We present Meta Pseudo Labels, a semi-supervised learning method that achieves a new state-of-the-art top-1 accuracy of 90.2% on ImageNet, which is 1.6% better than the existing state-of-the-art. Like Pseudo Labels, Meta Pseudo Labels has a teacher network to generate pseudo labels on unlabeled data to teach a student network. However, unlike Pseudo Labels where the teacher is fixed, the teacher in Meta Pseudo Labels is constantly adapted by the feedback of the student's performance on the labeled dataset. As a result, the teacher generates better pseudo labels to teach the student. Our code will be available at https://github.com/google-research/google-research/tree/master/meta_pseudo_labels.
Surface damage on concrete is important as the damage can affect the structural integrity of the structure. This paper proposes a two-step surface damage detection scheme using Convolutional Neural Network (CNN) and Artificial Neural Network (ANN). The CNN classifies given input images into two categories: positive and negative. The positive category is where the surface damage is present within the image, otherwise the image is classified as negative. This is an image-based classification. The ANN accepts image inputs that have been classified as positive by the ANN. This reduces the number of images that are further processed by the ANN. The ANN performs feature-based classification, in which the features are extracted from the detected edges within the image. The edges are detected using Canny edge detection. A total of 19 features are extracted from the detected edges. These features are inputs into the ANN. The purpose of the ANN is to highlight only the positive damaged edges within the image. The CNN achieves an accuracy of 80.7% for image classification and the ANN achieves an accuracy of 98.1% for surface detection. The decreased accuracy in the CNN is due to the false positive detection, however false positives are tolerated whereas false negatives are not. The false negative detection for both CNN and ANN in the two-step scheme are 0%.
Continual learning aims to learn new tasks without forgetting previously learned ones. We hypothesize that representations learned to solve each task in a sequence have a shared structure while containing some task-specific properties. We show that shared features are significantly less prone to forgetting and propose a novel hybrid continual learning framework that learns a disjoint representation for task-invariant and task-specific features required to solve a sequence of tasks. Our model combines architecture growth to prevent forgetting of task-specific skills and an experience replay approach to preserve shared skills. We demonstrate our hybrid approach is effective in avoiding forgetting and show it is superior to both architecture-based and memory-based approaches on class incrementally learning of a single dataset as well as a sequence of multiple datasets in image classification. Our code is available at \url{https://github.com/facebookresearch/Adversarial-Continual-Learning}.
Convolutional Neural Network (CNN) has become the most used method for image classification tasks. During its training the learning rate and the gradient are two key factors to tune for influencing the convergence speed of the model. Usual learning rate strategies are time-based i.e. monotonous decay over time. Recent state-of-the-art techniques focus on adaptive gradient algorithms i.e. Adam and its versions. In this paper we consider an online learning scenario and we propose two Event-Based control loops to adjust the learning rate of a classical algorithm E (Exponential)/PD (Proportional Derivative)-Control. The first Event-Based control loop will be implemented to prevent sudden drop of the learning rate when the model is approaching the optimum. The second Event-Based control loop will decide, based on the learning speed, when to switch to the next data batch. Experimental evaluationis provided using two state-of-the-art machine learning image datasets (CIFAR-10 and CIFAR-100). Results show the Event-Based E/PD is better than the original algorithm (higher final accuracy, lower final loss value), and the Double-Event-BasedE/PD can accelerate the training process, save up to 67% training time compared to state-of-the-art algorithms and even result in better performance.
Neural networks have led to major improvements in image classification but suffer from being non-robust to adversarial changes, unreliable uncertainty estimates on out-distribution samples and their inscrutable black-box decisions. In this work we propose RATIO, a training procedure for Robustness via Adversarial Training on In- and Out-distribution, which leads to robust models with reliable and robust confidence estimates on the out-distribution. RATIO has similar generative properties to adversarial training so that visual counterfactuals produce class specific features. While adversarial training comes at the price of lower clean accuracy, RATIO achieves state-of-the-art $l_2$-adversarial robustness on CIFAR10 and maintains better clean accuracy.
