Purpose: Image classification may be the fundamental task in imaging artificial intelligence. We have recently shown that reinforcement learning can achieve high accuracy for lesion localization and segmentation even with minuscule training sets. Here, we introduce reinforcement learning for image classification. In particular, we apply the approach to normal vs. tumor-containing 2D MRI brain images. Materials and Methods: We applied multi-step image classification to allow for combined Deep Q learning and TD(0) Q learning. We trained on a set of 30 images (15 normal and 15 tumor-containing). We tested on a separate set of 30 images (15 normal and 15 tumor-containing). For comparison, we also trained and tested a supervised deep-learning classification network on the same set of training and testing images. Results: Whereas the supervised approach quickly overfit the training data and as expected performed poorly on the testing set (57% accuracy, just over random guessing), the reinforcement learning approach achieved an accuracy of 100%. Conclusion: We have shown a proof-of-principle application of reinforcement learning to the classification of brain tumors. We achieved perfect testing set accuracy with a training set of merely 30 images.
In this paper, we introduce a new perspective on training deep neural networks capable of state-of-the-art performance without the need for the expensive over-parameterization by proposing the concept of In-Time Over-Parameterization (ITOP) in sparse training. By starting from a random sparse network and continuously exploring sparse connectivities during training, we can perform an Over-Parameterization in the space-time manifold, closing the gap in the expressibility between sparse training and dense training. We further use ITOP to understand the underlying mechanism of Dynamic Sparse Training (DST) and indicate that the benefits of DST come from its ability to consider across time all possible parameters when searching for the optimal sparse connectivity. As long as there are sufficient parameters that have been reliably explored during training, DST can outperform the dense neural network by a large margin. We present a series of experiments to support our conjecture and achieve the state-of-the-art sparse training performance with ResNet-50 on ImageNet. More impressively, our method achieves dominant performance over the overparameterization-based sparse methods at extreme sparsity levels. When trained on CIFAR-100, our method can match the performance of the dense model even at an extreme sparsity (98%). Code can be found https://github.com/Shiweiliuiiiiiii/In-Time-Over-Parameterization.
Learning subtle discriminative features plays a significant role in fine-grained image classification. Existing methods usually extract the distinguishable parts through the attention module for classification. Although these learned distinguishable parts contain valuable features that are beneficial for classification, part of irrelevant features are also preserved, which may confuse the model to make a correct classification, especially for the fine-grained tasks due to their similarities. How to keep the discriminative features while removing confusable features from the distinguishable parts is an interesting yet changeling task. In this paper, we introduce a novel classification approach, named Logical-based Feature Extraction Model (LAFE for short) to address this issue. The main advantage of LAFE lies in the fact that it can explicitly add the significance of discriminative features and subtract the confusable features. Specifically, LAFE utilizes the region attention modules and channel attention modules to extract discriminative features and confusable features respectively. Based on this, two novel loss functions are designed to automatically induce attention over these features for fine-grained image classification. Our approach demonstrates its robustness, efficiency, and state-of-the-art performance on three benchmark datasets.
Standard training techniques for neural networks involve multiple sources of randomness, e.g., initialization, mini-batch ordering and in some cases data augmentation. Given that neural networks are heavily over-parameterized in practice, such randomness can cause {\em churn} -- for the same input, disagreements between predictions of the two models independently trained by the same algorithm, contributing to the `reproducibility challenges' in modern machine learning. In this paper, we study this problem of churn, identify factors that cause it, and propose two simple means of mitigating it. We first demonstrate that churn is indeed an issue, even for standard image classification tasks (CIFAR and ImageNet), and study the role of the different sources of training randomness that cause churn. By analyzing the relationship between churn and prediction confidences, we pursue an approach with two components for churn reduction. First, we propose using \emph{minimum entropy regularizers} to increase prediction confidences. Second, \changes{we present a novel variant of co-distillation approach~\citep{anil2018large} to increase model agreement and reduce churn}. We present empirical results showing the effectiveness of both techniques in reducing churn while improving the accuracy of the underlying model.
