Recently, Barbu et al introduced a dataset called ObjectNet which includes objects in daily life situations. They showed a dramatic performance drop of the state of the art object recognition models on this dataset. Due to the importance and implications of their results regarding generalization ability of deep models, we take a second look at their findings. We highlight a major problem with their work which is applying object recognizers to the scenes containing multiple objects rather than isolated objects. The latter results in around 20-30% performance gain using our code. Compared with the results reported in the ObjectNet paper, we observe that around 10-15 % of the performance loss can be recovered, without any test time data augmentation. In accordance with Barbu et al.'s conclusions, however, we also conclude that deep models suffer drastically on this dataset. Thus, we believe that ObjectNet remains a challenging dataset for testing the generalization power of models beyond datasets on which they have been trained.
Collecting fine-grained labels usually requires expert-level domain knowledge and is prohibitive to scale up. In this paper, we propose Attribute Mix, a data augmentation strategy at attribute level to expand the fine-grained samples. The principle lies in that attribute features are shared among fine-grained sub-categories, and can be seamlessly transferred among images. Toward this goal, we propose an automatic attribute mining approach to discover attributes that belong to the same super-category, and Attribute Mix is operated by mixing semantically meaningful attribute features from two images. Attribute Mix is a simple but effective data augmentation strategy that can significantly improve the recognition performance without increasing the inference budgets. Furthermore, since attributes can be shared among images from the same super-category, we further enrich the training samples with attribute level labels using images from the generic domain. Experiments on widely used fine-grained benchmarks demonstrate the effectiveness of our proposed method.
We propose a scalable photonic architecture for implementation of feedforward and recurrent neural networks to perform the classification of handwritten digits from the MNIST database. Our experiment exploits off-the-shelf optical and electronic components to currently achieve a network size of 16,384 nodes. Both network types are designed within the the reservoir computing paradigm with randomly weighted input and hidden layers. Using various feature extraction techniques (e.g. histograms of oriented gradients, zoning, Gabor filters) and a simple training procedure consisting of linear regression and winner-takes-all decision strategy, we demonstrate numerically and experimentally that a feedforward network allows for classification error rate of 1%, which is at the state-of-the-art for experimental implementations and remains competitive with more advanced algorithmic approaches. We also investigate recurrent networks in numerical simulations by explicitly activating the temporal dynamics, and predict a performance improvement over the feedforward configuration.
Automatic designing computationally efficient neural networks has received much attention in recent years. Existing approaches either utilize network pruning or leverage the network architecture search methods. This paper presents a new framework named network adjustment, which considers network accuracy as a function of FLOPs, so that under each network configuration, one can estimate the FLOPs utilization ratio (FUR) for each layer and use it to determine whether to increase or decrease the number of channels on the layer. Note that FUR, like the gradient of a non-linear function, is accurate only in a small neighborhood of the current network. Hence, we design an iterative mechanism so that the initial network undergoes a number of steps, each of which has a small `adjusting rate' to control the changes to the network. The computational overhead of the entire search process is reasonable, i.e., comparable to that of re-training the final model from scratch. Experiments on standard image classification datasets and a wide range of base networks demonstrate the effectiveness of our approach, which consistently outperforms the pruning counterpart. The code is available at https://github.com/danczs/NetworkAdjustment.
A recent line of works studied wide deep neural networks (DNNs) by approximating them as Gaussian Processes (GPs). A DNN trained with gradient flow was shown to map to a GP governed by the Neural Tangent Kernel (NTK), whereas earlier works showed that a DNN with an i.i.d. prior over its weights maps to the so-called Neural Network Gaussian Process (NNGP). Here we consider a DNN training protocol, involving noise, weight decay and finite width, whose outcome corresponds to a certain non-Gaussian stochastic process. An analytical framework is then introduced to analyze this non-Gaussian process, whose deviation from a GP is controlled by the finite width. Our contribution is three-fold: (i) In the infinite width limit, we establish a correspondence between DNNs trained with noisy gradients and the NNGP, not the NTK. (ii) We provide a general analytical form for the finite width correction (FWC) for DNNs with arbitrary activation functions and depth and use it to predict the outputs of empirical finite networks with high accuracy. Analyzing the FWC behavior as a function of $n$, the training set size, we find that it is negligible for both the very small $n$ regime, and, surprisingly, for the large $n$ regime (where the GP error scales as $O(1/n)$). (iii) We flesh out algebraically how these FWCs can improve the performance of finite convolutional neural networks (CNNs) relative to their GP counterparts on image classification tasks.