Lightweight or mobile neural networks used for real-time computer vision tasks contain fewer parameters than normal networks, which lead to a constrained performance. In this work, we proposed a novel activation function named Tanh Exponential Activation Function (TanhExp) which can improve the performance for these networks on image classification task significantly. The definition of TanhExp is f(x) = xtanh(e^x). We demonstrate the simplicity, efficiency, and robustness of TanhExp on various datasets and network models and TanhExp outperforms its counterparts in both convergence speed and accuracy. Its behaviour also remains stable even with noise added and dataset altered. We show that without increasing the size of the network, the capacity of lightweight neural networks can be enhanced by TanhExp with only a few training epochs and no extra parameters added.
While the self-attention mechanism has been widely used in a wide variety of tasks, it has the unfortunate property of a quadratic cost with respect to the input length, which makes it difficult to deal with long inputs. In this paper, we present a method for accelerating and structuring self-attentions: Sparse Adaptive Connection (SAC). In SAC, we regard the input sequence as a graph and attention operations are performed between linked nodes. In contrast with previous self-attention models with pre-defined structures (edges), the model learns to construct attention edges to improve task-specific performances. In this way, the model is able to select the most salient nodes and reduce the quadratic complexity regardless of the sequence length. Based on SAC, we show that previous variants of self-attention models are its special cases. Through extensive experiments on neural machine translation, language modeling, graph representation learning and image classification, we demonstrate SAC is competitive with state-of-the-art models while significantly reducing memory cost.
Nowadays it is prevalent to take features extracted from pre-trained deep learning models as image representations which have achieved promising classification performance. Existing methods usually consider either object-based features or scene-based features only. However, both types of features are important for complex images like scene images, as they can complement each other. In this paper, we propose a novel type of features -- hybrid deep features, for scene images. Specifically, we exploit both object-based and scene-based features at two levels: part image level (i.e., parts of an image) and whole image level (i.e., a whole image), which produces a total number of four types of deep features. Regarding the part image level, we also propose two new slicing techniques to extract part based features. Finally, we aggregate these four types of deep features via the concatenation operator. We demonstrate the effectiveness of our hybrid deep features on three commonly used scene datasets (MIT-67, Scene-15, and Event-8), in terms of the scene image classification task. Extensive comparisons show that our introduced features can produce state-of-the-art classification accuracies which are more consistent and stable than the results of existing features across all datasets.
Federated learning (FL) is a novel machine learning setting that enables on-device intelligence via decentralized training and federated optimization. Deep neural networks' rapid development facilitates the learning techniques for modeling complex problems and emerges into federated deep learning under the federated setting. However, the tremendous amount of model parameters burdens the communication network with a high load of transportation. This paper introduces two approaches for improving communication efficiency by dynamic sampling and top-$k$ selective masking. The former controls the fraction of selected client models dynamically, while the latter selects parameters with top-$k$ largest values of difference for federated updating. Experiments on convolutional image classification and recurrent language modeling are conducted on three public datasets to show our proposed methods' effectiveness.
Grasp verification is advantageous for autonomous manipulation robots as they provide the feedback required for higher level planning components about successful task completion. However, a major obstacle in doing grasp verification is sensor selection. In this paper, we propose a vision based grasp verification system using machine vision cameras, with the verification problem formulated as an image classification task. Machine vision cameras consist of a camera and a processing unit capable of on-board deep learning inference. The inference in these low-power hardware are done near the data source, reducing the robot's dependence on a centralized server, leading to reduced latency, and improved reliability. Machine vision cameras provide the deep learning inference capabilities using different neural accelerators. Although, it is not clear from the documentation of these cameras what is the effect of these neural accelerators on performance metrics such as latency and throughput. To systematically benchmark these machine vision cameras, we propose a parameterized model generator that generates end to end models of Convolutional Neural Networks(CNN). Using these generated models we benchmark latency and throughput of two machine vision cameras, JeVois A33 and Sipeed Maix Bit. Our experiments demonstrate that the selected machine vision camera and the deep learning models can robustly verify grasp with 97% per frame accuracy.