Deep learning models have become increasingly useful in many different industries. On the domain of image classification, convolutional neural networks proved the ability to learn robust features for the closed set problem, as shown in many different datasets, such as MNIST FASHIONMNIST, CIFAR10, CIFAR100, and IMAGENET. These approaches use deep neural networks with dense layers with softmax activation functions in order to learn features that can separate classes in a latent space. However, this traditional approach is not useful for identifying classes unseen on the training set, known as the open set problem. A similar problem occurs in scenarios involving learning on small data. To tackle both problems, few-shot learning has been proposed. In particular, metric learning learns features that obey constraints of a metric distance in the latent space in order to perform classification. However, while this approach proves to be useful for the open set problem, current implementation requires pair-wise training, where both positive and negative examples of similar images are presented during the training phase, which limits the applicability of these approaches in large data or large class scenarios given the combinatorial nature of the possible inputs.In this paper, we present a constraint-based approach applied to the representations in the latent space under the normalized softmax loss, proposed by[18]. We experimentally validate the proposed approach for the classification of unseen classes on different datasets using both metric learning and the normalized softmax loss, on disjoint and joint scenarios. Our results show that not only our proposed strategy can be efficiently trained on larger set of classes, as it does not require pairwise learning, but also present better classification results than the metric learning strategies surpassing its accuracy by a significant margin.
With the widespread popularity of electronic devices, the emergence of biometric technology has brought significant convenience to user authentication compared with the traditional password and mode unlocking. Among many biological characteristics, the face is a universal and irreplaceable feature with simple detection methods and good recognition accuracy. Face recognition is one of the main functions of electronic equipment propaganda. The previous work in this field mainly focused on converting loss function in traditional deep convolution neural networks without changing the network structure. With the development of AutoML, neural architecture search (NAS) has shown remarkable performance in image classification tasks. In this paper, we first propose a new deep neural architecture search pipeline combined with NAS technology and reinforcement learning strategy into face recognition. We quote the framework of NAS, which trains the child and controller networks alternately. At the same time, we optimize NAS by incorporating evaluation latency into rewards of reinforcement learning and utilize the policy gradient algorithm to search the architecture automatically with the cross-entropy loss. The network architectures we searched out have achieved state-of-the-art accuracy in the large-scale face dataset, which achieved 98.77% top-1 in the MS-Celeb-1M dataset and 99.89% in LFW dataset with relatively small network size.
Convolutional neural networks (CNNs) are able to attain better visual recognition performance than fully connected neural networks despite having much fewer parameters due to their parameter sharing principle. Modern architectures usually contain a small number of fully-connected layers, often at the end, after multiple layers of convolutions. In some cases, most of the convolutions can be eliminated without suffering any loss in recognition performance. However, there is no solid recipe to detect the hidden subset of convolutional neurons that is responsible for the majority of the recognition work. In this work, we formulate this as a pruning problem where the aim is to prune as many kernels as possible while preserving the vanilla generalization performance. To this end, we use the matrix characteristics based on eigenvalues for pruning, in comparison to the average absolute weight of a kernel which is the de facto standard in the literature to assess the importance of an individual convolutional kernel, to shed light on the internal mechanisms of a widely used family of CNNs, namely residual neural networks (ResNets), for the image classification problem using CIFAR-10, CIFAR-100 and Tiny ImageNet datasets.
