arXiv subjects

Bai Li

Publications and source records attributed to Bai Li.

At least 37 records · Page 2Linked to original sources

arXiv2020

Towards Understanding Fast Adversarial Training

Current neural-network-based classifiers are susceptible to adversarial examples. The most empirically successful approach to defending against such adversarial examples is adversarial training, which incorporates a strong self-attack during training to enhance its robustness. This approach, however, is computationally expensive and hence is hard to scale up. A recent work, called fast adversarial training, has shown that it is possible to markedly reduce computation time without sacrificing significant performance. This approach incorporates simple self-attacks, yet it can only run for a limited number of training epochs, resulting in sub-optimal performance. In this paper, we conduct experiments to understand the behavior of fast adversarial training and show the key to its success is the ability to recover from overfitting to weak attacks. We then extend our findings to improve fast adversarial training, demonstrating superior robust accuracy to strong adversarial training, with much-reduced training time.

cs.LG↗

arXiv2020

Representation Learning for Discovering Phonemic Tone Contours

Tone is a prosodic feature used to distinguish words in many languages, some of which are endangered and scarcely documented. In this work, we use unsupervised representation learning to identify probable clusters of syllables that share the same phonemic tone. Our method extracts the pitch for each syllable, then trains a convolutional autoencoder to learn a low dimensional representation for each contour. We then apply the mean shift algorithm to cluster tones in high-density regions of the latent space. Furthermore, by feeding the centers of each cluster into the decoder, we produce a prototypical contour that represents each cluster. We apply this method to spoken multi-syllable words in Mandarin Chinese and Cantonese and evaluate how closely our clusters match the ground truth tone categories. Finally, we discuss some difficulties with our approach, including contextual tone variation and allophony effects.

cs.SD↗

arXiv2020

Autonomous Last-mile Delivery Vehicles in Complex Traffic Environments

E-commerce has evolved with the digital technology revolution over the years. Last-mile logistics service contributes a significant part of the e-commerce experience. In contrast to the traditional last-mile logistics services, smart logistics service with autonomous driving technologies provides a promising solution to reduce the delivery cost and to improve efficiency. However, the traffic conditions in complex traffic environments, such as those in China, are more challenging compared to those in well-developed countries. Many types of moving objects (such as pedestrians, bicycles, electric bicycles, and motorcycles, etc.) share the road with autonomous vehicles, and their behaviors are not easy to track and predict. This paper introduces a technical solution from JD.com, a leading E-commerce company in China, to the autonomous last-mile delivery in complex traffic environments. Concretely, the methodologies in each module of our autonomous vehicles are presented, together with safety guarantee strategies. Up to this point, JD.com has deployed more than 300 self-driving vehicles for trial operations in tens of provinces of China, with an accumulated 715,819 miles and up to millions of on-road testing hours.

cs.RO↗

arXiv2020

Towards Practical Lottery Ticket Hypothesis for Adversarial Training

Recent research has proposed the lottery ticket hypothesis, suggesting that for a deep neural network, there exist trainable sub-networks performing equally or better than the original model with commensurate training steps. While this discovery is insightful, finding proper sub-networks requires iterative training and pruning. The high cost incurred limits the applications of the lottery ticket hypothesis. We show there exists a subset of the aforementioned sub-networks that converge significantly faster during the training process and thus can mitigate the cost issue. We conduct extensive experiments to show such sub-networks consistently exist across various model structures for a restrictive setting of hyperparameters ($e.g.$, carefully selected learning rate, pruning ratio, and model capacity). As a practical application of our findings, we demonstrate that such sub-networks can help in cutting down the total time of adversarial training, a standard approach to improve robustness, by up to 49\% on CIFAR-10 to achieve the state-of-the-art robustness.

cs.LG↗

arXiv2019

Enhancing Cross-task Black-Box Transferability of Adversarial Examples with Dispersion Reduction

Neural networks are known to be vulnerable to carefully crafted adversarial examples, and these malicious samples often transfer, i.e., they remain adversarial even against other models. Although great efforts have been delved into the transferability across models, surprisingly, less attention has been paid to the cross-task transferability, which represents the real-world cybercriminal's situation, where an ensemble of different defense/detection mechanisms need to be evaded all at once. In this paper, we investigate the transferability of adversarial examples across a wide range of real-world computer vision tasks, including image classification, object detection, semantic segmentation, explicit content detection, and text detection. Our proposed attack minimizes the ``dispersion'' of the internal feature map, which overcomes existing attacks' limitation of requiring task-specific loss functions and/or probing a target model. We conduct evaluation on open source detection and segmentation models as well as four different computer vision tasks provided by Google Cloud Vision (GCV) APIs, to show how our approach outperforms existing attacks by degrading performance of multiple CV tasks by a large margin with only modest perturbations linf=16.