Affinity graphs are widely used in deep architectures, including graph convolutional neural networks and attention networks. Thus far, the literature has focused on abstracting features from such graphs, while the learning of the affinities themselves has been overlooked. Here we propose a principled method to directly supervise the learning of weights in affinity graphs, to exploit meaningful connections between entities in the data source. Applied to a visual attention network, our affinity supervision improves relationship recovery between objects, even without the use of manually annotated relationship labels. We further show that affinity learning between objects boosts scene categorization performance and that the supervision of affinity can also be applied to graphs built from mini-batches, for neural network training. In an image classification task we demonstrate consistent improvement over the baseline, with diverse network architectures and datasets.
We propose a scheme for supervised image classification that uses privileged information, in the form of keypoint annotations for the training data, to learn strong models from small and/or biased training sets. Our main motivation is the recognition of animal species for ecological applications such as biodiversity modelling, which is challenging because of long-tailed species distributions due to rare species, and strong dataset biases such as repetitive scene background in camera traps. To counteract these challenges, we propose a visual attention mechanism that is supervised via keypoint annotations that highlight important object parts. This privileged information, implemented as a novel privileged pooling operation, is only required during training and helps the model to focus on regions that are discriminative. In experiments with three different animal species datasets, we show that deep networks with privileged pooling can use small training sets more efficiently and generalize better.
Deep learning shows great potential for the domain of digital pathology. An automated digital pathology system could serve as a second reader, perform initial triage in large screening studies, or assist in reporting. However, it is expensive to exhaustively annotate large histology image databases, since medical specialists are a scarce resource. In this paper, we apply the semi-supervised teacher-student knowledge distillation technique proposed by Yalniz et al. (2019) to the task of quantifying prognostic features in colorectal cancer. We obtain accuracy improvements through extending this approach to a chain of students, where each student's predictions are used to train the next student i.e. the student becomes the teacher. Using the chain approach, and only 0.5% labelled data (the remaining 99.5% in the unlabelled pool), we match the accuracy of training on 100% labelled data. At lower percentages of labelled data, similar gains in accuracy are seen, allowing some recovery of accuracy even from a poor initial choice of labelled training set. In conclusion, this approach shows promise for reducing the annotation burden, thus increasing the affordability of automated digital pathology systems.
Convolutional Neural Networks (CNNs) have shown remarkable performance in general object recognition tasks. In this paper, we propose a new model called EnsNet which is composed of one base CNN and multiple Fully Connected SubNetworks (FCSNs). In this model, the set of feature-maps generated by the last convolutional layer in the base CNN is divided along channels into disjoint subsets, and these subsets are assigned to the FCSNs. Each of the FCSNs is trained independent of others so that it can predict the class label from the subset of the feature-maps assigned to it. The output of the overall model is determined by majority vote of the base CNN and the FCSNs. Experimental results using the MNIST, Fashion-MNIST and CIFAR-10 datasets show that the proposed approach further improves the performance of CNNs. In particular, an EnsNet achieves a state-of-the-art error rate of 0.16% on MNIST.
Image classifiers are typically scored on their test set accuracy, but high accuracy can mask a subtle type of model failure. We find that high scoring convolutional neural networks (CNNs) on popular benchmarks exhibit troubling pathologies that allow them to display high accuracy even in the absence of semantically salient features. When a model provides a high-confidence decision without salient supporting input features, we say the classifier has overinterpreted its input, finding too much class-evidence in patterns that appear nonsensical to humans. Here, we demonstrate that neural networks trained on CIFAR-10 and ImageNet suffer from overinterpretation, and we find models on CIFAR-10 make confident predictions even when 95% of input images are masked and humans cannot discern salient features in the remaining pixel-subsets. We introduce Batched Gradient SIS, a new method for discovering sufficient input subsets for complex datasets, and use this method to show the sufficiency of border pixels in ImageNet for training and testing. Although these patterns portend potential model fragility in real-world deployment, they are in fact valid statistical patterns of the benchmark that alone suffice to attain high test accuracy. Unlike adversarial examples, overinterpretation relies upon unmodified image pixels. We find ensembling and input dropout can each help mitigate overinterpretation.
