scholarly journals Deep learning approaches for neural decoding across architectures and recording modalities

Author(s):  
Jesse A Livezey ◽  
Joshua I Glaser

Abstract Decoding behavior, perception or cognitive state directly from neural signals is critical for brain–computer interface research and an important tool for systems neuroscience. In the last decade, deep learning has become the state-of-the-art method in many machine learning tasks ranging from speech recognition to image segmentation. The success of deep networks in other domains has led to a new wave of applications in neuroscience. In this article, we review deep learning approaches to neural decoding. We describe the architectures used for extracting useful features from neural recording modalities ranging from spikes to functional magnetic resonance imaging. Furthermore, we explore how deep learning has been leveraged to predict common outputs including movement, speech and vision, with a focus on how pretrained deep networks can be incorporated as priors for complex decoding targets like acoustic speech or images. Deep learning has been shown to be a useful tool for improving the accuracy and flexibility of neural decoding across a wide range of tasks, and we point out areas for future scientific development.

2020 ◽  
Author(s):  
Yuan Yuan ◽  
Lei Lin

Satellite image time series (SITS) classification is a major research topic in remote sensing and is relevant for a wide range of applications. Deep learning approaches have been commonly employed for SITS classification and have provided state-of-the-art performance. However, deep learning methods suffer from overfitting when labeled data is scarce. To address this problem, we propose a novel self-supervised pre-training scheme to initialize a Transformer-based network by utilizing large-scale unlabeled data. In detail, the model is asked to predict randomly contaminated observations given an entire time series of a pixel. The main idea of our proposal is to leverage the inherent temporal structure of satellite time series to learn general-purpose spectral-temporal representations related to land cover semantics. Once pre-training is completed, the pre-trained network can be further adapted to various SITS classification tasks by fine-tuning all the model parameters on small-scale task-related labeled data. In this way, the general knowledge and representations about SITS can be transferred to a label-scarce task, thereby improving the generalization performance of the model as well as reducing the risk of overfitting. Comprehensive experiments have been carried out on three benchmark datasets over large study areas. Experimental results demonstrate the effectiveness of the proposed method, leading to a classification accuracy increment up to 1.91% to 6.69%. <div><b>This work has been submitted to the IEEE for possible publication. Copyright may be transferred without notice, after which this version may no longer be accessible.</b></div>


Author(s):  
Jun-Li Xu ◽  
Cecilia Riccioli ◽  
Ana Herrero-Langreo ◽  
Aoife Gowen

Deep learning (DL) has recently achieved considerable successes in a wide range of applications, such as speech recognition, machine translation and visual recognition. This tutorial provides guidelines and useful strategies to apply DL techniques to address pixel-wise classification of spectral images. A one-dimensional convolutional neural network (1-D CNN) is used to extract features from the spectral domain, which are subsequently used for classification. In contrast to conventional classification methods for spectral images that examine primarily the spectral context, a three-dimensional (3-D) CNN is applied to simultaneously extract spatial and spectral features to enhance classificationaccuracy. This tutorial paper explains, in a stepwise manner, how to develop 1-D CNN and 3-D CNN models to discriminate spectral imaging data in a food authenticity context. The example image data provided consists of three varieties of puffed cereals imaged in the NIR range (943–1643 nm). The tutorial is presented in the MATLAB environment and scripts and dataset used are provided. Starting from spectral image pre-processing (background removal and spectral pre-treatment), the typical steps encountered in development of CNN models are presented. The example dataset provided demonstrates that deep learning approaches can increase classification accuracy compared to conventional approaches, increasing the accuracy of the model tested on an independent image from 92.33 % using partial least squares-discriminant analysis to 99.4 % using 3-CNN model at pixel level. The paper concludes with a discussion on the challenges and suggestions in the application of DL techniques for spectral image classification.


Sensors ◽  
2019 ◽  
Vol 19 (18) ◽  
pp. 3929 ◽  
Author(s):  
Grigorios Tsagkatakis ◽  
Anastasia Aidini ◽  
Konstantina Fotiadou ◽  
Michalis Giannopoulos ◽  
Anastasia Pentari ◽  
...  

Deep Learning, and Deep Neural Networks in particular, have established themselves as the new norm in signal and data processing, achieving state-of-the-art performance in image, audio, and natural language understanding. In remote sensing, a large body of research has been devoted to the application of deep learning for typical supervised learning tasks such as classification. Less yet equally important effort has also been allocated to addressing the challenges associated with the enhancement of low-quality observations from remote sensing platforms. Addressing such channels is of paramount importance, both in itself, since high-altitude imaging, environmental conditions, and imaging systems trade-offs lead to low-quality observation, as well as to facilitate subsequent analysis, such as classification and detection. In this paper, we provide a comprehensive review of deep-learning methods for the enhancement of remote sensing observations, focusing on critical tasks including single and multi-band super-resolution, denoising, restoration, pan-sharpening, and fusion, among others. In addition to the detailed analysis and comparison of recently presented approaches, different research avenues which could be explored in the future are also discussed.


