Dialogue systems play an increasingly important role in various aspects of our daily life. It is evident from recent research that dialogue systems trained on human conversation data are biased. In particular, they can produce responses that reflect people's gender prejudice. Many debiasing methods have been developed for various natural language processing tasks, such as word embedding. However, they are not directly applicable to dialogue systems because they are likely to force dialogue models to generate similar responses for different genders. This greatly degrades the diversity of the generated responses and immensely hurts the performance of the dialogue models. In this paper, we propose a novel adversarial learning framework Debiased-Chat to train dialogue models free from gender bias while keeping their performance. Extensive experiments on two real-world conversation datasets show that our framework significantly reduces gender bias in dialogue models while maintaining the response quality.
Representation learning has been proven to play an important role in the unprecedented success of machine learning models in numerous tasks, such as machine translation, face recognition and recommendation. The majority of existing representation learning approaches often require a large number of consistent and noise-free labels. However, due to various reasons such as budget constraints and privacy concerns, labels are very limited in many real-world scenarios. Directly applying standard representation learning approaches on small labeled data sets will easily run into over-fitting problems and lead to sub-optimal solutions. Even worse, in some domains such as education, the limited labels are usually annotated by multiple workers with diverse expertise, which yields noises and inconsistency in such crowdsourcing settings. In this paper, we propose a novel framework which aims to learn effective representations from limited data with crowdsourced labels. Specifically, we design a grouping based deep neural network to learn embeddings from a limited number of training samples and present a Bayesian confidence estimator to capture the inconsistency among crowdsourced labels. Furthermore, to expedite the training process, we develop a hard example selection procedure to adaptively pick up training examples that are misclassified by the model. Extensive experiments conducted on three real-world data sets demonstrate the superiority of our framework on learning representations from limited data with crowdsourced labels, comparing with various state-of-the-art baselines. In addition, we provide a comprehensive analysis on each of the main components of our proposed framework and also introduce the promising results it achieved in our real production to fully understand the proposed framework.
Meta learning algorithms have been widely applied in many tasks for efficient learning, such as few-shot image classification and fast reinforcement learning. During meta training, the meta learner develops a common learning strategy, or experience, from a variety of learning tasks. Therefore, during meta test, the meta learner can use the learned strategy to quickly adapt to new tasks even with a few training samples. However, there is still a dark side about meta learning in terms of reliability and robustness. In particular, is meta learning vulnerable to adversarial attacks? In other words, would a well-trained meta learner utilize its learned experience to build wrong or likely useless knowledge, if an adversary unnoticeably manipulates the given training set? Without the understanding of this problem, it is extremely risky to apply meta learning in safety-critical applications. Thus, in this paper, we perform the initial study about adversarial attacks on meta learning under the few-shot classification problem. In particular, we formally define key elements of adversarial attacks unique to meta learning and propose the first attacking algorithm against meta learning under various settings. We evaluate the effectiveness of the proposed attacking strategy as well as the robustness of several representative meta learning algorithms. Experimental results demonstrate that the proposed attacking strategy can easily break the meta learner and meta learning is vulnerable to adversarial attacks. The implementation of the proposed framework will be released upon the acceptance of this paper.
Discrete event sequences are ubiquitous, such as an ordered event series of process interactions in Information and Communication Technology systems. Recent years have witnessed increasing efforts in detecting anomalies with discrete-event sequences. However, it still remains an extremely difficult task due to several intrinsic challenges including data imbalance issues, the discrete property of the events, and sequential nature of the data. To address these challenges, in this paper, we propose OC4Seq, a multi-scale one-class recurrent neural network for detecting anomalies in discrete event sequences. Specifically, OC4Seq integrates the anomaly detection objective with recurrent neural networks (RNNs) to embed the discrete event sequences into latent spaces, where anomalies can be easily detected. In addition, given that an anomalous sequence could be caused by either individual events, subsequences of events, or the whole sequence, we design a multi-scale RNN framework to capture different levels of sequential patterns simultaneously. Experimental results on three benchmark datasets show that OC4Seq consistently outperforms various representative baselines by a large margin. Moreover, through both quantitative and qualitative analysis, the importance of capturing multi-scale sequential patterns for event anomaly detection is verified.
Communication compression has been extensively adopted to speed up large-scale distributed optimization. However, most existing decentralized algorithms with compression are unsatisfactory in terms of convergence rate and stability. In this paper, we delineate two key obstacles in the algorithm design -- data heterogeneity and compression error. Our attempt to explicitly overcome these obstacles leads to a novel decentralized algorithm named LEAD. This algorithm is the first \underline{L}in\underline{EA}r convergent \underline{D}ecentralized algorithm with communication compression. Our theory describes the coupled dynamics of the inaccurate model propagation and optimization process. We also provide the first consensus error bound without assuming bounded gradients. Empirical experiments validate our theoretical analysis and show that the proposed algorithm achieves state-of-the-art computation and communication efficiency.