eess.IV↗

arXiv2019

Graph-Driven Generative Models for Heterogeneous Multi-Task Learning

We propose a novel graph-driven generative model, that unifies multiple heterogeneous learning tasks into the same framework. The proposed model is based on the fact that heterogeneous learning tasks, which correspond to different generative processes, often rely on data with a shared graph structure. Accordingly, our model combines a graph convolutional network (GCN) with multiple variational autoencoders, thus embedding the nodes of the graph i.e., samples for the tasks) in a uniform manner while specializing their organization and usage to different tasks. With a focus on healthcare applications (tasks), including clinical topic modeling, procedure recommendation and admission-type prediction, we demonstrate that our method successfully leverages information across different tasks, boosting performance in all tasks and outperforming existing state-of-the-art approaches.

cs.LG↗

arXiv2019

Certified Adversarial Robustness with Additive Noise

The existence of adversarial data examples has drawn significant attention in the deep-learning community; such data are seemingly minimally perturbed relative to the original data, but lead to very different outputs from a deep-learning algorithm. Although a significant body of work on developing defensive models has been considered, most such models are heuristic and are often vulnerable to adaptive attacks. Defensive methods that provide theoretical robustness guarantees have been studied intensively, yet most fail to obtain non-trivial robustness when a large-scale model and data are present. To address these limitations, we introduce a framework that is scalable and provides certified bounds on the norm of the input manipulation for constructing adversarial examples. We establish a connection between robustness against adversarial perturbation and additive random noise, and propose a training strategy that can significantly improve the certified bounds. Our evaluation on MNIST, CIFAR-10 and ImageNet suggests that the proposed method is scalable to complicated models and large data sets, while providing competitive robustness to state-of-the-art provable defense methods.

cs.LG↗

arXiv2019

Distributed Recursive Filtering for Spatially Interconnected Systems with Randomly Occurred Missing Measurements

This paper proposed a distributed filter for spatially interconnected systems (SISs), which considers missing measurements in the sensors of sub-systems. An SIS is established by many similar sub-systems that directly interact or communicate with connective neighbors. Despite that the interactions are simple and tractable, the overall SIS can perform rich and complex behaviors. In actual projects, sensors of sub-systems in a sensor network may break down sometimes, which causes parts of the measurements unavailable unexpectedly. In this work, distributed characteristics of SISs are described by Andrea model and the losses of measurements are assumed to occur with known probabilities. Experimental results confirm that, this filtering method can be effectively employed for the state estimation of SISs, when missing measurements occur.

eess.SY↗

arXiv2019

Trajectory Planning for Autonomous Parking in Complex Environments: A Tunnel-based Optimal Control Approach

This paper proposes a fast and accurate trajectory planning algorithm for autonomous parking. Nominally, an optimal control problem should be formulated to describe this scheme, but the dimensionality of the optimal control problem is usually large, because the vehicle needs to avoid collision with every obstacle at every moment during the entire dynamic process. Although an initial guess obtained by a sample-and-search based planner facilitates the numerical optimization process, it is still far from being as fast as real-time. To address this issue, we replace all of the collision-avoidance constraints by series of within-tunnel conditions. Concretely, we develop a tunnel-based strategy such that the vehicle is restricted to move within the tunnels which naturally separate the vehicle from the obstacles. Unification, efficiency, and robustness of the proposed trajectory planning method have been verified by simulations.

cs.RO↗

arXiv2019

Real-world Conversational AI for Hotel Bookings

In this paper, we present a real-world conversational AI system to search for and book hotels through text messaging. Our architecture consists of a frame-based dialogue management system, which calls machine learning models for intent classification, named entity recognition, and information retrieval subtasks. Our chatbot has been deployed on a commercial scale, handling tens of thousands of hotel searches every day. We describe the various opportunities and challenges of developing a chatbot in the travel industry.

cs.LG↗

arXiv2019

Detecting dementia in Mandarin Chinese using transfer learning from a parallel corpus

Machine learning has shown promise for automatic detection of Alzheimer's disease (AD) through speech; however, efforts are hampered by a scarcity of data, especially in languages other than English. We propose a method to learn a correspondence between independently engineered lexicosyntactic features in two languages, using a large parallel corpus of out-of-domain movie dialogue data. We apply it to dementia detection in Mandarin Chinese, and demonstrate that our method outperforms both unilingual and machine translation-based baselines. This appears to be the first study that transfers feature domains in detecting cognitive decline.

cs.CL↗

arXiv2019

On Norm-Agnostic Robustness of Adversarial Training

Adversarial examples are carefully perturbed in-puts for fooling machine learning models. A well-acknowledged defense method against such examples is adversarial training, where adversarial examples are injected into training data to increase robustness. In this paper, we propose a new attack to unveil an undesired property of the state-of-the-art adversarial training, that is it fails to obtain robustness against perturbations in $\ell_2$ and $\ell_\infty$ norms simultaneously. We discuss a possible solution to this issue and its limitations as well.

