In this paper, we aim to understand the generalization properties of generative adversarial networks (GANs) from a new perspective of privacy protection. Theoretically, we prove that a differentially private learning algorithm used for training the GAN does not overfit to a certain degree, i.e., the generalization gap can be bounded. Moreover, some recent works, such as the Bayesian GAN, can be re-interpreted based on our theoretical insight from privacy protection. Quantitatively, to evaluate the information leakage of well-trained GAN models, we perform various membership attacks on these models. The results show that previous Lipschitz regularization techniques are effective in not only reducing the generalization gap but also alleviating the information leakage of the training dataset.

We focus on developing a novel scalable graph-based semi-supervised learning (SSL) method for a small number of labeled data and a large amount of unlabeled data. Due to the lack of labeled data and the availability of large-scale unlabeled data, existing SSL methods usually encounter either suboptimal performance because of an improper graph or the high computational complexity of the large-scale optimization problem. In this paper, we propose to address both challenging problems by constructing a proper graph for graph-based SSL methods. Different from existing approaches, we simultaneously learn a small set of vertexes to characterize the high-dense regions of the input data and a graph to depict the relationships among these vertexes. A novel approach is then proposed to construct the graph of the input data from the learned graph of a small number of vertexes with some preferred properties. Without explicitly calculating the constructed graph of inputs, two transductive graph-based SSL approaches are presented with the computational complexity in linear with the number of input data. Extensive experiments on synthetic data and real datasets of varied sizes demonstrate that the proposed method is not only scalable for large-scale data, but also achieve good classification performance, especially for extremely small number of labels.

Bayesian deep learning is recently regarded as an intrinsic way to characterize the weight uncertainty of deep neural networks~(DNNs). Stochastic Gradient Langevin Dynamics~(SGLD) is an effective method to enable Bayesian deep learning on large-scale datasets. Previous theoretical studies have shown various appealing properties of SGLD, ranging from the convergence properties to the generalization bounds. In this paper, we study the properties of SGLD from a novel perspective of membership privacy protection (i.e., preventing the membership attack). The membership attack, which aims to determine whether a specific sample is used for training a given DNN model, has emerged as a common threat against deep learning algorithms. To this end, we build a theoretical framework to analyze the information leakage (w.r.t. the training dataset) of a model trained using SGLD. Based on this framework, we demonstrate that SGLD can prevent the information leakage of the training dataset to a certain extent. Moreover, our theoretical analysis can be naturally extended to other types of Stochastic Gradient Markov Chain Monte Carlo (SG-MCMC) methods. Empirical results on different datasets and models verify our theoretical findings and suggest that the SGLD algorithm can not only reduce the information leakage but also improve the generalization ability of the DNN models in real-world applications.

We propose an efficient algorithm for solving orthogonal canonical correlation analysis (OCCA) in the form of trace-fractional structure and orthogonal linear projections. Even though orthogonality has been widely used and proved to be a useful criterion for pattern recognition and feature extraction, existing methods for solving OCCA problem are either numerical unstable by relying on a deflation scheme, or less efficient by directly using generic optimization methods. In this paper, we propose an alternating numerical scheme whose core is the sub-maximization problem in the trace-fractional form with an orthogonal constraint. A customized self-consistent-field (SCF) iteration for this sub-maximization problem is devised. It is proved that the SCF iteration is globally convergent to a KKT point and that the alternating numerical scheme always converges. We further formulate a new trace-fractional maximization problem for orthogonal multiset CCA (OMCCA) and then propose an efficient algorithm with an either Jacobi-style or Gauss-Seidel-style updating scheme based on the same SCF iteration. Extensive experiments are conducted to evaluate the proposed algorithms against existing methods including two real world applications: multi-label classification and multi-view feature extraction. Experimental results show that our methods not only perform competitively to or better than baselines but also are more efficient.

In this paper, we aim to understand the generalization properties of generative adversarial networks (GANs) from a new perspective of privacy protection. Theoretically, we prove that a differentially private learning algorithm used for training the GAN does not overfit to a certain degree, i.e., the generalization gap can be bounded. Moreover, some recent works, such as the Bayesian GAN, can be re-interpreted based on our theoretical insight from privacy protection. Quantitatively, to evaluate the information leakage of well-trained GAN models, we perform various membership attacks on these models. The results show that previous Lipschitz regularization techniques are effective in not only reducing the generalization gap but also alleviating the information leakage of the training dataset.

