Non-autoregressive models generate target words in a parallel way, which achieve a faster decoding speed but at the sacrifice of translation accuracy. To remedy a flawed translation by non-autoregressive models, a promising approach is to train a conditional masked translation model (CMTM), and refine the generated results within several iterations. Unfortunately, such approach hardly considers the \textit{sequential dependency} among target words, which inevitably results in a translation degradation. Hence, instead of solely training a Transformer-based CMTM, we propose a Self-Review Mechanism to infuse sequential information into it. Concretely, we insert a left-to-right mask to the same decoder of CMTM, and then induce it to autoregressively review whether each generated word from CMTM is supposed to be replaced or kept. The experimental results (WMT14 En$\leftrightarrow$De and WMT16 En$\leftrightarrow$Ro) demonstrate that our model uses dramatically less training computations than the typical CMTM, as well as outperforms several state-of-the-art non-autoregressive models by over 1 BLEU. Through knowledge distillation, our model even surpasses a typical left-to-right Transformer model, while significantly speeding up decoding.
This article presents a secure key exchange algorithm that exploits reciprocity in wireless channels to share a secret key between two nodes $A$ and $B$. Reciprocity implies that the channel phases in the links $A\rightarrow B$ and $B\rightarrow A$ are the same. A number of such reciprocal phase values are measured at nodes $A$ and $B$, called shared phase values hereafter. Each shared phase value is used to mask points of a Phase Shift Keying (PSK) constellation. Masking is achieved by rotating each PSK constellation with a shared phase value. Rotation of constellation is equivalent to adding phases modulo-$2\pi$, and as the channel phase is uniformly distributed in $[0,2\pi)$, the result of summation conveys zero information about summands. To enlarge the key size over a static or slow fading channel, the Radio Frequency (RF) propagation path is perturbed to create several independent realizations of multi-path fading, each used to share a new phase value. To eavesdrop a phase value shared in this manner, the Eavesdropper (Eve) will always face an under-determined system of linear equations which will not reveal any useful information about its actual solution value. This property is used to establish a secure key between two legitimate users.
Semantic concepts and relations encoded in domain-specific ontologies and other medical semantic resources play a crucial role in deciphering terms in medical queries and documents. The exploitation of these resources for tackling the semantic gap issue has been widely studied in the literature. However, there are challenges that hinder their widespread use in real-world applications. Among these challenges is the insufficient knowledge individually encoded in existing medical ontologies, which is magnified when users express their information needs using long-winded natural language queries. In this context, many of the users query terms are either unrecognized by the used ontologies, or cause retrieving false positives that degrade the quality of current medical information search approaches. In this article, we explore the combination of multiple extrinsic semantic resources in the development of a full-fledged medical information search framework to: i) highlight and expand head medical concepts in verbose medical queries (i.e. concepts among query terms that significantly contribute to the informativeness and intent of a given query), ii) build semantically enhanced inverted index documents, iii) contribute to a heuristical weighting technique in the query document matching process. To demonstrate the effectiveness of the proposed approach, we conducted several experiments over the CLEF eHealth 2014 dataset. Findings indicate that the proposed method combining several extrinsic semantic resources proved to be more effective than related approaches in terms of precision measure.
Acronyms are abbreviated units of a phrase constructed by using initial components of the phrase in a text. Automatic extraction of acronyms from a text can help various Natural Language Processing tasks like machine translation, information retrieval, and text summarisation. This paper discusses an ensemble approach for the task of Acronym Extraction, which utilises two different methods to extract acronyms and their corresponding long forms. The first method utilises a multilingual contextual language model and fine-tunes the model to perform the task. The second method relies on a convolutional neural network architecture to extract acronyms and append them to the output of the previous method. We also augment the official training dataset with additional training samples extracted from several open-access journals to help improve the task performance. Our dataset analysis also highlights the noise within the current task dataset. Our approach achieves the following macro-F1 scores on test data released with the task: Danish (0.74), English-Legal (0.72), English-Scientific (0.73), French (0.63), Persian (0.57), Spanish (0.65), Vietnamese (0.65). We release our code and models publicly.