In this work, we present a novel mask guided attention (MGA) method for fine-grained patchy image classification. The key challenge of fine-grained patchy image classification lies in two folds, ultra-fine-grained inter-category variances among objects and very few data available for training. This motivates us to consider employing more useful supervision signal to train a discriminative model within limited training samples. Specifically, the proposed MGA integrates a pre-trained semantic segmentation model that produces auxiliary supervision signal, i.e., patchy attention mask, enabling a discriminative representation learning. The patchy attention mask drives the classifier to filter out the insignificant parts of images (e.g., common features between different categories), which enhances the robustness of MGA for the fine-grained patchy image classification. We verify the effectiveness of our method on three publicly available patchy image datasets. Experimental results demonstrate that our MGA method achieves superior performance on three datasets compared with the state-of-the-art methods. In addition, our ablation study shows that MGA improves the accuracy by 2.25% and 2% on the SoyCultivarVein and BtfPIS datasets, indicating its practicality towards solving the fine-grained patchy image classification.
Convolutional Neural Networks (CNNs) have proven to be a powerful state-of-the-art method for image classification tasks. One drawback however is the high computational complexity and high memory consumption of CNNs which makes them unfeasible for execution on embedded platforms which are constrained on physical resources needed to support CNNs. Quantization has often been used to efficiently optimize CNNs for memory and computational complexity at the cost of a loss of prediction accuracy. We therefore propose a method to optimally quantize the weights, biases and activations of each layer of a pre-trained CNN while controlling the loss in inference accuracy to enable quantized inference. We quantize the 32-bit floating-point precision parameters to low bitwidth fixed-point representations thereby finding optimal bitwidths and fractional offsets for parameters of each layer of a given CNN. We quantize parameters of a CNN post-training without re-training it. Our method is designed to quantize parameters of a CNN taking into account how other parameters are quantized because ignoring quantization errors due to other quantized parameters leads to a low precision CNN with accuracy losses of up to 50% which is far beyond what is acceptable. Our final method therefore gives a low precision CNN with accuracy losses of less than 1%. As compared to a method used by commercial tools that quantize all parameters to 8-bits, our approach provides quantized CNN with averages of 53% lower memory consumption and 77.5% lower cost of executing multiplications for the two CNNs trained on the four datasets that we tested our work on. We find that layer-wise quantization of parameters significantly helps in this process.
Multi-Instance Learning(MIL) aims to learn the mapping between a bag of instances and the bag-level label. Therefore, the relationships among instances are very important for learning the mapping. In this paper, we propose an MIL algorithm based on a graph built by structural relationship among instances within a bag. Then, Graph Convolutional Network(GCN) and the graph-attention mechanism are used to learn bag-embedding. In the task of medical image classification, our GCN-based MIL algorithm makes full use of the structural relationships among patches(instances) in an original image space domain, and experimental results verify that our method is more suitable for handling medical high-resolution images. We also verify experimentally that the proposed method achieves better results than previous methods on five bechmark MIL datasets and four medical image datasets.
Large Scale image classification is a challenging problem within the field of computer vision. As the real world contains billions of different objects, understanding the performance of popular techniques and models is vital in order to apply them to real world tasks. In this paper, we evaluate techniques and popular CNN based deep learning architectures to perform large scale species classification on the dataset from iNaturalist 2019 Challenge. Methods utilizing dataset pruning and transfer learning are shown to outperform models trained without either of the two techniques. The ResNext based classifier outperforms other model architectures over 10 epochs and achieves a top-one validation error of 0.68 when classifying amongst the 1,010 species.
B-mode ultrasound imaging is a popular medical imaging technique. Like other image processing tasks, deep learning has been used for analysis of B-mode ultrasound images in the last few years. However, training deep learning models requires large labeled datasets, which is often unavailable for ultrasound images. The lack of large labeled data is a bottleneck for the use of deep learning in ultrasound image analysis. To overcome this challenge, in this work we exploit Auxiliary Classifier Generative Adversarial Network (ACGAN) that combines the benefits of data augmentation and transfer learning in the same framework. We conduct experiment on a dataset of breast ultrasound images that shows the effectiveness of the proposed approach.