2020 ◽  
Author(s):  
Yuan Yuan ◽  
Lei Lin

<div>Satellite image time series (SITS) classification is a major research topic in remote sensing and is relevant for a wide range of applications. Deep learning approaches have been commonly employed for SITS classification and have provided state-of-the-art performance. However, deep learning methods suffer from overfitting when labeled data is scarce. To address this problem, we propose a novel self-supervised pre-training scheme to initialize a Transformer-based network by utilizing large-scale unlabeled data. In detail, the model is asked to predict randomly contaminated observations given an entire time series of a pixel. The main idea of our proposal is to leverage the inherent temporal structure of satellite time series to learn general-purpose spectral-temporal representations related to land cover semantics. Once pre-training is completed, the pre-trained network can be further adapted to various SITS classification tasks by fine-tuning all the model parameters on small-scale task-related labeled data. In this way, the general knowledge and representations about SITS can be transferred to a label-scarce task, thereby improving the generalization performance of the model as well as reducing the risk of overfitting. Comprehensive experiments have been carried out on three benchmark datasets over large study areas. Experimental results demonstrate the effectiveness of the proposed method, leading to a classification accuracy increment up to 2.38% to 5.27%. The code and the pre-trained model will be available at https://github.com/linlei1214/SITS-BERT upon publication.</div><div><b>This work has been submitted to the IEEE for possible publication. Copyright may be transferred without notice, after which this version may no longer be accessible.</b></div>


2020 ◽  
Author(s):  
Harshvardhan Sikka

One of the popular directions in Deep Learning (DL) research has been to build larger and more complex deep networks that can perform well on several different learning tasks, commonly known as multitask learning. This work is usually done within specific domains, e.g. multitask models that perform captioning, translation, and text classification tasks. Some work has been done in building multimodal/crossmodal networks that use deep networks with a combination of different neural network primitives (Convolutional Layers, Recurrent Layers, Mixture of Expert layers, etc). This paper explores various topics and ideas that may prove relevant to large, sparse, multitask networks and explores the potential for a general approach to building and managing these networks. A framework to automatically build, update, and interpret modular LSMNs is presented in the context of current tooling and theory.


2021 ◽  
Vol 2021 (1) ◽  
pp. 128-144
Author(s):  
Jaewoo Lee ◽  
Daniel Kifer

AbstractRecent work on Renyi Differential Privacy has shown the feasibility of applying differential privacy to deep learning tasks. Despite their promise, however, differentially private deep networks often lag far behind their non-private counterparts in accuracy, showing the need for more research in model architectures, optimizers, etc. One of the barriers to this expanded research is the training time — often orders of magnitude larger than training non-private networks. The reason for this slowdown is a crucial privacy-related step called “per-example gradient clipping” whose naive implementation undoes the benefits of batch training with GPUs. By analyzing the back-propagation equations we derive new methods for per-example gradient clipping that are compatible with auto-differeniation (e.g., in Py-Torch and TensorFlow) and provide better GPU utilization. Our implementation in PyTorch showed significant training speed-ups (by factors of 54x - 94x for training various models with batch sizes of 128). These techniques work for a variety of architectural choices including convolutional layers, recurrent networks, attention, residual blocks, etc.


2022 ◽  
Vol 12 ◽  
Author(s):  
Radek Zenkl ◽  
Radu Timofte ◽  
Norbert Kirchgessner ◽  
Lukas Roth ◽  
Andreas Hund ◽  
...  

Robust and automated segmentation of leaves and other backgrounds is a core prerequisite of most approaches in high-throughput field phenotyping. So far, the possibilities of deep learning approaches for this purpose have not been explored adequately, partly due to a lack of publicly available, appropriate datasets. This study presents a workflow based on DeepLab v3+ and on a diverse annotated dataset of 190 RGB (350 x 350 pixels) images. Images of winter wheat plants of 76 different genotypes and developmental stages have been acquired throughout multiple years at high resolution in outdoor conditions using nadir view, encompassing a wide range of imaging conditions. Inconsistencies of human annotators in complex images have been quantified, and metadata information of camera settings has been included. The proposed approach achieves an intersection over union (IoU) of 0.77 and 0.90 for plants and soil, respectively. This outperforms the benchmarked machine learning methods which use Support Vector Classifier and/or Random Forrest. The results show that a small but carefully chosen and annotated set of images can provide a good basis for a powerful segmentation pipeline. Compared to earlier methods based on machine learning, the proposed method achieves better performance on the selected dataset in spite of using a deep learning approach with limited data. Increasing the amount of publicly available data with high human agreement on annotations and further development of deep neural network architectures will provide high potential for robust field-based plant segmentation in the near future. This, in turn, will be a cornerstone of data-driven improvement in crop breeding and agricultural practices of global benefit.


