Convolutional Neural Network Data Research Articles

Comparison of simulation and predictive efficacy for hemorrhagic fever with renal syndrome incidence in mainland China based on five time series models.

Hemorrhagic fever with renal syndrome (HFRS) is a zoonotic infectious disease commonly found in Asia and Europe, characterized by fever, hemorrhage, shock, and renal failure. China is the most severely affected region, necessitating an analysis of the temporal incidence patterns in the country. We employed Autoregressive Integrated Moving Average (ARIMA), Long Short-Term Memory (LSTM), Convolutional Neural Network (CNN), Nonlinear AutoRegressive with eXogenous inputs (NARX), and a hybrid CNN-LSTM model to model and forecast time series data spanning from January 2009 to November 2023 in the mainland China. By comparing the simulated performance of these models on training and testing sets, we determined the most suitable model. Overall, the CNN-LSTM model demonstrated optimal fitting performance (with Root Mean Square Error (RMSE), Mean Absolute Percentage Error (MAPE), and Mean Absolute Error (MAE) of 93.77/270.66, 7.59%/38.96%, and 64.37/189.73 for the training and testing sets, respectively, lower than those of individual CNN or LSTM models). The hybrid CNN-LSTM model seamlessly integrates CNN's data feature extraction and LSTM's recurrent prediction capabilities, rendering it theoretically applicable for simulating diverse distributed time series data. We recommend that the CNN-LSTM model be considered as a valuable time series analysis tool for disease prediction by policy-makers.

Комбинированная модель локализации акустических источников с применением технологии глубокого обучения

This paper presents a deep learning model for sound source localization, which considers localization as a source direction classification problem. An approach is proposed based on the integration of sound intensity features and GCC-PHAT (Generalized cross-correlation - phase transform) features as input data for convolutional neural networks. Taking into account the modeling conditions, datasets were created for the purpose of training, validating and testing the model with spatial resolutions of 10º and 2 º. Simulation results demonstrated the effectiveness of the proposed model in localizing the source with high accuracy in a closed environment and in the presence of reverberation. The proposed model with a resolution of 10º outperformed the model fed with only sound intensity features as input features, achieving improvement in accuracy by 6,57% and in prediction accuracy by 2,86%, while the model with a resolution of 2º achieved an improvement in accuracy by 15,57% and in prediction accuracy by 2,04%.

Different types of drug abusers prefrontal cortex activation patterns and based on machine-learning classification

Drug addiction can cause abnormal brain activation changes, which are the root cause of drug craving and brain function errors. This study enrolled drug abusers to determine the effects of different drugs on brain activation. A functional near-infrared spectroscopy (fNIRS) device was used for the research. This study was designed with an experimental paradigm that included the induction of resting and drug addiction cravings. We collected the fNIRS data of 30 drug users, including 10 who used heroin, 10 who used Methamphetamine, and 10 who used mixed drugs. First, using Statistical Analysis, the study analyzed the activations of eight functional areas of the left and right hemispheres of the prefrontal cortex of drug addicts who respectively used heroin, Methamphetamine, and mixed drugs, including Left/Right-Dorsolateral prefrontal cortex (L/R-DLPFC), Left/Right-Ventrolateral prefrontal cortex (L/R-VLPFC), Left/Right-Frontopolar prefrontal cortex (L/R-FPC), and Left/Right Orbitofrontal Cortex (L/R-OFC). Second, referencing the degrees of activation of oxyhaemoglobin concentration (HbO[Formula: see text], the study made an analysis and got the specific activation patterns of each group of the addicts. Finally, after taking out data which are related to the addicts who recorded high degrees of activation among the three groups of addicts, and which had the same channel numbers, the paper classified the different drug abusers using the data as the input data for Convolutional Neural Networks (CNNs). The average three-class accuracy is 67.13%. It is of great significance for the analysis of brain function errors and personalized rehabilitation.