Unsupervised Domain Adaptation (UDA) aims to transfer knowledge from a labeled source domain to an unlabeled target domain. Most existing UDA approaches enable knowledge transfer via learning domain-invariant representation and sharing one classifier across two domains. However, ignoring the domain-specific information that are related to the task, and forcing a unified classifier to fit both domains will limit the feature expressiveness in each domain. In this paper, by observing that the Transformer architecture with comparable parameters can generate more transferable representations than CNN counterparts, we propose a Win-Win TRansformer framework (WinTR) that separately explores the domain-specific knowledge for each domain and meanwhile interchanges cross-domain knowledge. Specifically, we learn two different mappings using two individual classification tokens in the Transformer, and design for each one a domain-specific classifier. The cross-domain knowledge is transferred via source guided label refinement and single-sided feature alignment with respect to source or target, which keeps the integrity of domain-specific information. Extensive experiments on three benchmark datasets show that our method outperforms the state-of-the-art UDA methods, validating the effectiveness of exploiting both domain-specific and invariant
Transliteration is a task of translating named entities from a language to another, based on phonetic similarity. The task has embraced deep learning approaches in recent years, yet, most ignore the phonetic features of the involved languages. In this work, we incorporate phonetic information into neural networks in two ways: we synthesize extra data using forward and back-translation but in a phonetic manner; and we pre-train models on a phonetic task before learning transliteration. Our experiments include three language pairs and six directions, namely English to and from Chinese, Hebrew and Thai. Results indicate that our proposed approach brings benefits to the model and achieves better or similar performance when compared to state of the art.
Motivated by the wide adoption of reinforcement learning (RL) in real-world personalized services, where users' sensitive and private information needs to be protected, we study regret minimization in finite-horizon Markov decision processes (MDPs) under the constraints of differential privacy (DP). Compared to existing private RL algorithms that work only on tabular finite-state, finite-actions MDPs, we take the first step towards privacy-preserving learning in MDPs with large state and action spaces. Specifically, we consider MDPs with linear function approximation (in particular linear mixture MDPs) under the notion of joint differential privacy (JDP), where the RL agent is responsible for protecting users' sensitive data. We design two private RL algorithms that are based on value iteration and policy optimization, respectively, and show that they enjoy sub-linear regret performance while guaranteeing privacy protection. Moreover, the regret bounds are independent of the number of states, and scale at most logarithmically with the number of actions, making the algorithms suitable for privacy protection in nowadays large-scale personalized services. Our results are achieved via a general procedure for learning in linear mixture MDPs under changing regularizers, which not only generalizes previous results for non-private learning, but also serves as a building block for general private reinforcement learning.
Adversarial training, the process of training a deep learning model with adversarial data, is one of the most successful adversarial defense methods for deep learning models. We have found that the robustness to white-box attack of an adversarially trained model can be further improved if we fine tune this model in inference stage to adapt to the adversarial input, with the extra information in it. We introduce an algorithm that "post trains" the model at inference stage between the original output class and a "neighbor" class, with existing training data. The accuracy of pre-trained Fast-FGSM CIFAR10 classifier base model against white-box projected gradient attack (PGD) can be significantly improved from 46.8% to 64.5% with our algorithm.
Deep metric learning aims to learn an embedding space where the distance between data reflects their class equivalence, even when their classes are unseen during training. However, the limited number of classes available in training precludes generalization of the learned embedding space. Motivated by this, we introduce a new data augmentation approach that synthesizes novel classes and their embedding vectors. Our approach can provide rich semantic information to an embedding model and improve its generalization by augmenting training data with novel classes unavailable in the original data. We implement this idea by learning and exploiting a conditional generative model, which, given a class label and a noise, produces a random embedding vector of the class. Our proposed generator allows the loss to use richer class relations by augmenting realistic and diverse classes, resulting in better generalization to unseen samples. Experimental results on public benchmark datasets demonstrate that our method clearly enhances the performance of proxy-based losses.
Representation knowledge distillation aims at transferring rich information from one model to another. Current approaches for representation distillation mainly focus on the direct minimization of distance metrics between the models' embedding vectors. Such direct methods may be limited in transferring high-order dependencies embedded in the representation vectors, or in handling the capacity gap between the teacher and student models. In this paper, we introduce two approaches for enhancing representation distillation using classifier sharing between the teacher and student. Specifically, we first show that connecting the teacher's classifier to the student backbone and freezing its parameters is beneficial for the process of representation distillation, yielding consistent improvements. Then, we propose an alternative approach that asks to tailor the teacher model to a student with limited capacity. This approach competes with and in some cases surpasses the first method. Via extensive experiments and analysis, we show the effectiveness of the proposed methods on various datasets and tasks, including image classification, fine-grained classification, and face verification. For example, we achieve state-of-the-art performance for face verification on the IJB-C dataset for a MobileFaceNet model: TAR@(FAR=1e-5)=93.7\%. Code is available at https://github.com/Alibaba-MIIL/HeadSharingKD.