Optimization techniques are of great importance to effectively and efficiently train a deep neural network (DNN). It has been shown that using the first and second order statistics (e.g., mean and variance) to perform Z-score standardization on network activations or weight vectors, such as batch normalization (BN) and weight standardization (WS), can improve the training performance. Different from these existing methods that mostly operate on activations or weights, we present a new optimization technique, namely gradient centralization (GC), which operates directly on gradients by centralizing the gradient vectors to have zero mean. GC can be viewed as a projected gradient descent method with a constrained loss function. We show that GC can regularize both the weight space and output feature space so that it can boost the generalization performance of DNNs. Moreover, GC improves the Lipschitzness of the loss function and its gradient so that the training process becomes more efficient and stable. GC is very simple to implement and can be easily embedded into existing gradient based DNN optimizers with only one line of code. It can also be directly used to fine-tune the pre-trained DNNs. Our experiments on various applications, including general image classification, fine-grained image classification, detection and segmentation, demonstrate that GC can consistently improve the performance of DNN learning. The code of GC can be found at https://github.com/Yonghongwei/Gradient-Centralization.
We investigate how different active learning (AL) query policies coupled with classification uncertainty visualizations affect analyst trust in automated classification systems. A current standard policy for AL is to query the oracle (e.g., the analyst) to refine labels for datapoints where the classifier has the highest uncertainty. This is an optimal policy for the automation system as it yields maximal information gain. However, model-centric policies neglect the effects of this uncertainty on the human component of the system and the consequent manner in which the human will interact with the system post-training. In this paper, we present an empirical study evaluating how AL query policies and visualizations lending transparency to classification influence trust in automated classification of image data. We found that query policy significantly influences an analyst's trust in an image classification system, and we use these results to propose a set of oracle query policies and visualizations for use during AL training phases that can influence analyst trust in classification.
We have recently seen great progress in image classification due to the success of deep convolutional neural networks and the availability of large-scale datasets. Most of the existing work focuses on single-label image classification. However, there are usually multiple tags associated with an image. The existing works on multi-label classification are mainly based on lab curated labels. Humans assign tags to their images differently, which is mainly based on their interests and personal tagging behavior. In this paper, we address the problem of personalized tag recommendation and propose an end-to-end deep network which can be trained on large-scale datasets. The user-preference is learned within the network in an unsupervised way where the network performs joint optimization for user-preference and visual encoding. A joint training of user-preference and visual encoding allows the network to efficiently integrate the visual preference with tagging behavior for a better user recommendation. In addition, we propose the use of adversarial learning, which enforces the network to predict tags resembling user-generated tags. We demonstrate the effectiveness of the proposed model on two different large-scale and publicly available datasets, YFCC100M and NUS-WIDE. The proposed method achieves significantly better performance on both the datasets when compared to the baselines and other state-of-the-art methods. The code is publicly available at https://github.com/vyzuer/ALTReco.
Deep learning techniques are currently the state of the art approach to deal with image classification problems. Nevertheless, non-expert users might find challenging the use of these techniques due to several reasons, including the lack of enough images, the necessity of trying different models and conducting a thorough comparison of the results obtained with them, and the technical difficulties of employing different libraries, tools and special purpose hardware like GPUs. In this work, we present FrImCla, an open-source and free tool that simplifies the construction of robust models for image classification from a dataset of images, and only using the computer CPU. Given a dataset of annotated images, FrImCla automatically constructs a classification model (both for single-label and multi-label classification problems) by trying several feature extractors (based both on transfer learning and traditional computer vision methods) and machine learning algorithms, and selecting the best combination after a thorough statistical analysis. Thus, this tool can be employed by non-expert users to create accurate models from small datasets of images without requiring any special purpose hardware. In addition, in this paper, we show that FrImCla can be employed to construct accurate models that are close, or even better, to the state-of-the-art models.
Deep neural networks (DNN) have shown remarkable success in a variety of machine learning applications. The capacity of these models (i.e., number of parameters), endows them with expressive power and allows them to reach the desired performance. In recent years, there is an increasing interest in deploying DNNs to resource-constrained devices (i.e., mobile devices) with limited energy, memory, and computational budget. To address this problem, we propose Entropy-Constrained Trained Ternarization (EC2T), a general framework to create sparse and ternary neural networks which are efficient in terms of storage (e.g., at most two binary-masks and two full-precision values are required to save a weight matrix) and computation (e.g., MAC operations are reduced to a few accumulations plus two multiplications). This approach consists of two steps. First, a super-network is created by scaling the dimensions of a pre-trained model (i.e., its width and depth). Subsequently, this super-network is simultaneously pruned (using an entropy constraint) and quantized (that is, ternary values are assigned layer-wise) in a training process, resulting in a sparse and ternary network representation. We validate the proposed approach in CIFAR-10, CIFAR-100, and ImageNet datasets, showing its effectiveness in image classification tasks.
