The brain, as the source of inspiration for Artificial Neural Networks (ANN), is based on a sparse structure. This sparse structure helps the brain to consume less energy, learn easier and generalize patterns better than any other ANN. In this paper, two evolutionary methods for adopting sparsity to ANNs are proposed. In the proposed methods, the sparse structure of a network as well as the values of its parameters are trained and updated during the learning process. The simulation results show that these two methods have better accuracy and faster convergence while they need fewer training samples compared to their sparse and non-sparse counterparts. Furthermore, the proposed methods significantly improve the generalization power and reduce the number of parameters. For example, the sparsification of the ResNet47 network by exploiting our proposed methods for the image classification of ImageNet dataset uses 40 % fewer parameters while the top-1 accuracy of the model improves by 12% and 5% compared to the dense network and their sparse counterpart, respectively. As another example, the proposed methods for the CIFAR10 dataset converge to their final structure 7 times faster than its sparse counterpart, while the final accuracy increases by 6%.
Image clustering is a particularly challenging computer vision task, which aims to generate annotations without human supervision. Recent advances focus on the use of self-supervised learning strategies in image clustering, by first learning valuable semantics and then clustering the image representations. These multiple-phase algorithms, however, increase the computational time and their final performance is reliant on the first stage. By extending the self-supervised approach, we propose a novel single-phase clustering method that simultaneously learns meaningful representations and assigns the corresponding annotations. This is achieved by integrating a discrete representation into the self-supervised paradigm through a classifier net. Specifically, the proposed clustering objective employs mutual information, and maximizes the dependency between the integrated discrete representation and a discrete probability distribution. The discrete probability distribution is derived though the self-supervised process by comparing the learnt latent representation with a set of trainable prototypes. To enhance the learning performance of the classifier, we jointly apply the mutual information across multi-crop views. Our empirical results show that the proposed framework outperforms state-of-the-art techniques with the average accuracy of 89.1% and 49.0%, respectively, on CIFAR-10 and CIFAR-100/20 datasets. Finally, the proposed method also demonstrates attractive robustness to parameter settings, making it ready to be applicable to other datasets.
Quantizing deep neural networks is an effective method for reducing memory consumption and improving inference speed, and is thus useful for implementation in resource-constrained devices. However, it is still hard for extremely low-bit models to achieve accuracy comparable with that of full-precision models. To address this issue, we propose learnable companding quantization (LCQ) as a novel non-uniform quantization method for 2-, 3-, and 4-bit models. LCQ jointly optimizes model weights and learnable companding functions that can flexibly and non-uniformly control the quantization levels of weights and activations. We also present a new weight normalization technique that allows more stable training for quantization. Experimental results show that LCQ outperforms conventional state-of-the-art methods and narrows the gap between quantized and full-precision models for image classification and object detection tasks. Notably, the 2-bit ResNet-50 model on ImageNet achieves top-1 accuracy of 75.1% and reduces the gap to 1.7%, allowing LCQ to further exploit the potential of non-uniform quantization.
Interleaving learning is a human learning technique where a learner interleaves the studies of multiple topics, which increases long-term retention and improves ability to transfer learned knowledge. Inspired by the interleaving learning technique of humans, in this paper we explore whether this learning methodology is beneficial for improving the performance of machine learning models as well. We propose a novel machine learning framework referred to as interleaving learning (IL). In our framework, a set of models collaboratively learn a data encoder in an interleaving fashion: the encoder is trained by model 1 for a while, then passed to model 2 for further training, then model 3, and so on; after trained by all models, the encoder returns back to model 1 and is trained again, then moving to model 2, 3, etc. This process repeats for multiple rounds. Our framework is based on multi-level optimization consisting of multiple inter-connected learning stages. An efficient gradient-based algorithm is developed to solve the multi-level optimization problem. We apply interleaving learning to search neural architectures for image classification on CIFAR-10, CIFAR-100, and ImageNet. The effectiveness of our method is strongly demonstrated by the experimental results.
Coronavirus disease 2019 (COVID-19) is an ongoing global pandemic in over 200 countries and territories, which has resulted in a great public health concern across the international community. Analysis of X-ray imaging data can play a critical role in timely and accurate screening and fighting against COVID-19. Supervised deep learning has been successfully applied to recognize COVID-19 pathology from X-ray imaging datasets. However, it requires a substantial amount of annotated X-ray images to train models, which is often not applicable to data analysis for emerging events such as COVID-19 outbreak, especially in the early stage of the outbreak. To address this challenge, this paper proposes a two-path semi-supervised deep learning model, ssResNet, based on Residual Neural Network (ResNet) for COVID-19 image classification, where two paths refer to a supervised path and an unsupervised path, respectively. Moreover, we design a weighted supervised loss that assigns higher weight for the minority classes in the training process to resolve the data imbalance. Experimental results on a large-scale of X-ray image dataset COVIDx demonstrate that the proposed model can achieve promising performance even when trained on very few labeled training images.