In recent years, Graph Convolutional Networks (GCNs) show competitive performance in different domains, such as social network analysis, recommendation, and smart city. However, training GCNs with insufficient supervision is very difficult. The performance of GCNs becomes unsatisfying with few labeled data. Although some pioneering work try to understand why GCNs work or fail, their analysis focus more on the entire model level. Profiling GCNs on different nodes is still underexplored. To address the limitations, we study GCNs with respect to the node degree distribution. We show that GCNs have a higher accuracy on nodes with larger degrees even if they are underrepresented in most graphs, with both empirical observation and theoretical proof. We then propose Self-Supervised-Learning Degree-Specific GCN (SL-DSGCN) which handles the degree-related biases of GCNs from model and data aspects. Firstly, we design a degree-specific GCN layer that models both discrepancies and similarities of nodes with different degrees, and reduces the inner model-aspect biases of GCNs caused by sharing the same parameters with all nodes. Secondly, we develop a self-supervised-learning algorithm that assigns pseudo labels with uncertainty scores on unlabeled nodes using a Bayesian neural network. Pseudo labels increase the chance of connecting to labeled neighbors for low-degree nodes, thus reducing the biases of GCNs from the data perspective. We further exploit uncertainty scores as dynamic weights on pseudo labels in the stochastic gradient descent for SL-DSGCN. We validate \ours on three benchmark datasets, and confirm SL-DSGCN not only outperforms state-of-the-art self-training/self-supervised-learning GCN methods, but also improves GCN accuracy dramatically for low-degree nodes.
The success of deep learning notoriously requires larger amounts of costly annotated data. This has led to the development of self-supervised learning (SSL) that aims to alleviate this limitation by creating domain specific pretext tasks on unlabeled data. Simultaneously, there are increasing interests in generalizing deep learning to the graph domain in the form of graph neural networks (GNNs). GNNs can naturally utilize unlabeled nodes through the simple neighborhood aggregation that is unable to thoroughly make use of unlabeled nodes. Thus, we seek to harness SSL for GNNs to fully exploit the unlabeled data. Different from data instances in the image and text domains, nodes in graphs present unique structure information and they are inherently linked indicating not independent and identically distributed (or i.i.d.). Such complexity is a double-edged sword for SSL on graphs. On the one hand, it determines that it is challenging to adopt solutions from the image and text domains to graphs and dedicated efforts are desired. On the other hand, it provides rich information that enables us to build SSL from a variety of perspectives. Thus, in this paper, we first deepen our understandings on when, why, and which strategies of SSL work with GNNs by empirically studying numerous basic SSL pretext tasks on graphs. Inspired by deep insights from the empirical studies, we propose a new direction SelfTask to build advanced pretext tasks that are able to achieve state-of-the-art performance on various real-world datasets. The specific experimental settings to reproduce our results can be found in \url{https://github.com/ChandlerBang/SelfTask-GNN}.
Graphs neural networks (GNNs) learn node features by aggregating and combining neighbor information, which have achieved promising performance on many graph tasks. However, GNNs are mostly treated as black-boxes and lack human intelligible explanations. Thus, they cannot be fully trusted and used in certain application domains if GNN models cannot be explained. In this work, we propose a novel approach, known as XGNN, to interpret GNNs at the model-level. Our approach can provide high-level insights and generic understanding of how GNNs work. In particular, we propose to explain GNNs by training a graph generator so that the generated graph patterns maximize a certain prediction of the model.We formulate the graph generation as a reinforcement learning task, where for each step, the graph generator predicts how to add an edge into the current graph. The graph generator is trained via a policy gradient method based on information from the trained GNNs. In addition, we incorporate several graph rules to encourage the generated graphs to be valid. Experimental results on both synthetic and real-world datasets show that our proposed methods help understand and verify the trained GNNs. Furthermore, our experimental results indicate that the generated graphs can provide guidance on how to improve the trained GNNs.
Recently, neural network based dialogue systems have become ubiquitous in our increasingly digitalized society. However, due to their inherent opaqueness, some recently raised concerns about using neural models are starting to be taken seriously. In fact, intentional or unintentional behaviors could lead to a dialogue system to generate inappropriate responses. Thus, in this paper, we investigate whether we can learn to craft input sentences that result in a black-box neural dialogue model being manipulated into having its outputs contain target words or match target sentences. We propose a reinforcement learning based model that can generate such desired inputs automatically. Extensive experiments on a popular well-trained state-of-the-art neural dialogue model show that our method can successfully seek out desired inputs that lead to the target outputs in a considerable portion of cases. Consequently, our work reveals the potential of neural dialogue models to be manipulated, which inspires and opens the door towards developing strategies to defend them.
Graph classification is an important task on graph-structured data with many real-world applications. The goal of graph classification task is to train a classifier using a set of training graphs. Recently, Graph Neural Networks (GNNs) have greatly advanced the task of graph classification. When building a GNN model for graph classification, the graphs in the training set are usually assumed to be identically distributed. However, in many real-world applications, graphs in the same dataset could have dramatically different structures, which indicates that these graphs are likely non-identically distributed. Therefore, in this paper, we aim to develop graph neural networks for graphs that are not non-identically distributed. Specifically, we propose a general non-IID graph neural network framework, i.e., Non-IID-GNN. Given a graph, Non-IID-GNN can adapt any existing graph neural network model to generate a sample-specific model for this graph. Comprehensive experiments on various graph classification benchmarks demonstrate the effectiveness of the proposed framework. We will release the code of the proposed framework upon the acceptance of the paper.