Neural Computing and Applications最新文献

Artificial neural networks (ANNs) are one of the most promising tools in the quest to develop general artificial intelligence. Their design was inspired by how neurons in natural brains connect and process, the only other substrate to harbor intelligence. Compared to biological brains that are sparsely connected and that form sparsely distributed representations, ANNs instead process information by connecting all nodes of one layer to all nodes of the next. In addition, modern ANNs are trained with backpropagation, while their natural counterparts have been optimized by natural evolution over eons. We study whether the training method influences how information propagates through the brain by measuring the transfer entropy, that is, the information that is transferred from one group of neurons to another. We find that while the distribution of connection weights in optimized networks is largely unaffected by the training method, neuroevolution leads to networks in which information transfer is significantly more focused on small groups of neurons (compared to those trained by backpropagation) while also being more robust to perturbations of the weights. We conclude that the specific attributes of a training method (local vs. global) can significantly affect how information is processed and relayed through the brain, even when the overall performance is similar.

Due to its perceptual limitations, an agent may have too little information about the environment to act optimally. In such cases, it is important to keep track of the action-observation history to uncover hidden state information. Recent deep reinforcement learning methods use recurrent neural networks (RNN) to memorize past observations. However, these models are expensive to train and have convergence difficulties, especially when dealing with high dimensional data. In this paper, we propose influence-aware memory, a theoretically inspired memory architecture that alleviates the training difficulties by restricting the input of the recurrent layers to those variables that influence the hidden state information. Moreover, as opposed to standard RNNs, in which every piece of information used for estimating Q values is inevitably fed back into the network for the next prediction, our model allows information to flow without being necessarily stored in the RNN's internal memory. Results indicate that, by letting the recurrent layers focus on a small fraction of the observation variables while processing the rest of the information with a feedforward neural network, we can outperform standard recurrent architectures both in training speed and policy performance. This approach also reduces runtime and obtains better scores than methods that stack multiple observations to remove partial observability.

This paper presents a new strategy that uses multiple neural networks in conjunction with the DEtection TRansformer (DETR) network to detect firearms in surveillance images. The strategy developed in this work presents a methodology that promotes collaboration and self-coordination of networks in the fully connected layers of DETR through the technique of multiple self-coordinating artificial neural networks (MANN), which does not require a coordinator. This self-coordination consists of training the networks one after the other and integrating their outputs without an extra element called a coordinator. The results indicate that the proposed network is highly effective, achieving high-level outcomes in firearm detection. The network’s high precision of 84% and its ability to perform classifications are noteworthy.

The X-ray method, together with image analysis tools, has been used to evaluate the internal structures of seeds and correlate them with the physical, physiological and sanitary quality, providing significant and accurate results. The objective of this study was to analyze radiographic images of rice seeds infested by the rice weevil Sitophilus oryzae (Linnaeus, 1763) (Coleoptera: Curculionidae). Rice seed samples from three different cultivars were infested with S. oryzae for 90 days. Next, seed samples collected at random were analyzed by X-ray testing. The radiographic images were analyzed by ImageJ® software to extract color and shape features. Scanning electron microscopy analyses were also performed. The results showed that X-ray testing was effective in detecting infestation. The gray distribution histograms revealed differences between healthy seeds and those infested by adult insects or empty seeds, confirmed by the significant differences obtained for the area and relative and integrated density variables. The study demonstrated that the analysis of radiographic images can provide quantitative information on insect infestation of rice seeds, which is useful in the evaluation of seed quality and for detecting the presence of pests in rice seeds.

Most recommendation systems are implemented using numerical or categorical data, that is, traditional data. This type of data can be a limiting factor when used to model complex concepts where there is internal variability or internal structure in the data. To overcome these limitations, symbolic data is used, where data can be represented by different types of values, such as intervals, lists, or histograms. This work introduces a single approach to constructing recommendation systems based on content or based on collaborative filtering using modal variables for users and items. In the content-based system, user profiles and item profiles are created from modal representations of their features, and a list of items is matched against a user profile. For collaborative filtering, user profiles are built, and users are grouped to form a neighborhood, products rated by users of this neighborhood are recommended based on the similarity between the neighbor and the user who will receive the recommendation. Experiments are carried out, using a movie domain dataset, to evaluate the effectiveness of the proposed approach. The outcomes suggest our ability to generate ranked lists of superior quality compared to previous methods utilizing symbolic data. Specifically, the lists created through the proposed method exhibit higher normalized discounted cumulative gain and, in qualitative terms, showcase more diverse content.