Author(s):  
Shulin Xiang ◽  
Tao Chen ◽  
Zhichao Fan ◽  
Xuedong Chen ◽  
Zhigang Wu ◽  
...  

Abstract With the development of Materials Genome Initiative (MGI) and data mining technology, machine learning (ML) has emerged as an important tool in the research of materials science. For the heat resistant alloys used in furnace tubes, the rapid prediction of the high-temperature properties is critical but difficult until now. In this work, the ML method based on the deep learning algorithm is developed to establish the direct correlation between microstructure inputs and output stress rupture properties of Fe-Cr-Ni based heat resistant alloys. Two simple convolutional neural networks (CNN) and the complex network with VGG16 architecture are implemented and evaluated. The simple CNN and VGG16 models are trained from scratch and pre-trained, respectively. Due to the relatively few training samples in the dataset, the data augmentation configuration and the improved architecture are effective to mitigate overfitting in simple CNN models. The result also shows that in the case of transfer learning, the features extracted from other datasets can be used directly to this new visual task. It is demonstrated that both the simple CNN and VGG16 models reach the high prediction accuracies (more than 90 %) of high-temperature properties with a wide range of microstructures. In addition, the good prediction performance achieved in the small dataset also reveals the deep learning approaches can be used to construct powerful vision models in engineering practice, where very limited data is the common situation.


2020 ◽  
Author(s):  
Yuan Yuan ◽  
Lei Lin

<div>Satellite image time series (SITS) classification is a major research topic in remote sensing and is relevant for a wide range of applications. Deep learning approaches have been commonly employed for SITS classification and have provided state-of-the-art performance. However, deep learning methods suffer from overfitting when labeled data is scarce. To address this problem, we propose a novel self-supervised pre-training scheme to initialize a Transformer-based network by utilizing large-scale unlabeled data. In detail, the model is asked to predict randomly contaminated observations given an entire time series of a pixel. The main idea of our proposal is to leverage the inherent temporal structure of satellite time series to learn general-purpose spectral-temporal representations related to land cover semantics. Once pre-training is completed, the pre-trained network can be further adapted to various SITS classification tasks by fine-tuning all the model parameters on small-scale task-related labeled data. In this way, the general knowledge and representations about SITS can be transferred to a label-scarce task, thereby improving the generalization performance of the model as well as reducing the risk of overfitting. Comprehensive experiments have been carried out on three benchmark datasets over large study areas. Experimental results demonstrate the effectiveness of the proposed method, leading to a classification accuracy increment up to 2.38% to 5.27%. The code and the pre-trained model will be available at https://github.com/linlei1214/SITS-BERT upon publication.</div><div><b>This work has been submitted to the IEEE for possible publication. Copyright may be transferred without notice, after which this version may no longer be accessible.</b></div>


2020 ◽  
Author(s):  
Yuan Yuan ◽  
Lei Lin

<div>Satellite image time series (SITS) classification is a major research topic in remote sensing and is relevant for a wide range of applications. Deep learning approaches have been commonly employed for SITS classification and have provided state-of-the-art performance. However, deep learning methods suffer from overfitting when labeled data is scarce. To address this problem, we propose a novel self-supervised pre-training scheme to initialize a Transformer-based network by utilizing large-scale unlabeled data. In detail, the model is asked to predict randomly contaminated observations given an entire time series of a pixel. The main idea of our proposal is to leverage the inherent temporal structure of satellite time series to learn general-purpose spectral-temporal representations related to land cover semantics. Once pre-training is completed, the pre-trained network can be further adapted to various SITS classification tasks by fine-tuning all the model parameters on small-scale task-related labeled data. In this way, the general knowledge and representations about SITS can be transferred to a label-scarce task, thereby improving the generalization performance of the model as well as reducing the risk of overfitting. Comprehensive experiments have been carried out on three benchmark datasets over large study areas. Experimental results demonstrate the effectiveness of the proposed method, leading to a classification accuracy increment up to 2.38% to 5.27%. The code and the pre-trained model will be available at https://github.com/linlei1214/SITS-BERT upon publication.</div><div><b>This work has been submitted to the IEEE for possible publication. Copyright may be transferred without notice, after which this version may no longer be accessible.</b></div>


Sign in / Sign up

Export Citation Format

Share Document