The ability of a deep learning network to distinguish between in-distribution (ID) and out-of-distribution (OOD) inputs is crucial for ensuring the reliability and trustworthiness of AI systems. Existing OOD detection methods often involve complex architectural innovations, such as ensemble models, which, while enhancing detection accuracy, significantly increase model complexity and training time. Other methods utilize surrogate samples to simulate OOD inputs, but these may not generalize well across different types of OOD data. In this paper, we propose a straightforward yet novel technique to enhance OOD detection in pre-trained networks without altering its original parameters. Our approach defines probabilistic trust intervals for each network weight, determined using in-distribution data. During inference, additional weight values are sampled, and the resulting disagreements among outputs are utilized for OOD detection. We propose a metric to quantify this disagreement and validate its effectiveness with empirical evidence. Our method significantly outperforms various baseline methods across multiple OOD datasets without requiring actual or surrogate OOD samples. We evaluate our approach on MNIST, Fashion-MNIST, CIFAR-10, CIFAR-100 and CIFAR-10-C (a corruption-augmented version of CIFAR-10), across various neural network architectures (e.g., VGG-16, ResNet-20, DenseNet-100). On the MNIST-FashionMNIST setup, our method achieves a False Positive Rate (FPR) of 12.46\% at 95\% True Positive Rate (TPR), compared to 27.09\% achieved by the best baseline. On adversarial and corrupted datasets such as CIFAR-10-C, our proposed method easily differentiate between clean and noisy inputs. These results demonstrate the robustness of our approach in identifying corrupted and adversarial inputs, all without requiring OOD samples during training.
Fine-grained visual classification is a challenging task that recognizes the sub-classes belonging to the same meta-class. Large inter-class similarity and intra-class variance is the main challenge of this task. Most exiting methods try to solve this problem by designing complex model structures to explore more minute and discriminative regions. In this paper, we argue that mining multi-regional multi-grained features is precisely the key to this task. Specifically, we introduce a new loss function, termed top-down spatial attention loss (TDSA-Loss), which contains a multi-stage channel constrained module and a top-down spatial attention module. The multi-stage channel constrained module aims to make the feature channels in different stages category-aligned. Meanwhile, the top-down spatial attention module uses the attention map generated by high-level aligned feature channels to make middle-level aligned feature channels to focus on particular regions. Finally, we can obtain multiple discriminative regions on high-level feature channels and obtain multiple more minute regions within these discriminative regions on middle-level feature channels. In summary, we obtain multi-regional multi-grained features. Experimental results over four widely used fine-grained image classification datasets demonstrate the effectiveness of the proposed method. Ablative studies further show the superiority of two modules in the proposed method. Codes are available at: https://github.com/dongliangchang/Top-Down-Spatial-Attention-Loss.
The beam interruptions (interlocks) of particle accelerators, despite being necessary safety measures, lead to abrupt operational changes and a substantial loss of beam time. A novel time series classification approach is applied to decrease beam time loss in the High Intensity Proton Accelerator complex by forecasting interlock events. The forecasting is performed through binary classification of windows of multivariate time series. The time series are transformed into Recurrence Plots which are then classified by a Convolutional Neural Network, which not only captures the inner structure of the time series but also utilizes the advances of image classification techniques. Our best performing interlock-to-stable classifier reaches an Area under the ROC Curve value of $0.71 \pm 0.01$ compared to $0.65 \pm 0.01$ of a Random Forest model, and it can potentially reduce the beam time loss by $0.5 \pm 0.2$ seconds per interlock.
Fractures occur in the shoulder area, which has a wider range of motion than other joints in the body, for various reasons. To diagnose these fractures, data gathered from Xradiation (X-ray), magnetic resonance imaging (MRI), or computed tomography (CT) are used. This study aims to help physicians by classifying shoulder images taken from X-ray devices as fracture / non-fracture with artificial intelligence. For this purpose, the performances of 26 deep learning-based pretrained models in the detection of shoulder fractures were evaluated on the musculoskeletal radiographs (MURA) dataset, and two ensemble learning models (EL1 and EL2) were developed. The pretrained models used are ResNet, ResNeXt, DenseNet, VGG, Inception, MobileNet, and their spinal fully connected (Spinal FC) versions. In the EL1 and EL2 models developed using pretrained models with the best performance, test accuracy was 0.8455,0.8472, Cohens kappa was 0.6907, 0.6942 and the area that was related with fracture class under the receiver operating characteristic (ROC) curve (AUC) was 0.8862,0.8695. As a result of 28 different classifications in total, the highest test accuracy and Cohens kappa values were obtained in the EL2 model, and the highest AUC value was obtained in the EL1 model.
