We propose a deep amortized clustering (DAC), a neural architecture which learns to cluster datasets efficiently using a few forward passes. DAC implicitly learns what makes a cluster, how to group data points into clusters, and how to count the number of clusters in datasets. DAC is meta-learned using labelled datasets for training, a process distinct from traditional clustering algorithms which usually require hand-specified prior knowledge about cluster shapes/structures. We empirically show, on both synthetic and image data, that DAC can efficiently and accurately cluster new datasets coming from the same distribution used to generate training datasets.
Boltzmann Machines constitute a class of neural networks with applications to image reconstruction, pattern classification and unsupervised learning in general. Their most common variants, called Restricted Boltzmann Machines (RBMs) exhibit a good trade-off between computability on existing silicon-based hardware and generality of possible applications. Still, the diffusion of RBMs is quite limited, since their training process proves to be hard. The advent of commercial Adiabatic Quantum Computers (AQCs) raised the expectation that the implementations of RBMs on such quantum devices could increase the training speed with respect to conventional hardware. To date, however, the implementation of RBM networks on AQCs has been limited by the low qubit connectivity when each qubit acts as a node of the neural network. Here we demonstrate the feasibility of a complete RBM on AQCs, thanks to an embedding that associates its nodes to virtual qubits, thus outperforming previous implementations based on incomplete graphs. Moreover, to accelerate the learning, we implement a semantic quantum search which, contrary to previous proposals, takes the input data as initial boundary conditions to start each learning step of the RBM, thanks to a reverse annealing schedule. Such an approach, unlike the more conventional forward annealing schedule, allows sampling configurations in a meaningful neighborhood of the training data, mimicking the behavior of the classical Gibbs sampling algorithm. We show that the learning based on reverse annealing quickly raises the sampling probability of a meaningful subset of the set of the configurations. Even without a proper optimization of the annealing schedule, the RBM semantically trained by reverse annealing achieves better scores on reconstruction tasks.
State-of-the-art adversarial attacks on neural networks use expensive iterative methods and numerous random restarts from different initial points. Iterative FGSM-based methods without restarts trade off performance for computational efficiency because they do not adequately explore the image space and are highly sensitive to the choice of step size. We propose a variant of Projected Gradient Descent (PGD) that uses a random step size to improve performance without resorting to expensive random restarts. Our method, Wide Iterative Stochastic crafting (WITCHcraft), achieves results superior to the classical PGD attack on the CIFAR-10 and MNIST data sets but without additional computational cost. This simple modification of PGD makes crafting attacks more economical, which is important in situations like adversarial training where attacks need to be crafted in real time.
Diversity and unpredictability of artifacts potentially presented to an iris sensor calls for presentation attack detection methods that are agnostic to specificity of presentation attack instruments. This paper proposes a method that combines two-dimensional and three-dimensional properties of the observed iris to address the problem of spoof detection in case when some properties of artifacts are unknown. The 2D (textural) iris features are extracted by a state-of-the-art method employing Binary Statistical Image Features (BSIF) and an ensemble of classifiers is used to deliver 2D modality-related decision. The 3D (shape) iris features are reconstructed by a photometric stereo method from only two images captured under near-infrared illumination placed at two different angles, as in many current commercial iris recognition sensors. The map of normal vectors is used to assess the convexity of the observed iris surface. The combination of these two approaches has been applied to detect whether a subject is wearing a textured contact lens to disguise their identity. Extensive experiments with NDCLD'15 dataset, and a newly collected NDIris3D dataset show that the proposed method is highly robust under various open-set testing scenarios, and that it outperforms all available open-source iris PAD methods tested in identical scenarios. The source code and the newly prepared benchmark are made available along with this paper.
This paper introduces a novel contour-based approach named deep snake for real-time instance segmentation. Unlike some recent methods that directly regress the coordinates of the object boundary points from an image, deep snake uses a neural network to iteratively deform an initial contour to the object boundary, which implements the classic idea of snake algorithms with a learning-based approach. For structured feature learning on the contour, we propose to use circular convolution in deep snake, which better exploits the cycle-graph structure of a contour compared against generic graph convolution. Based on deep snake, we develop a two-stage pipeline for instance segmentation: initial contour proposal and contour deformation, which can handle errors in initial object localization. Experiments show that the proposed approach achieves state-of-the-art performances on the Cityscapes, Kins and Sbd datasets while being efficient for real-time instance segmentation, 32.3 fps for 512$\times$512 images on a 1080Ti GPU. The code will be available at https://github.com/zju3dv/snake/.
