Contrastive learning applied to self-supervised representation learning has seen a resurgence in recent years, leading to state of the art performance in the unsupervised training of deep image models. Modern batch contrastive approaches subsume or significantly outperform traditional contrastive losses such as triplet, max-margin and the N-pairs loss. In this work, we extend the self-supervised batch contrastive approach to the fully-supervised setting, allowing us to effectively leverage label information. Clusters of points belonging to the same class are pulled together in embedding space, while simultaneously pushing apart clusters of samples from different classes. We analyze two possible versions of the supervised contrastive (SupCon) loss, identifying the best-performing formulation of the loss. On ResNet-200, we achieve top-1 accuracy of 81.4% on the ImageNet dataset, which is 0.8% above the best number reported for this architecture. We show consistent outperformance over cross-entropy on other datasets and two ResNet variants. The loss shows benefits for robustness to natural corruptions and is more stable to hyperparameter settings such as optimizers and data augmentations. Our loss function is simple to implement, and reference TensorFlow code is released at https://t.ly/supcon.
We present ktrain, a low-code Python library that makes machine learning more accessible and easier to apply. As a wrapper to TensorFlow and many other libraries (e.g., transformers, scikit-learn, stellargraph), it is designed to make sophisticated, state-of-the-art machine learning models simple to build, train, inspect, and apply by both beginners and experienced practitioners. Featuring modules that support text data (e.g., text classification, sequence tagging, open-domain question-answering), vision data (e.g., image classification), graph data (e.g., node classification, link prediction), and tabular data, ktrain presents a simple unified interface enabling one to quickly solve a wide range of tasks in as little as three or four "commands" or lines of code.
Integrated Gradients as an attribution method for deep neural network models offers simple implementability. However, it suffers from noisiness of explanations which affects the ease of interpretability. The SmoothGrad technique is proposed to solve the noisiness issue and smoothen the attribution maps of any gradient-based attribution method. In this paper, we present SmoothTaylor as a novel theoretical concept bridging Integrated Gradients and SmoothGrad, from the Taylor's theorem perspective. We apply the methods to the image classification problem, using the ILSVRC2012 ImageNet object recognition dataset, and a couple of pretrained image models to generate attribution maps. These attribution maps are empirically evaluated using quantitative measures for sensitivity and noise level. We further propose adaptive noising to optimize for the noise scale hyperparameter value. From our experiments, we find that the SmoothTaylor approach together with adaptive noising is able to generate better quality saliency maps with lesser noise and higher sensitivity to the relevant points in the input space as compared to Integrated Gradients.
With the increasing adoption of machine learning tools like neural networks across several domains, interesting connections and comparisons to concepts from other domains are coming to light. In this work, we focus on the class of Tensor Networks, which has been a work horse for physicists in the last two decades to analyse quantum many-body systems. Building on the recent interest in tensor networks for machine learning, we extend the Matrix Product State tensor networks (which can be interpreted as linear classifiers operating in exponentially high dimensional spaces) to be useful in medical image analysis tasks. We focus on classification problems as a first step where we motivate the use of tensor networks and propose adaptions for 2D images using classical image domain concepts such as local orderlessness of images. With the proposed locally orderless tensor network model (LoTeNet), we show that tensor networks are capable of attaining performance that is comparable to state-of-the-art deep learning methods. We evaluate the model on two publicly available medical imaging datasets and show performance improvements with fewer model hyperparameters and lesser computational resources compared to relevant baseline methods.
Deep-learning architectures for classification problems involve the cross-entropy loss sometimes assisted with auxiliary loss functions like center loss, contrastive loss and triplet loss. These auxiliary loss functions facilitate better discrimination between the different classes of interest. However, recent studies hint at the fact that these loss functions do not take into account the intrinsic angular distribution exhibited by the low-level and high-level feature representations. This results in less compactness between samples from the same class and unclear boundary separations between data clusters of different classes. In this paper, we address this issue by proposing the use of geometric constraints, rooted in Riemannian geometry. Specifically, we propose Angular Margin Contrastive Loss (AMC-Loss), a new loss function to be used along with the traditional cross-entropy loss. The AMC-Loss employs the discriminative angular distance metric that is equivalent to geodesic distance on a hypersphere manifold such that it can serve a clear geometric interpretation. We demonstrate the effectiveness of AMC-Loss by providing quantitative and qualitative results. We find that although the proposed geometrically constrained loss-function improves quantitative results modestly, it has a qualitatively surprisingly beneficial effect on increasing the interpretability of deep-net decisions as seen by the visual explanations generated by techniques such as the Grad-CAM. Our code is available at https://github.com/hchoi71/AMC-Loss.
