Learning Transformation Equivariant Representations (TERs) seeks to capture the intrinsic visual structures of images through the representations that equivary to the applied transformations. It assumes that a transformation should be decoded from expressive representations of images before and after transformations. It greatly expands the scope of {\em translation} equivariance pinpointing the success of the Convolutional Neural Networks (CNNs) to develop a generic class of {\em transformation} equivariant representations. Unlike group equivariant convolutions that are limited to discrete transformations or linear transformation equivariance, we present a more flexible and tractable AutoEncoding Transformation (AET) model that can handle various types of transformations. Both deterministic AET and probabilistic Autoencoding Variational Transformations (AVT) models are presented. While the former trains transformation equivariant representations by directly reconstructing applied transformations, the latter is trained by maximizing the joint mutual information between the representations and the transformations. It leads to the Generalized TERs (GTERs) that could equivary against transformations in a more general manner by enabling them to capture more complex patterns of transformed visual structures beyond the linear TERs of a transformation group. We will further show that the presented approach can be extended to (semi-)supervised models by jointly maximizing the mutual information in the learned representations about the input labels and transformations. Experiment results following the standard evaluation protocols demonstrate the superior performances of the proposed models to the existing state-of-the-art unsupervised and (semi-)supervised approaches in literature.
The Long Short-Term Memory (LSTM) recurrent neural network is capable of processing complex sequential information since it utilizes special gating schemes for learning representations from long input sequences. It has the potential to model any sequential time-series data, where the current hidden state has to be considered in the context of the past hidden states. This property makes LSTM an ideal choice to learn the complex dynamics present in long sequences. Unfortunately, the conventional LSTMs do not consider the impact of spatio-temporal dynamics corresponding to the given salient motion patterns, when they gate the information that ought to be memorized through time. To address this problem, we propose a differential gating scheme for the LSTM neural network, which emphasizes on the change in information gain caused by the salient motions between the successive video frames. This change in information gain is quantified by Derivative of States (DoS), and thus the proposed LSTM model is termed as differential Recurrent Neural Network (dRNN). In addition, the original work used the hidden state at the last time-step to model the entire video sequence. Based on the energy profiling of DoS, we further propose to employ the State Energy Profile (SEP) to search for salient dRNN states and construct more informative representations. The effectiveness of the proposed model was demonstrated by automatically recognizing human actions from the real-world 2D and 3D single-person action datasets. We point out that LSTM is a special form of dRNN. As a result, we have introduced a new family of LSTMs. Our study is one of the first works towards demonstrating the potential of learning complex time-series representations via high-order derivatives of states.
The learning of Transformation-Equivariant Representations (TERs), which is introduced by Hinton et al. \cite{hinton2011transforming}, has been considered as a principle to reveal visual structures under various transformations. It contains the celebrated Convolutional Neural Networks (CNNs) as a special case that only equivary to the translations. In contrast, we seek to train TERs for a generic class of transformations and train them in an {\em unsupervised} fashion. To this end, we present a novel principled method by Autoencoding Variational Transformations (AVT), compared with the conventional approach to autoencoding data. Formally, given transformed images, the AVT seeks to train the networks by maximizing the mutual information between the transformations and representations. This ensures the resultant TERs of individual images contain the {\em intrinsic} information about their visual structures that would equivary {\em extricably} under various transformations. Technically, we show that the resultant optimization problem can be efficiently solved by maximizing a variational lower-bound of the mutual information. This variational approach introduces a transformation decoder to approximate the intractable posterior of transformations, resulting in an autoencoding architecture with a pair of the representation encoder and the transformation decoder. Experiments demonstrate the proposed AVT model sets a new record for the performances on unsupervised tasks, greatly closing the performance gap to the supervised models.