Orthogonality is widely used for training deep neural networks (DNNs) due to its ability to maintain all singular values of the Jacobian close to 1 and reduce redundancy in representation. This paper proposes a computationally efficient and numerically stable orthogonalization method using Newton's iteration (ONI), to learn a layer-wise orthogonal weight matrix in DNNs. ONI works by iteratively stretching the singular values of a weight matrix towards 1. This property enables it to control the orthogonality of a weight matrix by its number of iterations. We show that our method improves the performance of image classification networks by effectively controlling the orthogonality to provide an optimal tradeoff between optimization benefits and representational capacity reduction. We also show that ONI stabilizes the training of generative adversarial networks (GANs) by maintaining the Lipschitz continuity of a network, similar to spectral normalization (SN), and further outperforms SN by providing controllable orthogonality.
Image classification has been studied extensively but there has been limited work in the direction of using non-conventional, external guidance other than traditional image-label pairs to train such models. In this thesis we present a set of methods to leverage information about the semantic hierarchy induced by class labels. In the first part of the thesis, we inject label-hierarchy knowledge to an arbitrary classifier and empirically show that availability of such external semantic information in conjunction with the visual semantics from images boosts overall performance. Taking a step further in this direction, we model more explicitly the label-label and label-image interactions by using order-preserving embedding-based models, prevalent in natural language, and tailor them to the domain of computer vision to perform image classification. Although, contrasting in nature, both the CNN-classifiers injected with hierarchical information, and the embedding-based models outperform a hierarchy-agnostic model on the newly presented, real-world ETH Entomological Collection image dataset https://www.research-collection.ethz.ch/handle/20.500.11850/365379.
These days deep neural networks are ubiquitously used in a wide range of tasks, from image classification and machine translation to face identification and self-driving cars. In many applications, a single model error can lead to devastating financial, reputational and even life-threatening consequences. Therefore, it is crucially important to correct model mistakes quickly as they appear. In this work, we investigate the problem of neural network editing $-$ how one can efficiently patch a mistake of the model on a particular sample, without influencing the model behavior on other samples. Namely, we propose Editable Training, a model-agnostic training technique that encourages fast editing of the trained model. We empirically demonstrate the effectiveness of this method on large-scale image classification and machine translation tasks.
Recent advances in neural networks have made great progress in the hyperspectral image (HSI) classification. However, the overfitting effect, which is mainly caused by complicated model structure and small training set, remains a major concern. Reducing the complexity of the neural networks could prevent overfitting to some extent, but also declines the networks' ability to express more abstract features. Enlarging the training set is also difficult, for the high expense of acquisition and manual labeling. In this paper, we propose an abundance-based multi-HSI classification method. Firstly, we convert every HSI from the spectral domain to the abundance domain by a dataset-specific autoencoder. Secondly, the abundance representations from multiple HSIs are collected to form an enlarged dataset. Lastly, we train an abundance-based classifier and employ the classifier to predict over all the involved HSI datasets. Different from the spectra that are usually highly mixed, the abundance features are more representative in reduced dimension with less noise. This benefits the proposed method to employ simple classifiers and enlarged training data, and to expect less overfitting issues. The effectiveness of the proposed method is verified by the ablation study and the comparative experiments.
Convolutional neural networks have witnessed remarkable improvements in computational efficiency in recent years. A key driving force has been the idea of trading-off model expressivity and efficiency through a combination of $1\times 1$ and depth-wise separable convolutions in lieu of a standard convolutional layer. The price of the efficiency, however, is the sub-optimal flow of information across space and channels in the network. To overcome this limitation, we present MUXConv, a layer that is designed to increase the flow of information by progressively multiplexing channel and spatial information in the network, while mitigating computational complexity. Furthermore, to demonstrate the effectiveness of MUXConv, we integrate it within an efficient multi-objective evolutionary algorithm to search for the optimal model hyper-parameters while simultaneously optimizing accuracy, compactness, and computational efficiency. On ImageNet, the resulting models, dubbed MUXNets, match the performance (75.3% top-1 accuracy) and multiply-add operations (218M) of MobileNetV3 while being 1.6$\times$ more compact, and outperform other mobile models in all the three criteria. MUXNet also performs well under transfer learning and when adapted to object detection. On the ChestX-Ray 14 benchmark, its accuracy is comparable to the state-of-the-art while being $3.3\times$ more compact and $14\times$ more efficient. Similarly, detection on PASCAL VOC 2007 is 1.2% more accurate, 28% faster and 6% more compact compared to MobileNetV2. Code is available from https://github.com/human-analysis/MUXConv
Deep neural networks with millions of parameters may suffer from poor generalization due to overfitting. To mitigate the issue, we propose a new regularization method that penalizes the predictive distribution between similar samples. In particular, we distill the predictive distribution between different samples of the same label during training. This results in regularizing the dark knowledge (i.e., the knowledge on wrong predictions) of a single network (i.e., a self-knowledge distillation) by forcing it to produce more meaningful and consistent predictions in a class-wise manner. Consequently, it mitigates overconfident predictions and reduces intra-class variations. Our experimental results on various image classification tasks demonstrate that the simple yet powerful method can significantly improve not only the generalization ability but also the calibration performance of modern convolutional neural networks.