Recent semi-supervised learning methods use pseudo supervision as core idea, especially self-training methods that generate pseudo labels. However, pseudo labels are unreliable. Self-training methods usually rely on single model prediction confidence to filter low-confidence pseudo labels, thus remaining high-confidence errors and wasting many low-confidence correct labels. In this paper, we point out it is difficult for a model to counter its own errors. Instead, leveraging inter-model disagreement between different models is a key to locate pseudo label errors. With this new viewpoint, we propose mutual training between two different models by a dynamically re-weighted loss function, called Dynamic Mutual Training (DMT). We quantify inter-model disagreement by comparing predictions from two different models to dynamically re-weight loss in training, where a larger disagreement indicates a possible error and corresponds to a lower loss value. Extensive experiments show that DMT achieves state-of-the-art performance in both image classification and semantic segmentation. Our codes are released at https://github.com/voldemortX/DST-CBC .
Neural Architecture Search (NAS), i.e., the automation of neural network design, has gained much popularity in recent years with increasingly complex search algorithms being proposed. Yet, solid comparisons with simple baselines are often missing. At the same time, recent retrospective studies have found many new algorithms to be no better than random search (RS). In this work we consider, for the first time, a simple Local Search (LS) algorithm for NAS. We particularly consider a multi-objective NAS formulation, with network accuracy and network complexity as two objectives, as understanding the trade-off between these two objectives is arguably the most interesting aspect of NAS. The proposed LS algorithm is compared with RS and two evolutionary algorithms (EAs), as these are often heralded as being ideal for multi-objective optimization. To promote reproducibility, we create and release two benchmark datasets, named MacroNAS-C10 and MacroNAS-C100, containing 200K saved network evaluations for two established image classification tasks, CIFAR-10 and CIFAR-100. Our benchmarks are designed to be complementary to existing benchmarks, especially in that they are better suited for multi-objective search. We additionally consider a version of the problem with a much larger architecture space. While we find and show that the considered algorithms explore the search space in fundamentally different ways, we also find that LS substantially outperforms RS and even performs nearly as good as state-of-the-art EAs. We believe that this provides strong evidence that LS is truly a competitive baseline for NAS against which new NAS algorithms should be benchmarked.
It is well known that featuremap attention and multi-path representation are important for visual recognition. In this paper, we present a modularized architecture, which applies the channel-wise attention on different network branches to leverage their success in capturing cross-feature interactions and learning diverse representations. Our design results in a simple and unified computation block, which can be parameterized using only a few variables. Our model, named ResNeSt, outperforms EfficientNet in accuracy and latency trade-off on image classification. In addition, ResNeSt has achieved superior transfer learning results on several public benchmarks serving as the backbone, and has been adopted by the winning entries of COCO-LVIS challenge. The source code for complete system and pretrained models are publicly available.
Ground Terrain Recognition is a difficult task as the context information varies significantly over the regions of a ground terrain image. In this paper, we propose a novel approach towards ground-terrain recognition via modeling the Extent-of-Texture information to establish a balance between the order-less texture component and ordered-spatial information locally. At first, the proposed method uses a CNN backbone feature extractor network to capture meaningful information of a ground terrain image, and model the extent of texture and shape information locally. Then, the order-less texture information and ordered shape information are encoded in a patch-wise manner, which is utilized by intra-domain message passing module to make every patch aware of each other for rich feature learning. Next, the Extent-of-Texture (EoT) Guided Inter-domain Message Passing module combines the extent of texture and shape information with the encoded texture and shape information in a patch-wise fashion for sharing knowledge to balance out the order-less texture information with ordered shape information. Further, Bilinear model generates a pairwise correlation between the order-less texture information and ordered shape information. Finally, the ground-terrain image classification is performed by a fully connected layer. The experimental results indicate superior performance of the proposed model over existing state-of-the-art techniques on publicly available datasets like DTD, MINC and GTOS-mobile.
GPUs are a key enabler of the revolution in machine learning and high performance computing, functioning as de facto co-processors to accelerate large-scale computation. As the programming stack and tool support have matured, GPUs have also become accessible to programmers, who may lack detailed knowledge of the underlying architecture and fail to fully leverage the GPU's computation power. GEVO (Gpu optimization using EVOlutionary computation) is a tool for automatically discovering optimization opportunities and tuning the performance of GPU kernels in the LLVM representation. GEVO uses population-based search to find edits to GPU code compiled to LLVM-IR and improves performance on desired criteria while retaining required functionality. We demonstrate that GEVO improves the execution time of the GPU programs in the Rodinia benchmark suite and the machine learning models, SVM and ResNet18, on NVIDIA Tesla P100. For the Rodinia benchmarks, GEVO improves GPU kernel runtime performance by an average of 49.48% and by as much as 412% over the fully compiler-optimized baseline. If kernel output accuracy is relaxed to tolerate up to 1% error, GEVO can find kernel variants that outperform the baseline version by an average of 51.08%. For the machine learning workloads, GEVO achieves kernel performance improvement for SVM on the MNIST handwriting recognition (3.24X) and the a9a income prediction (2.93X) datasets with no loss of model accuracy. GEVO achieves 1.79X kernel performance improvement on image classification using ResNet18/CIFAR-10, with less than 1% model accuracy reduction.
