupled allows safer...