1,721,070 research outputs found
Author Correction: Small Molecule Accurate Recognition Technology (SMART) to Enhance Natural Products Research
No full textAn amendment to this paper has been published and can be accessed via a link at the top of the paper
Recommended from our members
Better Cardiac Image Segmentation by Highly Recurrent Neural Networks
Full text linkCardiac magnetic resonance (CMR) image segmentation has been a crucial tool for medical professionals to diagnose cardiovascular diseases (CVDs), which are the leading causes of death throughout the world. Segmenting CMR images is very time consuming and increases the cost of CVD diagnoses and treatment, making them inaccessible to many. Automated CMR image segmentation models strive to lower the cost of CVD diagnosis, but such models must be efficient and accurate in such failure-sensitive domains as human medicine. This thesis proposes to apply γ-Net, a recurrent extension of the popular U-Net, to automatically perform high-quality CMR image segmentation. γ-Net is a recent development by Linsley et al. of Brown University, and has exhibited the ability to outperform U-Net on very small datasets, which is beneficial given the very limited amount of patient CMR data available to the scientific community. γ-Net leverages biological principles backed by anatomical evidence as well as attention mechanisms in order to achieve its high efficiency.In this thesis, we examine the following topics: (a) γ-Net’s resilience to smaller training set sizes, which is cruicial when little patient data is available; (b) resilience to variation in training and validation data, which is shown to significantly degrade performance in state-of-the- art models; and (c) the ability to transfer to new datasets with minimal fine tuning, which saves training cost for practical applications. We have found that (a) γ-Net significantly outperforms an equivalent U-Net in validation performance when trained using a reduced training set; (b) γ-Net is much more resilient to input variations than U-Net; and (c) γ-Net generalizes to new datasets better than comparable U-Nets
Recommended from our members
Where's the Smoke and Fire? Exploiting Spatial and Temporal Context for Improved Wildfire Detection
Full text linkThis thesis covers techniques to improve wildfire detection accuracy through the use of spatial and temporal context. Our dataset of wildfires consists of high-resolution images with smoke plumes that need to be detected early, when they are smallest, to allow firefighters enough time to react. We propose two novel architectures - the first combines a region proposal network on a lower-resolution image with a sequence network on the full-resolution image for efficient, accurate predictions; and the second combines a ResNet feature extractor with a vision transformer for high resolution tiled image classification. We also propose novel loss functions, combining a precise per-grid-element loss with a coarser per-image loss to achieve better precision in our detected fires without increasing the amount of false positives. With these techniques, we achieve state of the art results on this dataset with an accuracy of 89% and an average time to detection of 12 seconds
Recommended from our members
Individual differences in proof structures following multimodal logic teaching
Full text linkWe have been studying how students respond to multimodal logic teaching with Hyperproof. Performance measures have already indicated that students’ pre-existing cognitive styles have a significant impact on teaching outcome. Furthermore, a substantial corpus of proofs has been gathered via automatic logging of proof development. We report results from analyses of final proof structure, exploiting (i) ‘proofograms’, a novel method of proof visualisation, and (ii) corpus-linguistic bigram analysis of rule use. Results suggest that students’ cognitive styles do indeed influence the structure of their logical discourse, and that the effect may be attributable to the relative skill with which students manipulate graphical abstractions
Recommended from our members
Action Recognition from Videos using Deep Neural Networks
Full text linkConvolutional neural network(CNN) models have been extensively used in recent years to solve the problem of image understanding giving state-of-the-art results in tasks like classification, recognition, retrieval, segmentation and object detection. Motivated by this success there have been several attempts to extend convolutional neural networks for video understanding and classification. An important distinction between images and videos is the temporal information that is encoded by the sequence of frames. Most CNN models fail to capture this temporal information. Recurrent neural networks have shown promising results in modelling sequences. In this work we present a neural network model which combines convolutional neural networks and recurrent neural networks. We first evaluate the effect of the convolutional network used for understanding static frames on action recognition. Following this we explore properties that are inherent in the dataset. We combine the representation we get from the convolutional network, the temporal information we get from the sequence of video frames and other properties of the dataset to create a unified model which is trained on the UCF-101 dataset for action recognition. We evaluate our model on the pre-defined test set splits of the UCF-101 dataset. We show that our model is able to achieve an improvement over the baseline model. We show comparison between our models and various models proposed in other related works on the UCF-101 dataset. We observe that a good model for action recognition not only needs to understand static frames but also needs to encode the temporal information across a sequence of frames
Recommended from our members
Leveraging Human Reasoning to Understand and Improve Visual Question Answering