A deep convolutional neural network (CNN) has been widely used in image classification and gives better classification accuracy than the other techniques. The softmax cross-entropy loss function is often used for classification tasks. There are some works to introduce the additional terms in the objective function for training to make the features of the output layer more discriminative. The neuron-wise discriminant criterion makes the input feature of each neuron in the output layer discriminative by introducing the discriminant criterion to each of the features. Similarly, the center loss was introduced to the features before the softmax activation function for face recognition to make the deep features discriminative. The ReLU function is often used for the network as an active function in the hidden layers of the CNN. However, it is observed that the deep features trained by using the ReLU function are not discriminative enough and show elongated shapes. In this paper, we propose to use the neuron-wise discriminant criterion at the output layer and the center-loss at the hidden layer. Also, we introduce the online computation of the means of each class with the exponential forgetting. We named them adaptive neuron-wise discriminant criterion and adaptive center loss, respectively. The effectiveness of the integration of the adaptive neuron-wise discriminant criterion and the adaptive center loss is shown by the experiments with MNSIT, FashionMNIST, CIFAR10, CIFAR100, and STL10. Source code is at https://github.com/i13abe/Adaptive-discriminant-and-center
We present DarkneTZ, a framework that uses an edge device's Trusted Execution Environment (TEE) in conjunction with model partitioning to limit the attack surface against Deep Neural Networks (DNNs). Increasingly, edge devices (smartphones and consumer IoT devices) are equipped with pre-trained DNNs for a variety of applications. This trend comes with privacy risks as models can leak information about their training data through effective membership inference attacks (MIAs). We evaluate the performance of DarkneTZ, including CPU execution time, memory usage, and accurate power consumption, using two small and six large image classification models. Due to the limited memory of the edge device's TEE, we partition model layers into more sensitive layers (to be executed inside the device TEE), and a set of layers to be executed in the untrusted part of the operating system. Our results show that even if a single layer is hidden, we can provide reliable model privacy and defend against state of the art MIAs, with only 3% performance overhead. When fully utilizing the TEE, DarkneTZ provides model protections with up to 10% overhead.
In the past decades the set of human tasks that are solved by machines was extended dramatically. From simple image classification problems researchers now move towards solving more sophisticated and vital problems, like, autonomous driving and language translation. The case of language translation includes a challenging area of sign language translation that incorporates both image and language processing. We make a step in that direction by proposing a lightweight network for ASL gesture recognition with a performance sufficient for practical applications. The proposed solution demonstrates impressive robustness on MS-ASL dataset and in live mode for continuous sign gesture recognition scenario. Additionally, we describe how to combine action recognition model training with metric-learning to train the network on the database of limited size. The training code is available as part of Intel OpenVINO Training Extensions.
Residual networks (ResNets) represent a powerful type of convolutional neural network (CNN) architecture, widely adopted and used in various tasks. In this work we propose an improved version of ResNets. Our proposed improvements address all three main components of a ResNet: the flow of information through the network layers, the residual building block, and the projection shortcut. We are able to show consistent improvements in accuracy and learning convergence over the baseline. For instance, on ImageNet dataset, using the ResNet with 50 layers, for top-1 accuracy we can report a 1.19% improvement over the baseline in one setting and around 2% boost in another. Importantly, these improvements are obtained without increasing the model complexity. Our proposed approach allows us to train extremely deep networks, while the baseline shows severe optimization issues. We report results on three tasks over six datasets: image classification (ImageNet, CIFAR-10 and CIFAR-100), object detection (COCO) and video action recognition (Kinetics-400 and Something-Something-v2). In the deep learning era, we establish a new milestone for the depth of a CNN. We successfully train a 404-layer deep CNN on the ImageNet dataset and a 3002-layer network on CIFAR-10 and CIFAR-100, while the baseline is not able to converge at such extreme depths. Code is available at: https://github.com/iduta/iresnet
This paper presents X3D, a family of efficient video networks that progressively expand a tiny 2D image classification architecture along multiple network axes, in space, time, width and depth. Inspired by feature selection methods in machine learning, a simple stepwise network expansion approach is employed that expands a single axis in each step, such that good accuracy to complexity trade-off is achieved. To expand X3D to a specific target complexity, we perform progressive forward expansion followed by backward contraction. X3D achieves state-of-the-art performance while requiring 4.8x and 5.5x fewer multiply-adds and parameters for similar accuracy as previous work. Our most surprising finding is that networks with high spatiotemporal resolution can perform well, while being extremely light in terms of network width and parameters. We report competitive accuracy at unprecedented efficiency on video classification and detection benchmarks. Code will be available at: https://github.com/facebookresearch/SlowFast
In this paper, we propose a dimension reduction method specifically designed for tensor-structured feature data in deep neural networks. The method is implemented as a hidden layer, called the TensorProjection layer, which transforms input tensors into output tensors with reduced dimensions through mode-wise projections. The projection directions are treated as model parameters of the layer and are optimized during model training. Our method can serve as an alternative to pooling layers for summarizing image data, or to convolutional layers as a technique for reducing the number of channels. We conduct experiments on tasks such as medical image classification and segmentation, integrating the TensorProjection layer into commonly used baseline architectures to evaluate its effectiveness. Numerical experiments indicate that the proposed method can outperform traditional downsampling methods, such as pooling layers, in our tasks, suggesting it as a promising alternative for feature summarization.