Augmenting training datasets has been shown to improve the learning effectiveness for several computer vision tasks. A good augmentation produces an augmented dataset that adds variability while retaining the statistical properties of the original dataset. Some techniques, such as AutoAugment and Fast AutoAugment, have introduced a search phase to find a set of suitable augmentation policies for a given model and dataset. This comes at the cost of great computational overhead, adding up to several thousand GPU hours. More recently RandAugment was proposed to substantially speedup the search phase by approximating the search space by a couple of hyperparameters, but still incurring non-negligible cost for tuning those. In this paper we show that, under the assumption that the augmentation space is approximately distribution invariant, a uniform sampling over the continuous space of augmentation transformations is sufficient to train highly effective models. Based on that result we propose UniformAugment, an automated data augmentation approach that completely avoids a search phase. In addition to discussing the theoretical underpinning supporting our approach, we also use the standard datasets, as well as established models for image classification, to show that UniformAugment's effectiveness is comparable to the aforementioned methods, while still being highly efficient by virtue of not requiring any search.
We revisit the long-standing problem of learning from a small sample, to which end we propose a novel method called GLICO (Generative Latent Implicit Conditional Optimization). GLICO learns a mapping from the training examples to a latent space and a generator that generates images from vectors in the latent space. Unlike most recent works, which rely on access to large amounts of unlabeled data, GLICO does not require access to any additional data other than the small set of labeled points. In fact, GLICO learns to synthesize completely new samples for every class using as little as 5 or 10 examples per class, with as few as 10 such classes without imposing any prior. GLICO is then used to augment the small training set while training a classifier on the small sample. To this end, our proposed method samples the learned latent space using spherical interpolation, and generates new examples using the trained generator. Empirical results show that the new sampled set is diverse enough, leading to improvement in image classification in comparison with the state of the art, when trained on small samples obtained from CIFAR-10, CIFAR-100, and CUB-200.
Bayesian inference is used extensively to quantify the uncertainty in an inferred field given the measurement of a related field when the two are linked by a mathematical model. Despite its many applications, Bayesian inference faces challenges when inferring fields that have discrete representations of large dimension, and/or have prior distributions that are difficult to characterize mathematically. In this work we demonstrate how the approximate distribution learned by a deep generative adversarial network (GAN) may be used as a prior in a Bayesian update to address both these challenges. We demonstrate the efficacy of this approach on two distinct, and remarkably broad, classes of problems. The first class leads to supervised learning algorithms for image classification with superior out of distribution detection and accuracy, and for image inpainting with built-in variance estimation. The second class leads to unsupervised learning algorithms for image denoising and for solving physics-driven inverse problems.
Modern deep learning approaches have achieved great success in many vision applications by training a model using all available task-specific data. However, there are two major obstacles making it challenging to implement for real life applications: (1) Learning new classes makes the trained model quickly forget old classes knowledge, which is referred to as catastrophic forgetting. (2) As new observations of old classes come sequentially over time, the distribution may change in unforeseen way, making the performance degrade dramatically on future data, which is referred to as concept drift. Current state-of-the-art incremental learning methods require a long time to train the model whenever new classes are added and none of them takes into consideration the new observations of old classes. In this paper, we propose an incremental learning framework that can work in the challenging online learning scenario and handle both new classes data and new observations of old classes. We address problem (1) in online mode by introducing a modified cross-distillation loss together with a two-step learning technique. Our method outperforms the results obtained from current state-of-the-art offline incremental learning methods on the CIFAR-100 and ImageNet-1000 (ILSVRC 2012) datasets under the same experiment protocol but in online scenario. We also provide a simple yet effective method to mitigate problem (2) by updating exemplar set using the feature of each new observation of old classes and demonstrate a real life application of online food image classification based on our complete framework using the Food-101 dataset.