A convolutional neural network accelerator for real-time underwater image recognition of autonomous underwater vehicle

In the field of underwater image recognition, a chip with smaller footprint and lower energy consumption is required to be implanted into autonomous intelligent underwater vehicle to make real-time response to the surrounding objects. Therefore, a promising accelerator with high performance and low energy consumption is designed, which optimizes the features possessed by convolutional neural network. The sharing of weights between neurons reduces the memory requirement. With all convolutional neural network data stored within on-chip static random-access memory, the need for memory access is drastically decreased. Besides, several small processing elements are used to form neural functional unit, which considerably reduces the bandwidth requirement through inter-processing element data transmission. By sending control signals to autonomous underwater vehicle, this accelerator enables it to avoid dangerous areas such as rocks and algae in time. The result suggests the proposed accelerator successfully achieves a higher processing speed than that of CPU and GPU with a footprint of 6.09 mm2 only and the energy consumption of 327.3 mW at 1 GHz.

Learning from Web Data Using Adversarial Discriminative Neural Networks for Fine-Grained Classification

Fine-grained classification is absorbed in recognizing the subordinate categories of one field, which need a large number of labeled images, while it is expensive to label these images. Utilizing web data has been an attractive option to meet the demands of training data for convolutional neural networks (CNNs), especially when the well-labeled data is not enough. However, directly training on such easily obtained images often leads to unsatisfactory performance due to factors such as noisy labels. This has been conventionally addressed by reducing the noise level of web data. In this paper, we take a fundamentally different view and propose an adversarial discriminative loss to advocate representation coherence between standard and web data. This is further encapsulated in a simple, scalable and end-to-end trainable multi-task learning framework. We experiment on three public datasets using large-scale web data to evaluate the effectiveness and generalizability of the proposed approach. Extensive experiments demonstrate that our approach performs favorably against the state-of-the-art methods.

Dealing with Lack of Training Data for Convolutional Neural Networks: The Case of Digital Pathology

Thanks to their capability to learn generalizable descriptors directly from images, deep Convolutional Neural Networks (CNNs) seem the ideal solution to most pattern recognition problems. On the other hand, to learn the image representation, CNNs need huge sets of annotated samples that are unfeasible in many every-day scenarios. This is the case, for example, of Computer-Aided Diagnosis (CAD) systems for digital pathology, where additional challenges are posed by the high variability of the cancerous tissue characteristics. In our experiments, state-of-the-art CNNs trained from scratch on histological images were less accurate and less robust to variability than a traditional machine learning framework, highlighting all the issues of fully training deep networks with limited data from real patients. To solve this problem, we designed and compared three transfer learning frameworks, leveraging CNNs pre-trained on non-medical images. This approach obtained very high accuracy, requiring much less computational resource for the training. Our findings demonstrate that transfer learning is a solution to the automated classification of histological samples and solves the problem of designing accurate and computationally-efficient CAD systems with limited training data.

Pedestrian detection with motion features via two-stream ConvNets

Motion information can be important for detecting objects, but it has been used less for pedestrian detection, particularly with deep-learning-based methods. We propose a method that uses deep motion features as well as deep still-image features, following the success of two-stream convolutional networks, each of which are trained separately for spatial and temporal streams. To extract motion clues for detection differentiated from other background motions, the temporal stream takes as input the difference in frames that are weakly stabilized by optical flow. To make the networks applicable to bounding-box-level detection, the mid-level features are concatenated and combined with a sliding-window detector. We also introduce transfer learning from multiple sources in the two-stream networks, which can transfer still image and motion features from ImageNet and an action recognition dataset respectively, to overcome the insufficiency of training data for convolutional neural networks in pedestrian datasets. We conducted an evaluation on two popular large-scale pedestrian benchmarks, namely the Caltech Pedestrian Detection Benchmark and Daimler Mono Pedestrian Detection Benchmark. We observed 10% improvement compared to the same method but without motion features.