Small data challenges have emerged in many learning problems, since the success of deep neural networks often relies on the availability of a huge amount of labeled data that is expensive to collect. To address it, many efforts have been made on training complex models with small data in an unsupervised and semi-supervised fashion. In this paper, we will review the recent progresses on these two major categories of methods. A wide spectrum of small data models will be categorized in a big picture, where we will show how they interplay with each other to motivate explorations of new ideas. We will review the criteria of learning the transformation equivariant, disentangled, self-supervised and semi-supervised representations, which underpin the foundations of recent developments. Many instantiations of unsupervised and semi-supervised generative models have been developed on the basis of these criteria, greatly expanding the territory of existing autoencoders, generative adversarial nets (GANs) and other deep networks by exploring the distribution of unlabeled data for more powerful representations. While we focus on the unsupervised and semi-supervised methods, we will also provide a broader review of other emerging topics, from unsupervised and semi-supervised domain adaptation to the fundamental roles of transformation equivariance and invariance in training a wide spectrum of deep networks. It is impossible for us to write an exclusive encyclopedia to include all related works. Instead, we aim at exploring the main ideas, principles and methods in this area to reveal where we are heading on the journey towards addressing the small data challenges in this big data era.
Recent advancements in recurrent neural network (RNN) research have demonstrated the superiority of utilizing multiscale structures in learning temporal representations of time series. Currently, most of multiscale RNNs use fixed scales, which do not comply with the nature of dynamical temporal patterns among sequences. In this paper, we propose Adaptively Scaled Recurrent Neural Networks (ASRNN), a simple but efficient way to handle this problem. Instead of using predefined scales, ASRNNs are able to learn and adjust scales based on different temporal contexts, making them more flexible in modeling multiscale patterns. Compared with other multiscale RNNs, ASRNNs are bestowed upon dynamical scaling capabilities with much simpler structures, and are easy to be integrated with various RNN cells. The experiments on multiple sequence modeling tasks indicate ASRNNs can efficiently adapt scales based on different sequence contexts and yield better performances than baselines without dynamical scaling abilities.
The success of deep neural networks often relies on a large amount of labeled examples, which can be difficult to obtain in many real scenarios. To address this challenge, unsupervised methods are strongly preferred for training neural networks without using any labeled data. In this paper, we present a novel paradigm of unsupervised representation learning by Auto-Encoding Transformation (AET) in contrast to the conventional Auto-Encoding Data (AED) approach. Given a randomly sampled transformation, AET seeks to predict it merely from the encoded features as accurately as possible at the output end. The idea is the following: as long as the unsupervised features successfully encode the essential information about the visual structures of original and transformed images, the transformation can be well predicted. We will show that this AET paradigm allows us to instantiate a large variety of transformations, from parameterized, to non-parameterized and GAN-induced ones. Our experiments show that AET greatly improves over existing unsupervised approaches, setting new state-of-the-art performances being greatly closer to the upper bounds by their fully supervised counterparts on CIFAR-10, ImageNet and Places datasets.
Hashing has been widely applied to multimodal retrieval on large-scale multimedia data due to its efficiency in computation and storage. Particularly, deep hashing has received unprecedented research attention in recent years, owing to its perfect retrieval performance. However, most of existing deep hashing methods learn binary hash codes by preserving the similarity relationship while without exploiting the semantic labels, which result in suboptimal binary codes. In this work, we propose a novel Deep Semantic Multimodal Hashing Network (DSMHN) for scalable multimodal retrieval. In DSMHN, two sets of modality-specific hash functions are jointly learned by explicitly preserving both the inter-modality similarities and the intra-modality semantic labels. Specifically, with the assumption that the learned hash codes should be optimal for task-specific classification, two stream networks are jointly trained to learn the hash functions by embedding the semantic labels on the resultant hash codes. Different from previous deep hashing methods, which are tied to some particular forms of loss functions, our deep hashing framework can be flexibly integrated with different types of loss functions. In addition, the bit balance property is investigated to generate binary codes with each bit having $50\%$ probability to be $1$ or $-1$. Moreover, a unified deep multimodal hashing framework is proposed to learn compact and high-quality hash codes by exploiting the feature representation learning, inter-modality similarity preserving learning, semantic label preserving learning and hash functions learning with bit balanced constraint simultaneously. We conduct extensive experiments for both unimodal and cross-modal retrieval tasks on three widely-used multimodal retrieval datasets. The experimental result demonstrates that DSMHN significantly outperforms state-of-the-art methods.