cs.LG↗

arXiv2019

Improving Sequence-to-Sequence Learning via Optimal Transport

Sequence-to-sequence models are commonly trained via maximum likelihood estimation (MLE). However, standard MLE training considers a word-level objective, predicting the next word given the previous ground-truth partial sentence. This procedure focuses on modeling local syntactic patterns, and may fail to capture long-range semantic structure. We present a novel solution to alleviate these issues. Our approach imposes global sequence-level guidance via new supervision based on optimal transport, enabling the overall characterization and preservation of semantic features. We further show that this method can be understood as a Wasserstein gradient flow trying to match our model to the ground truth sequence distribution. Extensive experiments are conducted to validate the utility of the proposed approach, showing consistent improvements over a wide variety of NLP tasks, including machine translation, abstractive text summarization, and image captioning.

cs.CL↗

arXiv2018

State Complexity of Overlap Assembly

The \emph{state complexity} of a regular language $L_m$ is the number $m$ of states in a minimal deterministic finite automaton (DFA) accepting $L_m$. The state complexity of a regularity-preserving binary operation on regular languages is defined as the maximal state complexity of the result of the operation where the two operands range over all languages of state complexities $\le m$ and $\le n$, respectively. We find a tight upper bound on the state complexity of the binary operation \emph{overlap assembly} on regular languages. This operation was introduced by Csuhaj-Varjú, Petre, and Vaszil to model the process of self-assembly of two linear DNA strands into a longer DNA strand, provided that their ends "overlap". We prove that the state complexity of the overlap assembly of languages $L_m$ and $L_n$, where $m\ge 2$ and $n\ge1$, is at most $2 (m-1) 3^{n-1} + 2^n$. Moreover, for $m \ge 2$ and $n \ge 3$ there exist languages $L_m$ and $L_n$ over an alphabet of size $n$ whose overlap assembly meets the upper bound and this bound cannot be met with smaller alphabets. Finally, we prove that $m+n$ is a tight upper bound on the overlap assembly of unary languages, and that there are binary languages whose overlap assembly has exponential state complexity at least $m(2^{n-1}-2)+2$.

cs.FL↗

arXiv2018

Dropout during inference as a model for neurological degeneration in an image captioning network

We replicate a variation of the image captioning architecture by Vinyals et al. (2015), then introduce dropout during inference mode to simulate the effects of neurodegenerative diseases like Alzheimer's disease (AD) and Wernicke's aphasia (WA). We evaluate the effects of dropout on language production by measuring the KL-divergence of word frequency distributions and other linguistic metrics as dropout is added. We find that the generated sentences most closely approximate the word frequency distribution of the training corpus when using a moderate dropout of 0.4 during inference.

cs.CL↗

arXiv2018

A Unified Particle-Optimization Framework for Scalable Bayesian Sampling

There has been recent interest in developing scalable Bayesian sampling methods such as stochastic gradient MCMC (SG-MCMC) and Stein variational gradient descent (SVGD) for big-data analysis. A standard SG-MCMC algorithm simulates samples from a discrete-time Markov chain to approximate a target distribution, thus samples could be highly correlated, an undesired property for SG-MCMC. In contrary, SVGD directly optimizes a set of particles to approximate a target distribution, and thus is able to obtain good approximations with relatively much fewer samples. In this paper, we propose a principle particle-optimization framework based on Wasserstein gradient flows to unify SG-MCMC and SVGD, and to allow new algorithms to be developed. Our framework interprets SG-MCMC as particle optimization on the space of probability measures, revealing a strong connection between SG-MCMC and SVGD. The key component of our framework is several particle-approximate techniques to efficiently solve the original partial differential equations on the space of probability measures. Extensive experiments on both synthetic data and deep neural networks demonstrate the effectiveness and efficiency of our framework for scalable Bayesian sampling.

stat.ML↗

arXiv2017

On Connecting Stochastic Gradient MCMC and Differential Privacy

Significant success has been realized recently on applying machine learning to real-world applications. There have also been corresponding concerns on the privacy of training data, which relates to data security and confidentiality issues. Differential privacy provides a principled and rigorous privacy guarantee on machine learning models. While it is common to design a model satisfying a required differential-privacy property by injecting noise, it is generally hard to balance the trade-off between privacy and utility. We show that stochastic gradient Markov chain Monte Carlo (SG-MCMC) -- a class of scalable Bayesian posterior sampling algorithms proposed recently -- satisfies strong differential privacy with carefully chosen step sizes. We develop theory on the performance of the proposed differentially-private SG-MCMC method. We conduct experiments to support our analysis and show that a standard SG-MCMC sampler without any modification (under a default setting) can reach state-of-the-art performance in terms of both privacy and utility on Bayesian learning.

stat.ML↗

arXiv2010

Non-Euclidean statistical analysis of covariance matrices and diffusion tensors

The statistical analysis of covariance matrices occurs in many important applications, e.g. in diffusion tensor imaging and longitudinal data analysis. We consider the situation where it is of interest to estimate an average covariance matrix, describe its anisotropy, to carry out principal geodesic analysis and to interpolate between covariance matrices. There are many choices of metric available, each with its advantages. The particular choice of what is best will depend on the particular application. The use of the Procrustes size-and-shape metric is particularly appropriate when the covariance matrices are close to being deficient in rank. We discuss the use of different metrics for diffusion tensor analysis, and we also introduce certain types of regularization for tensors.

stat.ME↗