Recent strategies achieved ensembling "for free" by fitting concurrently diverse subnetworks inside a single base network. The main idea during training is that each subnetwork learns to classify only one of the multiple inputs simultaneously provided. However, the question of how to best mix these multiple inputs has not been studied so far. In this paper, we introduce MixMo, a new generalized framework for learning multi-input multi-output deep subnetworks. Our key motivation is to replace the suboptimal summing operation hidden in previous approaches by a more appropriate mixing mechanism. For that purpose, we draw inspiration from successful mixed sample data augmentations. We show that binary mixing in features - particularly with rectangular patches from CutMix - enhances results by making subnetworks stronger and more diverse. We improve state of the art for image classification on CIFAR-100 and Tiny ImageNet datasets. Our easy to implement models notably outperform data augmented deep ensembles, without the inference and memory overheads. As we operate in features and simply better leverage the expressiveness of large networks, we open a new line of research complementary to previous works.
Self-supervised learning has been widely used to obtain transferrable representations from unlabeled images. Especially, recent contrastive learning methods have shown impressive performances on downstream image classification tasks. While these contrastive methods mainly focus on generating invariant global representations at the image-level under semantic-preserving transformations, they are prone to overlook spatial consistency of local representations and therefore have a limitation in pretraining for localization tasks such as object detection and instance segmentation. Moreover, aggressively cropped views used in existing contrastive methods can minimize representation distances between the semantically different regions of a single image. In this paper, we propose a spatially consistent representation learning algorithm (SCRL) for multi-object and location-specific tasks. In particular, we devise a novel self-supervised objective that tries to produce coherent spatial representations of a randomly cropped local region according to geometric translations and zooming operations. On various downstream localization tasks with benchmark datasets, the proposed SCRL shows significant performance improvements over the image-level supervised pretraining as well as the state-of-the-art self-supervised learning methods. Code is available at https://github.com/kakaobrain/scrl
Over the past decade, convolutional neural networks (CNN) have shown very competitive performance in medical image analysis tasks, such as disease classification, tumor segmentation, and lesion detection. CNN has great advantages in extracting local features of images. However, due to the locality of convolution operation, it can not deal with long-range relationships well. Recently, transformers have been applied to computer vision and achieved remarkable success in large-scale datasets. Compared with natural images, multi-modal medical images have explicit and important long-range dependencies, and effective multi-modal fusion strategies can greatly improve the performance of deep models. This prompts us to study transformer-based structures and apply them to multi-modal medical images. Existing transformer-based network architectures require large-scale datasets to achieve better performance. However, medical imaging datasets are relatively small, which makes it difficult to apply pure transformers to medical image analysis. Therefore, we propose TransMed for multi-modal medical image classification. TransMed combines the advantages of CNN and transformer to efficiently extract low-level features of images and establish long-range dependencies between modalities. We evaluated our model for the challenging problem of preoperative diagnosis of parotid gland tumors, and the experimental results show the advantages of our proposed method. We argue that the combination of CNN and transformer has tremendous potential in a large number of medical image analysis tasks. To our best knowledge, this is the first work to apply transformers to medical image classification.
Region modification-based data augmentation techniques have shown to improve performance for high level vision tasks (object detection, semantic segmentation, image classification, etc.) by encouraging underlying algorithms to focus on multiple discriminative features. However, as these techniques destroy spatial relationship with neighboring regions, performance can be deteriorated when using them to train algorithms designed for low level vision tasks (low light image enhancement, image dehazing, deblurring, etc.) where textural consistency between recovered and its neighboring regions is important to ensure effective performance. In this paper, we examine the efficacy of a simple copy-blend data augmentation technique that copies patches from noisy images and blends onto a clean image and vice versa to ensure that an underlying algorithm localizes and recovers affected regions resulting in increased perceptual quality of a recovered image. To assess performance improvement, we perform extensive experiments alongside different region modification-based augmentation techniques and report observations such as improved performance, reduced requirement for training dataset, and early convergence across tasks such as low light image enhancement, image dehazing and image deblurring without any modification to baseline algorithm.