X-ray Bragg coherent diffraction imaging is a powerful technique for 3D materials characterization. However, obtaining X-ray diffraction data is difficult and computationally intensive, motivating the need for automated processing of coherent diffraction images, with the goal of minimizing the number of X-ray datasets needed. We automate a machine learning approach to identify crystalline line defects in samples from the raw coherent diffraction data, in a workflow coupling coherent diffraction data generation with training and inference of deep neural network defect classifiers. In particular, we adopt a continual learning approach, where we generate training data as needed based on the accuracy of the defect classifier instead of generating all training data a priori. Moreover, we develop a novel data generation mechanism to improve the efficiency of defect identification beyond the previously published continual learning approach. We call the improved method smart continual learning. The results show that our approach improves the accuracy of defect classifiers and reduces training data requirements by up to 98% compared with prior approaches.

This study investigates the use of electroencephalography (EEG) to characterize emotions and provides insights into the consistency between self-reported and machine learning outcomes. Thirty participants engaged in five virtual reality environments designed to elicit specific emotions, while their brain activity was recorded. The participants self-assessed their ground truth emotional state in terms of Arousal and Valence through a Self-Assessment Manikin. Gradient Boosted Decision Tree was adopted as a classification algorithm to test the EEG feasibility in the characterization of emotional states. Distinctive patterns of neural activation corresponding to different levels of Valence and Arousal emerged, and a noteworthy correspondence between the outcomes of the self-assessments and the classifier suggested that EEG-based affective indicators can be successfully applied in emotional characterization, shedding light on the possibility of using them as ground truth measurements. These findings provide compelling evidence for the validity of EEG as a tool for emotion characterization and its contribution to a better understanding of emotional activation.

A novel methodology is introduced for extracting entities from noisy scanned documents by using end-to-end data and reformulating the entity extraction task as a text summarization problem. This approach offers two significant advantages over traditional entity extraction methods while maintaining comparable performance. First, it utilizes preexisting data to construct datasets, thereby eliminating the need for labor-intensive annotation procedures. Second, it employs multitask learning, enabling the training of a model via a single dataset. To evaluate our approach against state-of-the-art methods, we adapted three commonly used datasets, namely, Conference on Natural Language Learning (CoNLL++), few-shot named entity recognition (Few-NERD), and WikiNEuRal domain adaptation (WikiNEuRal + DA), to the format required by our methodology. We subsequently fine-tuned four sequence-to-sequence models: text-to-text transfer transformer (T5), fine-tuned language net T5 (FLAN-T5), bidirectional autoregressive transformer (BART), and pretraining with extracted gap sentences for abstractive summarization sequence-to-sequence models (PEGASUS). The results indicate that, in the absence of optical character recognition (OCR) noise, the BART model performs comparably to state-of-the-art methods. Furthermore, the performance degradation was limited to 3.49–5.23% when 39–62% of the sentences contained OCR noise. This performance is significantly superior to that of previous studies, which reported a 10–20% decrease in the F1 score with texts that had a 20% OCR error rate. Our experimental results demonstrate that a single model trained via our methodology can reliably extract entities from noisy OCRed texts, unlike existing state-of-the-art approaches, which require separate models for correcting OCR errors and extracting entities.

The object detection model based on the transfer learning approach comprises feature extraction and detection layers. YOLOv2 is among the fastest detection algorithms, which can utilize various pretrained classifier networks for feature extraction. However, reducing the number of network layers and increasing the mean average precision (mAP) together have challenges. Darknet-19-based YOLOv2 model achieved an mAP of 76.78% by having a smaller number of layers than other existing models. This work proposes modification by adding layers that help enhance feature extraction for further increasing the mAP of the model. Above that, the initial weights of the new layers can be random or deterministic, fine-tuned during training. In our work, we introduce a block of layers initialized with deterministic weights derived from several edge detection filter weights. Integrating such a block to the darknet-19-based object detection model improves the mAP to 85.94%, outperforming the other existing model in terms of mAP and number of layers.

In computer-aided diagnosis tools employed for skin cancer treatment and early diagnosis, skin lesion segmentation is important. However, achieving precise segmentation is challenging due to inherent variations in appearance, contrast, texture, and blurry lesion boundaries. This research presents a robust approach utilizing a dilated convolutional residual network, which incorporates an attention-based spatial feature enhancement block (ASFEB) and employs a guided decoder strategy. In each dilated convolutional residual block, dilated convolution is employed to broaden the receptive field with varying dilation rates. To improve the spatial feature information of the encoder, we employed an attention-based spatial feature enhancement block in the skip connections. The ASFEB in our proposed method combines feature maps obtained from average and maximum-pooling operations. These combined features are then weighted using the active outcome of global average pooling and convolution operations. Additionally, we have incorporated a guided decoder strategy, where each decoder block is optimized using an individual loss function to enhance the feature learning process in the proposed AD-Net. The proposed AD-Net presents a significant benefit by necessitating fewer model parameters compared to its peer methods. This reduction in parameters directly impacts the number of labeled data required for training, facilitating faster convergence during the training process. The effectiveness of the proposed AD-Net was evaluated using four public benchmark datasets. We conducted a Wilcoxon signed-rank test to verify the efficiency of the AD-Net. The outcomes suggest that our method surpasses other cutting-edge methods in performance, even without the implementation of data augmentation strategies.