Full text linkVisual Question Answering (VQA) is the task of answering questions based on an image. The field has seen significant advances recently, with systems achieving high accuracy even on open-ended questions. However, a number of recent studies have shown that many of these advanced systems exploit biases in datasets, text of the question or similarity of images in the dataset. To study these reported biases, proposed approaches seek to identify areas of images or words of the questions as evidence that the model focuses on while answering questions. These mechanisms often tend to be limited as the model can answer incorrectly while focusing on the correct region of the image or vice versa.In this thesis, we seek to incorporate and leverage human reasoning to improve interpretability of these VQA models. Essentially, we train models to generate human-like language as evidence or reasons/rationales for the answers that they predict. Further, we show that this type of system has the potential to improve the accuracy on VQA task itself as well
Recommended from our members
Comparison and Fine-Grained Analysis of Sequence Encoders for Natural Language Processing
Full text linkMost machine learning algorithms require a fixed length input to be able to perform commonly desired tasks such as classification, clustering, and regression. For natural language processing, the inherently unbounded and recursive nature of the input poses a unique challenge when deriving such fixed length representations. Although today there is a general consensus on how to generate fixed length representations of individual words which preserve their meaning, the same cannot be said for sequences of words in sentences, paragraphs, or documents. In this work, we study the encoders commonly used to generate fixed length representations of natural language sequences, and analyze their effectiveness across a variety of high and low level tasks including sentence classification and question answering. Additionally, we propose novel improvements to the existing Skip-Thought and End-to-End Memory Network architectures and study their performance on both the original and auxiliary tasks. Ultimately, we show that the setting in which the encoders are trained, and the corpus used for training, have a greater influence of the final learned representation than the underlying sequence encoders themselves
Recommended from our members
Debiasing Image Generative Models
Full text linkGenerative models have become increasingly popular in various domains to solve challenging tasks, including image generation, dialogue generation, and story generation. Unlike discriminative models, they can learn the underlying probability distribution of data and generate new examples. In particular, image generative models have gained significant attention due to their remarkable ability to produce images of unparalleled quality. However, while there has been a lot of attention to biases in discriminative models, bias in generative models has received little attention. The presence of biases in generative models, particularly related to race and gender, can have significant consequences in downstream applications. Therefore, efforts to address this issue are essential to promote fair and ethical use of generative models in various domains. To achieve this goal, this dissertation presents a comprehensive study of debiasing image generative models by incorporating diversity and fairness constraints into the training process.In this dissertation, we investigate three different approaches to debiasing image generative models. In the first approach, a new task of high-fidelity image generation conditioned on multiple attributes from imbalanced datasets is proposed. This task poses new challenges for state-of-the-art GANs models, and a new training framework is proposed to address thesechallenges. The second approach investigates bias in image-to-image translation models and proposes debiasing using contrastive learning. Finally, the study highlights the prevalence of bias in large-pretrained models like CLIP and its impact on text-to-image generative models. Identity preserving losses are proposed to rectify the problem without retraining the pretrained model. In all of these approaches, we evaluate the impact of debiasing on image generation and the effectiveness of existing methods in reducing biases in generated images. We show the proposed task and framework offer new avenues for further research in debiasing generative models. Overall, this dissertation contributes to the field of generative models by providing a comprehensive study of debiasing generative models and proposing a new task and framework for high-fidelity image generation
Recommended from our members
Approximate Spatial Layout Processing in Early Vision
Full text linkImagine yourself running through rough terrain, perhaps
fleeing a predator, or perhaps chasing after prey. Your visual
system does not have time to scrutinize the countless trees,
rocks, and other objects you pass by. What you need most is
enough spatial information to avoid obstacles, to orient yourself,
to pick a path. In this situation, even a rough sketch of
the spatial layout of the environment can provide crucial information
Recommended from our members
Explainable Deep Learning for Biomedical Time Series Classification
Full text linkThrough recent advances in wearable medical devices and subsequent explosion of biological data, deep learning has emerged as a promising approach for the automatic analysis of biomedical time series signals. However, currently deep learning models work as black boxes, and most efforts to explain classification decisions are 1) designed for image classification, 2) only produce local explanations or 3) trade off accuracy for explainability by learning a symbolic interpretable model. In this study, we introduce a post hoc explainability framework for deep networks in the clinical world, which provides model explanations at both global and local levels. Global explanations help to get a birds-eye view of how a model behaves and whether it aligns with the expectations of clinical experts. Local explanations can be used to confer useful information about the model’s behavior for a specific input by highlighting the most important regions or features. We present a comprehensive analysis of this framework through the important clinical problem of detection of atrial fibrillation from single-lead electrocardiography signals
- …