Convolution has been the core ingredient of modern neural networks, triggering the surge of deep learning in vision. In this work, we rethink the inherent principles of standard convolution for vision tasks, specifically spatial-agnostic and channel-specific. Instead, we present a novel atomic operation for deep neural networks by inverting the aforementioned design principles of convolution, coined as involution. We additionally demystify the recent popular self-attention operator and subsume it into our involution family as an over-complicated instantiation. The proposed involution operator could be leveraged as fundamental bricks to build the new generation of neural networks for visual recognition, powering different deep learning models on several prevalent benchmarks, including ImageNet classification, COCO detection and segmentation, together with Cityscapes segmentation. Our involution-based models improve the performance of convolutional baselines using ResNet-50 by up to 1.6% top-1 accuracy, 2.5% and 2.4% bounding box AP, and 4.7% mean IoU absolutely while compressing the computational cost to 66%, 65%, 72%, and 57% on the above benchmarks, respectively. Code and pre-trained models for all the tasks are available at https://github.com/d-li14/involution.
The intuition that local flatness of the loss landscape is correlated with better generalization for deep neural networks (DNNs) has been explored for decades, spawning many different flatness measures. Recently, this link with generalization has been called into question by a demonstration that many measures of flatness are vulnerable to parameter re-scaling which arbitrarily changes their value without changing neural network outputs. Here we show that, in addition, some popular variants of SGD such as Adam and Entropy-SGD, can also break the flatness-generalization correlation. As an alternative to flatness measures, we use a function based picture and propose using the log of Bayesian prior upon initialization, $\log P(f)$, as a predictor of the generalization when a DNN converges on function $f$ after training to zero error. The prior is directly proportional to the Bayesian posterior for functions that give zero error on a test set. For the case of image classification, we show that $\log P(f)$ is a significantly more robust predictor of generalization than flatness measures are. Whilst local flatness measures fail under parameter re-scaling, the prior/posterior, which is global quantity, remains invariant under re-scaling. Moreover, the correlation with generalization as a function of data complexity remains good for different variants of SGD.
In this paper, we propose a novel data augmentation strategy named Cut-Thumbnail, that aims to improve the shape bias of the network. We reduce an image to a certain size and replace the random region of the original image with the reduced image. The generated image not only retains most of the original image information but also has global information in the reduced image. We call the reduced image as thumbnail. Furthermore, we find that the idea of thumbnail can be perfectly integrated with Mixed Sample Data Augmentation, so we put one image's thumbnail on another image while the ground truth labels are also mixed, making great achievements on various computer vision tasks. Extensive experiments show that Cut-Thumbnail works better than state-of-the-art augmentation strategies across classification, fine-grained image classification, and object detection. On ImageNet classification, ResNet-50 architecture with our method achieves 79.21\% accuracy, which is more than 2.8\% improvement on the baseline.
Being expensive and time-consuming to collect massive COVID-19 image samples to train deep classification models, transfer learning is a promising approach by transferring knowledge from the abundant typical pneumonia datasets for COVID-19 image classification. However, negative transfer may deteriorate the performance due to the feature distribution divergence between two datasets and task semantic difference in diagnosing pneumonia and COVID-19 that rely on different characteristics. It is even more challenging when the target dataset has no labels available, i.e., unsupervised task transfer learning. In this paper, we propose a novel Task Adaptation Network (TAN) to solve this unsupervised task transfer problem. In addition to learning transferable features via domain-adversarial training, we propose a novel task semantic adaptor that uses the learning-to-learn strategy to adapt the task semantics. Experiments on three public COVID-19 datasets demonstrate that our proposed method achieves superior performance. Especially on COVID-DA dataset, TAN significantly increases the recall and F1 score by 5.0% and 7.8% compared to recently strong baselines. Moreover, we show that TAN also achieves superior performance on several public domain adaptation benchmarks.
Tables on the web constitute a valuable data source for many applications, like factual search and knowledge base augmentation. However, as genuine tables containing relational knowledge only account for a small proportion of tables on the web, reliable genuine web table classification is a crucial first step of table extraction. Previous works usually rely on explicit feature construction from the HTML code. In contrast, we propose an approach for web table classification by exploiting the full visual appearance of a table, which works purely by applying a convolutional neural network on the rendered image of the web table. Since these visual features can be extracted automatically, our approach circumvents the need for explicit feature construction. A new hand labeled gold standard dataset containing HTML source code and images for 13,112 tables was generated for this task. Transfer learning techniques are applied to well known VGG16 and ResNet50 architectures. The evaluation of CNN image classification with fine tuned ResNet50 (F1 93.29%) shows that this approach achieves results comparable to previous solutions using explicitly defined HTML code based features. By combining visual and explicit features, an F-measure of 93.70% can be achieved by Random Forest classification, which beats current state of the art methods.