We introduce a parameterization method called Neural Bayes which allows computing statistical quantities that are in general difficult to compute and opens avenues for formulating new objectives for unsupervised representation learning. Specifically, given an observed random variable $\mathbf{x}$ and a latent discrete variable $z$, we can express $p(\mathbf{x}|z)$, $p(z|\mathbf{x})$ and $p(z)$ in closed form in terms of a sufficiently expressive function (Eg. neural network) using our parameterization without restricting the class of these distributions. To demonstrate its usefulness, we develop two independent use cases for this parameterization: 1. Mutual Information Maximization (MIM): MIM has become a popular means for self-supervised representation learning. Neural Bayes allows us to compute mutual information between observed random variables $\mathbf{x}$ and latent discrete random variables $z$ in closed form. We use this for learning image representations and show its usefulness on downstream classification tasks. 2. Disjoint Manifold Labeling: Neural Bayes allows us to formulate an objective which can optimally label samples from disjoint manifolds present in the support of a continuous distribution. This can be seen as a specific form of clustering where each disjoint manifold in the support is a separate cluster. We design clustering tasks that obey this formulation and empirically show that the model optimally labels the disjoint manifolds. Our code is available at \url{https://github.com/salesforce/NeuralBayes}
Visual relationship detection is an intermediate image understanding task that detects two objects and classifies a predicate that explains the relationship between two objects in an image. The three components are linguistically and visually correlated (e.g. "wear" is related to "person" and "shirt", while "laptop" is related to "table" and "on") thus, the solution space is huge because there are many possible cases between them. Language and visual modules are exploited and a sophisticated spatial vector is proposed. The models in this work outperformed the state of arts without costly linguistic knowledge distillation from a large text corpus and building complex loss functions. All experiments were only evaluated on Visual Relationship Detection and Visual Genome dataset.
The two-dimensional principal component analysis (2DPCA) has become one of the most powerful tools of artificial intelligent algorithms. In this paper, we review 2DPCA and its variations, and propose a general ridge regression model to extract features from both row and column directions. To enhance the generalization ability of extracted features, a novel relaxed 2DPCA (R2DPCA) is proposed with a new ridge regression model. R2DPCA generates a weighting vector with utilizing the label information, and maximizes a relaxed criterion with applying an optimal algorithm to get the essential features. The R2DPCA-based approaches for face recognition and image reconstruction are also proposed and the selected principle components are weighted to enhance the role of main components. Numerical experiments on well-known standard databases indicate that R2DPCA has high generalization ability and can achieve a higher recognition rate than the state-of-the-art methods, including in the deep learning methods such as CNNs, DBNs, and DNNs.
Modern deep learning techniques can be employed to generate effective feature extractors for the task of iris recognition. The question arises: should we train such structures from scratch on a relatively large iris image dataset, or it is better to fine-tune the existing models to adapt them to a new domain? In this work we explore five different sets of weights for the popular ResNet-50 architecture to find out whether iris-specific feature extractors perform better than models trained for non-iris tasks. Features are extracted from each convolutional layer and the classification accuracy achieved by a Support Vector Machine is measured on a dataset that is disjoint from the samples used in training of the ResNet-50 model. We show that the optimal training strategy is to fine-tune an off-the-shelf set of weights to the iris recognition domain. This approach results in greater accuracy than both off-the-shelf weights and a model trained from scratch. The winning, fine-tuned approach also shows an increase in performance when compared to previous work, in which only off-the-shelf (not fine-tuned) models were used in iris feature extraction. We make the best-performing ResNet-50 model, fine-tuned with more than 360,000 iris images, publicly available along with this paper.
The recent years have witnessed the great potential of deep learning for video compression. In this paper, we propose the Hierarchical Learned Video Compression (HLVC) approach with three hierarchical quality layers and recurrent enhancement. To be specific, the frames in the first layer are compressed by image compression method with the highest quality. Using them as references, we propose the Bi-Directional Deep Compression (BDDC) network to compress the second layer with relatively high quality. Then, the third layer frames are compressed with the lowest quality, by the proposed Single Motion Deep Compression (SMDC) network, which adopts a single motion map to estimate the motions of multiple frames, thus saving the bit-rate for motion information. In our deep decoder, we develop the Weighted Recurrent Quality Enhancement (WRQE) network with the inputs of both compressed frames and bit stream. In the recurrent cell of WRQE, the memory and update signal are weighted by quality features to reasonably leverage multi-frame information for enhancement. In our HLVC approach, the hierarchical quality benefits the coding efficiency, since the high quality information facilitates the compression and enhancement of low quality frames at encoder and decoder sides, respectively. Finally, the experiments validate that our HLVC approach advances the state-of-the-art deep video compression methods, and outperforms x265 low delay P very fast mode in terms of both PSNR and MS-SSIM. The project page is at https://github.com/RenYang-home/HLVC.