Human brain is composed of roughly 100 billion neurons and brain functions are carried out by complex firing and interactions among the neurons, accompanied with electromagnetic, hemodynamic, and metabolic changes [1]. As the electromagnetic is directly related to the neural firing activities, it reflects the real-time dynamical process of the brain, which can be directly measured by Electroencephalogram (EEG) and Magnetoencephalography (MEG). Both EEG and MEG yield a much higher temporal resolution up to a few milliseconds than other brain imaging modalities such as functional magnetic resonance imaging (fMRI), positron emission tomography (PET), and single-photon emission computed tomography (SPECT) [1-4]. However, one limitation of EEG/MEG is the low spatial resolution, as the corresponding measurements are acquired on the scalp with little information regarding neural activations inside the brain. Reconstructing a brain source signal from EEG/MEG measurements is known as EEG/MEG source localization or EEG/MEG source imaging (ESI) [5]. The ESI techniques have been used in several clinical and/or brain research applications such as the study of language mechanisms, cognition process and sensory function with a brain-computer interface [6], the localization of primary sensory cortex in evoked potentials for surgical candidates [7], and the localization of the irritative zone in focal epilepsy [8] [9]. In general, the number of EEG/MEG sensors is much less than the number of brain sources and hence the ESI problem is highly ill-posed. In order to find a reasonable solution, it is necessary to impose certain neurophysiologically plausible assumptions as regularizations [5] [10].

Value functions are crucial for model-free Reinforcement Learning (RL) to obtain a policy implicitly or guide the policy updates. Value estimation heavily depends on the stochasticity of environmental dynamics and the quality of reward signals. In this paper, we propose a two-step understanding of value estimation from the perspective of future prediction, through decomposing the value function into a reward-independent future dynamics part and a policy-independent trajectory return part. We then derive a practical deep RL algorithm from the above decomposition, consisting of a convolutional trajectory representation model, a conditional variational dynamics model to predict the expected representation of future trajectory and a convex trajectory return model that maps a trajectory representation to its return. Our algorithm is evaluated in MuJoCo continuous control tasks and shows superior results under both common settings and delayed reward settings.

Glioma grading before the surgery is very critical for the prognosis prediction and treatment plan making. In this paper, we present a novel scattering wavelet-based radiomics method to predict noninvasively and accurately the glioma grades. The multimodal magnetic resonance images of 285 patients were used, with the intratumoral and peritumoral regions well labeled. The wavelet scattering-based features and traditional radiomics features were firstly extracted from both intratumoral and peritumoral regions respectively. The support vector machine (SVM), logistic regression (LR) and random forest (RF) were then trained with 5-fold cross validation to predict the glioma grades. The prediction obtained with different features was finally evaluated in terms of quantitative metrics. The area under the receiver operating characteristic curve (AUC) of glioma grade prediction based on scattering wavelet features was up to 0.99 when considering both intratumoral and peritumoral features in multimodal images, which increases by about 17% compared to traditional radiomics. Such results shown that the local invariant features extracted from the scattering wavelet transform allows improving the prediction accuracy for glioma grading. In addition, the features extracted from peritumoral regions further increases the accuracy of glioma grading.

Despite our unique ability to use natural languages, we know little about their origins like how they are created and evolved. The answer lies deeply in the evolution of our cognitive and social abilities over a very long period of time which is beyond our scrutiny. Existing studies on the origin of languages are often focused on the emergence of specific language features (such as recursion) without supporting a comprehensive view. Investigation of restricted language representations, such as temporal logic, unfortunately does not reveal much about the impetus underlying language formation and evolution, since much of their construction is based on natural languages themselves. In this paper, we investigate the origin of "natural languages" in a restricted setting involving only planning agents. Similar to a common view that considers languages as a tool for grounding symbols to semantic meanings, we take the view that a language for planning agents is a tool for grounding symbols to physical configurations. From this perspective, a language is used by the agents to coordinate their behaviors during planning. With a few assumptions, we show that language is closely connected to a type of domain abstractions, based on which a language can be constructed. We study how such abstractions can be identified and discuss how to use them during planning. We apply our method to several domains, discuss the results, and relaxation of the assumptions made.

Despite deep reinforcement learning has recently achieved great successes, however in multiagent environments, a number of challenges still remain. Multiagent reinforcement learning (MARL) is commonly considered to suffer from the problem of non-stationary environments and exponentially increasing policy space. It would be even more challenging to learn effective policies in circumstances where the rewards are sparse and delayed over long trajectories. In this paper, we study Hierarchical Deep Multiagent Reinforcement Learning (hierarchical deep MARL) in cooperative multiagent problems with sparse and delayed rewards, where efficient multiagent learning methods are desperately needed. We decompose the original MARL problem into hierarchies and investigate how effective policies can be learned hierarchically in synchronous/asynchronous hierarchical MARL frameworks. Several hierarchical deep MARL architectures, i.e., Ind-hDQN, hCom and hQmix, are introduced for different learning paradigms. Moreover, to alleviate the issues of sparse experiences in high-level learning and non-stationarity in multiagent settings, we propose a new experience replay mechanism, named as Augmented Concurrent Experience Replay (ACER). We empirically demonstrate the effects and efficiency of our approaches in several classic Multiagent Trash Collection tasks, as well as in an extremely challenging team sports game, i.e., Fever Basketball Defense.