In this paper, we aim to address the problem of human interaction recognition in videos by exploring the long-term inter-related dynamics among multiple persons. Recently, Long Short-Term Memory (LSTM) has become a popular choice to model individual dynamic for single-person action recognition due to its ability of capturing the temporal motion information in a range. However, existing RNN models focus only on capturing the dynamics of human interaction by simply combining all dynamics of individuals or modeling them as a whole. Such models neglect the inter-related dynamics of how human interactions change over time. To this end, we propose a novel Hierarchical Long Short-Term Concurrent Memory (H-LSTCM) to model the long-term inter-related dynamics among a group of persons for recognizing the human interactions. Specifically, we first feed each person's static features into a Single-Person LSTM to learn the single-person dynamic. Subsequently, the outputs of all Single-Person LSTM units are fed into a novel Concurrent LSTM (Co-LSTM) unit, which mainly consists of multiple sub-memory units, a new cell gate and a new co-memory cell. In a Co-LSTM unit, each sub-memory unit stores individual motion information, while this Co-LSTM unit selectively integrates and stores inter-related motion information between multiple interacting persons from multiple sub-memory units via the cell gate and co-memory cell, respectively. Extensive experiments on four public datasets validate the effectiveness of the proposed H-LSTCM by comparing against baseline and state-of-the-art methods.
In this paper, we formalize the idea behind capsule nets of using a capsule vector rather than a neuron activation to predict the label of samples. To this end, we propose to learn a group of capsule subspaces onto which an input feature vector is projected. Then the lengths of resultant capsules are used to score the probability of belonging to different classes. We train such a Capsule Projection Network (CapProNet) by learning an orthogonal projection matrix for each capsule subspace, and show that each capsule subspace is updated until it contains input feature vectors corresponding to the associated class. We will also show that the capsule projection can be viewed as normalizing the multiple columns of the weight matrix simultaneously to form an orthogonal basis, which makes it more effective in incorporating novel components of input features to update capsule representations. In other words, the capsule projection can be viewed as a multi-dimensional weight normalization in capsule subspaces, where the conventional weight normalization is simply a special case of the capsule projection onto 1D lines. Only a small negligible computing overhead is incurred to train the network in low-dimensional capsule subspaces or through an alternative hyper-power iteration to estimate the normalization matrix. Experiment results on image datasets show the presented model can greatly improve the performance of the state-of-the-art ResNet backbones by $10-20\%$ and that of the Densenet by $5-7\%$ respectively at the same level of computing and memory expenses. The CapProNet establishes the competitive state-of-the-art performance for the family of capsule nets by significantly reducing test errors on the benchmark datasets.
In this paper, we present novel sharp attention networks by adaptively sampling feature maps from convolutional neural networks (CNNs) for person re-identification (re-ID) problem. Due to the introduction of sampling-based attention models, the proposed approach can adaptively generate sharper attention-aware feature masks. This greatly differs from the gating-based attention mechanism that relies soft gating functions to select the relevant features for person re-ID. In contrast, the proposed sampling-based attention mechanism allows us to effectively trim irrelevant features by enforcing the resultant feature masks to focus on the most discriminative features. It can produce sharper attentions that are more assertive in localizing subtle features relevant to re-identifying people across cameras. For this purpose, a differentiable Gumbel-Softmax sampler is employed to approximate the Bernoulli sampling to train the sharp attention networks. Extensive experimental evaluations demonstrate the superiority of this new sharp attention model for person re-ID over the other state-of-the-art methods on three challenging benchmarks including CUHK03, Market-1501, and DukeMTMC-reID.