MED-TEX: Transferring and Explaining Knowledge with Less Data from Pretrained Medical Imaging Models

Deep neural network based image classification methods usually require a large amount of training data and lack interpretability, which are critical in the medical imaging domain. In this paper, we develop a novel knowledge distillation and model interpretation framework for medical image classification that jointly solves the above two issues. Specifically, to address the data-hungry issue, we propose to learn a small student model with less data by distilling knowledge only from a cumbersome pretrained teacher model. To interpret the teacher model as well as assisting the learning of the student, an explainer module is introduced to highlight the regions of an input medical image that are important for the predictions of the teacher model. Furthermore, the joint framework is trained by a principled way derived from the information-theoretic perspective. Our framework performance is demonstrated by the comprehensive experiments on the knowledge distillation and model interpretation tasks compared to state-of-the-art methods on a fundus disease dataset.

Improving Adversarial Robustness by Enforcing Local and Global Compactness

The fact that deep neural networks are susceptible to crafted perturbations severely impacts the use of deep learning in certain domains of application. Among many developed defense models against such attacks, adversarial training emerges as the most successful method that consistently resists a wide range of attacks. In this work, based on an observation from a previous study that the representations of a clean data example and its adversarial examples become more divergent in higher layers of a deep neural net, we propose the Adversary Divergence Reduction Network which enforces local/global compactness and the clustering assumption over an intermediate layer of a deep neural network. We conduct comprehensive experiments to understand the isolating behavior of each component (i.e., local/global compactness and the clustering assumption) and compare our proposed model with state-of-the-art adversarial training methods. The experimental results demonstrate that augmenting adversarial training with our proposed components can further improve the robustness of the network, leading to higher unperturbed and adversarial predictive performances.

A Self-Attention Network based Node Embedding Model

Despite several signs of progress have been made recently, limited research has been conducted for an inductive setting where embeddings are required for newly unseen nodes -- a setting encountered commonly in practical applications of deep learning for graph networks. This significantly affects the performances of downstream tasks such as node classification, link prediction or community extraction. To this end, we propose SANNE -- a novel unsupervised embedding model -- whose central idea is to employ a transformer self-attention network to iteratively aggregate vector representations of nodes in random walks. Our SANNE aims to produce plausible embeddings not only for present nodes, but also for newly unseen nodes. Experimental results show that the proposed SANNE obtains state-of-the-art results for the node classification task on well-known benchmark datasets.

OptiGAN: Generative Adversarial Networks for Goal Optimized Sequence Generation

One of the challenging problems in sequence generation tasks is the optimized generation of sequences with specific desired goals. Current sequential generative models mainly generate sequences to closely mimic the training data, without direct optimization of desired goals or properties specific to the task. We introduce OptiGAN, a generative model that incorporates both Generative Adversarial Networks (GAN) and Reinforcement Learning (RL) to optimize desired goal scores using policy gradients. We apply our model to text and real-valued sequence generation, where our model is able to achieve higher desired scores out-performing GAN and RL baselines, while not sacrificing output sample diversity.

A Capsule Network-based Model for Learning Node Embeddings

In this paper, we focus on learning low-dimensional embeddings of entity nodes from graph-structured data, where we can use the learned node embeddings for a downstream task of node classification. Existing node embedding models often suffer from a limitation of exploiting graph information to infer plausible embeddings of unseen nodes. To address this issue, we propose Caps2NE---a new unsupervised embedding model using a network of two capsule layers. Given a target node and its context nodes, Caps2NE applies a routing process to aggregate features of the context nodes at the first capsule layer, then feed these features into the second capsule layer to produce an embedding vector. This embedding vector is then used to infer a plausible embedding for the target node. Experimental results for the node classification task on six well-known benchmark datasets show that our Caps2NE obtains state-of-the-art performances.

On Scalable Variant of Wasserstein Barycenter

We study a variant of Wasserstein barycenter problem, which we refer to as \emph{tree-sliced Wasserstein barycenter}, by leveraging the structure of tree metrics for the ground metrics in the formulation of Wasserstein distance. Drawing on the tree structure, we propose efficient algorithms for solving the unconstrained and constrained versions of tree-sliced Wasserstein barycenter. The algorithms have fast computational time and efficient memory usage, especially for high dimensional settings while demonstrating favorable results when the tree metrics are appropriately constructed. Experimental results on large-scale synthetic and real datasets from Wasserstein barycenter for documents with word embedding, multilevel clustering, and scalable Bayes problems show the advantages of tree-sliced Wasserstein barycenter over (Sinkhorn) Wasserstein barycenter.

Perturbations are not Enough: Generating Adversarial Examples with Spatial Distortions

Deep neural network image classifiers are reported to be susceptible to adversarial evasion attacks, which use carefully crafted images created to mislead a classifier. Recently, various kinds of adversarial attack methods have been proposed, most of which focus on adding small perturbations to input images. Despite the success of existing approaches, the way to generate realistic adversarial images with small perturbations remains a challenging problem. In this paper, we aim to address this problem by proposing a novel adversarial method, which generates adversarial examples by imposing not only perturbations but also spatial distortions on input images, including scaling, rotation, shear, and translation. As humans are less susceptible to small spatial distortions, the proposed approach can produce visually more realistic attacks with smaller perturbations, able to deceive classifiers without affecting human predictions. We learn our method by amortized techniques with neural networks and generate adversarial examples efficiently by a forward pass of the networks. Extensive experiments on attacking different types of non-robustified classifiers and robust classifiers with defence show that our method has state-of-the-art performance in comparison with advanced attack parallels.

Unsupervised Universal Self-Attention Network for Graph Classification

Existing graph embedding models often have weaknesses in exploiting graph structure similarities, potential dependencies among nodes and global network properties. To this end, we present U2GAN, a novel unsupervised model leveraging on the strength of the recently introduced universal self-attention network (Dehghani et al., 2019), to learn low-dimensional embeddings of graphs which can be used for graph classification. In particular, given an input graph, U2GAN first applies a self-attention computation, which is then followed by a recurrent transition to iteratively memorize its attention on vector representations of each node and its neighbors across each iteration. Thus, U2GAN can address the weaknesses in the existing models in order to produce plausible node embeddings whose sum is the final embedding of the whole graph. Experimental results show that our unsupervised U2GAN produces new state-of-the-art performances on a range of well-known benchmark datasets for the graph classification task. It even outperforms supervised methods in most of benchmark cases.