Accelerating deep convolutional neural networks has become an active topic and sparked an interest in academia and industry. In this paper, we propose an efficient low-precision quantized Winograd convolution algorithm, called LANCE, which combines the advantages of fast convolution and quantization techniques. By embedding linear quantization operations into the Winograd-domain, the fast convolution can be performed efficiently under low-precision computation on graphics processing units. We test neural network models with LANCE on representative image classification datasets, including SVHN, CIFAR, and ImageNet. The experimental results show that our 8-bit quantized Winograd convolution improves the performance by up to 2.40x over the full-precision convolution with trivial accuracy loss.
Classifying whole images is a classic problem in machine learning, and graph neural networks are a powerful methodology to learn highly irregular geometries. It is often the case that certain parts of a point cloud are more important than others when determining overall classification. On graph structures this started by pooling information at the end of convolutional filters, and has evolved to a variety of staged pooling techniques on static graphs. In this paper, a dynamic graph formulation of pooling is introduced that removes the need for predetermined graph structure. It achieves this by dynamically learning the most important relationships between data via an intermediate clustering. The network architecture yields interesting results considering representation size and efficiency. It also adapts easily to a large number of tasks from image classification to energy regression in high energy particle physics.
Automated medical image classification with convolutional neural networks (CNNs) has great potential to impact healthcare, particularly in resource-constrained healthcare systems where fewer trained radiologists are available. However, little is known about how well a trained CNN can perform on images with the increased noise levels, different acquisition protocols, or additional artifacts that may arise when using low-cost scanners, which can be underrepresented in datasets collected from well-funded hospitals. In this work, we investigate how a model trained to triage head computed tomography (CT) scans performs on images acquired with reduced x-ray tube current, fewer projections per gantry rotation, and limited angle scans. These changes can reduce the cost of the scanner and demands on electrical power but come at the expense of increased image noise and artifacts. We first develop a model to triage head CTs and report an area under the receiver operating characteristic curve (AUROC) of 0.77. We then show that the trained model is robust to reduced tube current and fewer projections, with the AUROC dropping only 0.65% for images acquired with a 16x reduction in tube current and 0.22% for images acquired with 8x fewer projections. Finally, for significantly degraded images acquired by a limited angle scan, we show that a model trained specifically to classify such images can overcome the technological limitations to reconstruction and maintain an AUROC within 0.09% of the original model.
This paper provides an extensive analysis of the performance of the EfficientNet image classifiers with several recent training procedures, in particular one that corrects the discrepancy between train and test images. The resulting network, called FixEfficientNet, significantly outperforms the initial architecture with the same number of parameters. For instance, our FixEfficientNet-B0 trained without additional training data achieves 79.3% top-1 accuracy on ImageNet with 5.3M parameters. This is a +0.5% absolute improvement over the Noisy student EfficientNet-B0 trained with 300M unlabeled images. An EfficientNet-L2 pre-trained with weak supervision on 300M unlabeled images and further optimized with FixRes achieves 88.5% top-1 accuracy (top-5: 98.7%), which establishes the new state of the art for ImageNet with a single crop. These improvements are thoroughly evaluated with cleaner protocols than the one usually employed for Imagenet, and particular we show that our improvement remains in the experimental setting of ImageNet-v2, that is less prone to overfitting, and with ImageNet Real Labels. In both cases we also establish the new state of the art.
Plant diseases serve as one of main threats to food security and crop production. It is thus valuable to exploit recent advances of artificial intelligence to assist plant disease diagnosis. One popular approach is to transform this problem as a leaf image classification task, which can be then addressed by the powerful convolutional neural networks (CNNs). However, the performance of CNN-based classification approach depends on a large amount of high-quality manually labeled training data, which are inevitably introduced noise on labels in practice, leading to model overfitting and performance degradation. To overcome this problem, we propose a novel framework that incorporates rectified meta-learning module into common CNN paradigm to train a noise-robust deep network without using extra supervision information. The proposed method enjoys the following merits: i) A rectified meta-learning is designed to pay more attention to unbiased samples, leading to accelerated convergence and improved classification accuracy. ii) Our method is free on assumption of label noise distribution, which works well on various kinds of noise. iii) Our method serves as a plug-and-play module, which can be embedded into any deep models optimized by gradient descent based method. Extensive experiments are conducted to demonstrate the superior performance of our algorithm over the state-of-the-arts.