Recently, unsupervised local learning, based on Hebb's idea that change in synaptic efficacy depends on the activity of the pre- and postsynaptic neuron only, has shown potential as an alternative training mechanism to backpropagation. Unfortunately, Hebbian learning remains experimental and rarely makes it way into standard deep learning frameworks. In this work, we investigate the potential of Hebbian learning in the context of standard deep learning workflows. To this end, a framework for thorough and systematic evaluation of local learning rules in existing deep learning pipelines is proposed. Using this framework, the potential of Hebbian learned feature extractors for image classification is illustrated. In particular, the framework is used to expand the Krotov-Hopfield learning rule to standard convolutional neural networks without sacrificing accuracy compared to end-to-end backpropagation. The source code is available at https://github.com/Joxis/pytorch-hebbian.
Useful information (UI) is an elusive concept in neural networks. A quantitative measurement of UI is absent, despite the variations of UI can be recognized by prior knowledge. The communication bandwidth of feature maps decreases after downscaling operations, but UI flows smoothly after training due to lower Nyquist frequency. Inspired by the low-Nyqusit-frequency nature of UI, we propose the use of spectral roll-off points (SROPs) to estimate UI on variations. The computation of an SROP is extended from a 1-D signal to a 2-D image by the required rotation invariance in image classification tasks. SROP statistics across feature maps are implemented as layer-wise useful information estimates. We design sanity checks to explore SROP variations when UI variations are produced by variations in model input, model architecture and training stages. The variations of SROP is synchronizes with UI variations in various randomized and sufficiently trained model structures. Therefore, SROP variations is an accurate and convenient sign of UI variations, which promotes the explainability of data representations with respect to frequency-domain knowledge.
Accuracy predictor is a key component in Neural Architecture Search (NAS) for ranking architectures. Building a high-quality accuracy predictor usually costs enormous computation. To address this issue, instead of using an accuracy predictor, we propose a novel zero-shot index dubbed Zen-Score to rank the architectures. The Zen-Score represents the network expressivity and positively correlates with the model accuracy. The calculation of Zen-Score only takes a few forward inferences through a randomly initialized network, without training network parameters. Built upon the Zen-Score, we further propose a new NAS algorithm, termed as Zen-NAS, by maximizing the Zen-Score of the target network under given inference budgets. Within less than half GPU day, Zen-NAS is able to directly search high performance architectures in a data-free style. Comparing with previous NAS methods, the proposed Zen-NAS is magnitude times faster on multiple server-side and mobile-side GPU platforms with state-of-the-art accuracy on ImageNet. Our source code and pre-trained models are released on https://github.com/idstcv/ZenNAS.
The aim of few-shot learning (FSL) is to learn how to recognize image categories from a small number of training examples. A central challenge is that the available training examples are normally insufficient to determine which visual features are most characteristic of the considered categories. To address this challenge, we organize these visual features into facets, which intuitively group features of the same kind (e.g. features that are relevant to shape, color, or texture). This is motivated from the assumption that (i) the importance of each facet differs from category to category and (ii) it is possible to predict facet importance from a pre-trained embedding of the category names. In particular, we propose an adaptive similarity measure, relying on predicted facet importance weights for a given set of categories. This measure can be used in combination with a wide array of existing metric-based methods. Experiments on miniImageNet and CUB show that our approach improves the state-of-the-art in metric-based FSL.