We present a new supervised image classification method applicable to a broad class of image deformation models. The method makes use of the previously described Radon Cumulative Distribution Transform (R-CDT) for image data, whose mathematical properties are exploited to express the image data in a form that is more suitable for machine learning. While certain operations such as translation, scaling, and higher-order transformations are challenging to model in native image space, we show the R-CDT can capture some of these variations and thus render the associated image classification problems easier to solve. The method -- utilizing a nearest-subspace algorithm in R-CDT space -- is simple to implement, non-iterative, has no hyper-parameters to tune, is computationally efficient, label efficient, and provides competitive accuracies to state-of-the-art neural networks for many types of classification problems. In addition to the test accuracy performances, we show improvements (with respect to neural network-based methods) in terms of computational efficiency (it can be implemented without the use of GPUs), number of training samples needed for training, as well as out-of-distribution generalization. The Python code for reproducing our results is available at https://github.com/rohdelab/rcdt_ns_classifier.
The binary neural network, largely saving the storage and computation, serves as a promising technique for deploying deep models on resource-limited devices. However, the binarization inevitably causes severe information loss, and even worse, its discontinuity brings difficulty to the optimization of the deep network. To address these issues, a variety of algorithms have been proposed, and achieved satisfying progress in recent years. In this paper, we present a comprehensive survey of these algorithms, mainly categorized into the native solutions directly conducting binarization, and the optimized ones using techniques like minimizing the quantization error, improving the network loss function, and reducing the gradient error. We also investigate other practical aspects of binary neural networks such as the hardware-friendly design and the training tricks. Then, we give the evaluation and discussions on different tasks, including image classification, object detection and semantic segmentation. Finally, the challenges that may be faced in future research are prospected.
To reduce the overwhelming size of Deep Neural Networks (DNN) teacher-student methodology tries to transfer knowledge from a complex teacher network to a simple student network. We instead propose a novel method called the teacher-class network consisting of a single teacher and multiple student networks (i.e. class of students). Instead of transferring knowledge to one student only, the proposed method transfers a chunk of knowledge to each student. Our students are not trained for problem-specific logits, they are trained to mimic knowledge (dense representation) learned by the teacher network thus the combined knowledge learned by the class of students can be used to solve other problems as well. The proposed teacher-class architecture is evaluated on several benchmark datasets such as MNIST, Fashion MNIST, IMDB Movie Reviews, CAMVid, CIFAR-10 and ImageNet on multiple tasks including image classification, sentiment classification and segmentation. Our approach outperforms the state of-the-art single student approach in terms of accuracy as well as computational cost while achieving 10-30 times reduction in parameters.
Non-Local (NL) blocks have been widely studied in various vision tasks. However, it has been rarely explored to embed the NL blocks in mobile neural networks, mainly due to the following challenges: 1) NL blocks generally have heavy computation cost which makes it difficult to be applied in applications where computational resources are limited, and 2) it is an open problem to discover an optimal configuration to embed NL blocks into mobile neural networks. We propose AutoNL to overcome the above two obstacles. Firstly, we propose a Lightweight Non-Local (LightNL) block by squeezing the transformation operations and incorporating compact features. With the novel design choices, the proposed LightNL block is 400x computationally cheaper} than its conventional counterpart without sacrificing the performance. Secondly, by relaxing the structure of the LightNL block to be differentiable during training, we propose an efficient neural architecture search algorithm to learn an optimal configuration of LightNL blocks in an end-to-end manner. Notably, using only 32 GPU hours, the searched AutoNL model achieves 77.7% top-1 accuracy on ImageNet under a typical mobile setting (350M FLOPs), significantly outperforming previous mobile models including MobileNetV2 (+5.7%), FBNet (+2.8%) and MnasNet (+2.1%). Code and models are available at https://github.com/LiYingwei/AutoNL.