We propose a highly data-efficient active learning framework for image classification. Our novel framework combines: (1) unsupervised representation learning of a Convolutional Neural Network and (2) the Gaussian Process (GP) method, in sequence to achieve highly data and label efficient classifications. Moreover, both elements are less sensitive to the prevalent and challenging class imbalance issue, thanks to the (1) feature learned without labels and (2) the Bayesian nature of GP. The GP-provided uncertainty estimates enable active learning by ranking samples based on the uncertainty and selectively labeling samples showing higher uncertainty. We apply this novel combination to the severely imbalanced case of COVID-19 chest X-ray classification and the Nerthus colonoscopy classification. We demonstrate that only . 10% of the labeled data is needed to reach the accuracy from training all available labels. We also applied our model architecture and proposed framework to a broader class of datasets with expected success.
In recent years, the development of precision agriculture is a new technology. The main reason for the automation of agricultural processes is to save the time and energy required to perform repeated farming tasks and to increase production by treating each crop separately and applying smart agricultural concepts. In this paper, an automatic detection and image recognition of citrus diseases is presented that can help farmer find the disease and identify it from the captured images. This method use YOLO(You Only Look Once) algorithm which is an object detection model to detect and recognize the diseases from citrus leaf images. YOLO can realtime detect the disease and circle around it on the image and video. The dataset includes images of citrus leaf with two kinds of diseases: Citrus Canker, Citrus Greening.
In order to improve the performance of a deep neural network, the activation function is an important aspect that we must research continuously, that is why we have expanded the research in this direction. We introduced a novel P-Swish activation function (Parametric Swish), which is able to bring performance improvements on object classification tasks using datasets such as CIFAR-10, CIFAR-100, but we will see that we also used datasets for Natural Language Processing (NLP). To test it, we used several types of architectures, including LeNet-5, Network in Network (NiN), and ResNet34 compared to popular activation functions such as sigmoid, ReLU, Swish, and our proposals. In particular, the P-Swish function facilitates fast network training, which makes it suitable for the Transfer Learning technique.
In Deep Learning, Convolutional Neural Networks (CNNs) are widely used for Computer Vision applications. With the advent of new technology, there is an inevitable necessity for CNNs to be computationally less expensive. It has become a key factor in determining its competence. CNN models must be compact in size and work efficiently when deployed on embedded systems. In order to achieve this goal, researchers have invented new algorithms which make CNNs lightweight yet accurate enough to be used for applications like object detection. In this paper, we have tried to do the same by modifying an architecture to make it compact with a fair trade-off between model size and accuracy. A new architecture, R-MnasNet (Reduced MnasNet), has been introduced which has a model size of 3 MB. It is trained on CIFAR-10 [4] and has a validation accuracy of 91.13%. Whereas the baseline architecture, MnasNet [1], has a model size of 12.7 MB with a validation accuracy of 80.8% when trained with CIFAR-10 dataset. R-MnasNet can be used on resource-constrained devices. It can be deployed on embedded systems for vision applications.
With regards to Advanced Driver Assistance Systems in vehicles, vision and image-based ADAS is profoundly well known since it utilizes Computer vision algorithms, for example, object detection, street sign identification, vehicle control, impact cautioning, and so on., to aid sheltered and smart driving. Deploying these algorithms directly in resource-constrained devices like mobile and embedded devices etc. is not possible. Reduced Mobilenet V2 (RMNv2) is one of those models which is specifically designed for deploying easily in embedded and mobile devices. In this paper, we implemented a real-time RMNv2 image classifier in NXP Bluebox 2.0 and NXP i.MX RT1060. Because of its low model size of 4.3MB, it is very successful to implement this model in those devices. The model is trained and tested with the CIFAR10 dataset.
The emergence of a convolutional neural network (CNN) has greatly promoted the development of hyperspectral image (HSI) classification technology. However, the acquisition of HSI is difficult. The lack of training samples is the primary cause of low classification performance. The traditional CNN-based methods mainly use the 2-D CNN for feature extraction, which makes the interband correlations of HSIs underutilized. The 3-D CNN extracts the joint spectral-spatial information representation, but it depends on a more complex model. Also, too deep or too shallow network cannot extract the image features well. To tackle these issues, we propose an HSI classification method based on the 2D-3D CNN and multibranch feature fusion. We first combine 2-D CNN and 3-D CNN to extract image features. Then, by means of the multibranch neural network, three kinds of features from shallow to deep are extracted and fused in the spectral dimension. Finally, the fused features are passed into several fully connected layers and a softmax layer to obtain the classification results. In addition, our network model utilizes the state-of-the-art activation function Mish to further improve the classification performance. Our experimental results, conducted on four widely used HSI datasets, indicate that the proposed method achieves better performance than the existing alternatives.