End-to-end automatic speech recognition (ASR), unlike conventional ASR, does not have modules to learn the semantic representation from speech encoder. Moreover, the higher frame-rate of speech representation prevents the model to learn the semantic representation properly. Therefore, the models that are constructed by the lower frame-rate of speech encoder lead to better performance. For Transformer-based ASR, the lower frame-rate is not only important for learning better semantic representation but also for reducing the computational complexity due to the self-attention mechanism which has O(n^2) order of complexity in both training and inference. In this paper, we propose a Transformer-based ASR model with the time reduction layer, in which we incorporate time reduction layer inside transformer encoder layers in addition to traditional sub-sampling methods to input features that further reduce the frame-rate. This can help in reducing the computational cost of the self-attention process for training and inference with performance improvement. Moreover, we introduce a fine-tuning approach for pre-trained ASR models using self-knowledge distillation (S-KD) which further improves the performance of our ASR model. Experiments on LibriSpeech datasets show that our proposed methods outperform all other Transformer-based ASR systems. Furthermore, with language model (LM) fusion, we achieve new state-of-the-art word error rate (WER) results for Transformer-based ASR models with just 30 million parameters trained without any external data.

Recent works in Generative Adversarial Networks (GANs) are actively revisiting various data augmentation techniques as an effective way to prevent discriminator overfitting. It is still unclear, however, that which augmentations could actually improve GANs, and in particular, how to apply a wider range of augmentations in training. In this paper, we propose a novel way to address these questions by incorporating a recent contrastive representation learning scheme into the GAN discriminator, coined ContraD. This "fusion" enables the discriminators to work with much stronger augmentations without increasing their training instability, thereby preventing the discriminator overfitting issue in GANs more effectively. Even better, we observe that the contrastive learning itself also benefits from our GAN training, i.e., by maintaining discriminative features between real and fake samples, suggesting a strong coherence between the two worlds: good contrastive representations are also good for GAN discriminators, and vice versa. Our experimental results show that GANs with ContraD consistently improve FID and IS compared to other recent techniques incorporating data augmentations, still maintaining highly discriminative features in the discriminator in terms of the linear evaluation. Finally, as a byproduct, we also show that our GANs trained in an unsupervised manner (without labels) can induce many conditional generative models via a simple latent sampling, leveraging the learned features of ContraD. Generative adversarial networks (GANs) (Goodfellow et al., 2014) have become one of the most prominent approaches for generative modeling with a wide range of applications (Ho & Ermon, 2016; Zhu et al., 2017; Karras et al., 2019; Rott Shaham et al., 2019). In general, a GAN is defined by a minimax game between two neural networks: a generator network that maps a random vector into the data domain, and a discriminator network that classifies whether a given sample is real (from the training dataset) or fake (from the generator).

Natural Human-Robot Interaction (HRI) is one of the key components for service robots to be able to work in human-centric environments. In such dynamic environments, the robot needs to understand the intention of the user to accomplish a task successfully. Towards addressing this point, we propose a software architecture that segments a target object from a crowded scene, indicated verbally by a human user. At the core of our system, we employ a multi-modal deep neural network for visual grounding. Unlike most grounding methods that tackle the challenge using pre-trained object detectors via a two-stepped process, we develop a single stage zero-shot model that is able to provide predictions in unseen data. We evaluate the performance of the proposed model on real RGB-D data collected from public scene datasets. Experimental results showed that the proposed model performs well in terms of accuracy and speed, while showcasing robustness to variation in the natural language input.

Cyber attacks pose crucial threats to computer system security, and put digital treasuries at excessive risks. This leads to an urgent call for an effective intrusion detection system that can identify the intrusion attacks with high accuracy. It is challenging to classify the intrusion events due to the wide variety of attacks. Furthermore, in a normal network environment, a majority of the connections are initiated by benign behaviors. The class imbalance issue in intrusion detection forces the classifier to be biased toward the majority/benign class, thus leave many attack incidents undetected. Spurred by the success of deep neural networks in computer vision and natural language processing, in this paper, we design a new system named DeepIDEA that takes full advantage of deep learning to enable intrusion detection and classification. To achieve high detection accuracy on imbalanced data, we design a novel attack-sharing loss function that can effectively move the decision boundary towards the attack classes and eliminates the bias towards the majority/benign class. By using this loss function, DeepIDEA respects the fact that the intrusion mis-classification should receive higher penalty than the attack mis-classification. Extensive experimental results on three benchmark datasets demonstrate the high detection accuracy of DeepIDEA. In particular, compared with eight state-of-the-art approaches, DeepIDEA always provides the best class-balanced accuracy.

Real-time surgical phase recognition is a fundamental task in modern operating rooms. Previous works tackle this task relying on architectures arranged in spatio-temporal order, however, the supportive benefits of intermediate spatial features are not considered. In this paper, we introduce, for the first time in surgical workflow analysis, Transformer to reconsider the ignored complementary effects of spatial and temporal features for accurate surgical phase recognition. Our hybrid embedding aggregation Transformer fuses cleverly designed spatial and temporal embeddings by allowing for active queries based on spatial information from temporal embedding sequences. More importantly, our framework is lightweight and processes the hybrid embeddings in parallel to achieve a high inference speed. Our method is thoroughly validated on two large surgical video datasets, i.e., Cholec80 and M2CAI16 Challenge datasets, and significantly outperforms the state-of-the-art approaches at a processing speed of 91 fps.

Machine learning on sets towards sequential output is an important and ubiquitous task, with applications ranging from language modelling and meta-learning to multi-agent strategy games and power grid optimization. Combining elements of representation learning and structured prediction, its two primary challenges include obtaining a meaningful, permutation invariant set representation and subsequently utilizing this representation to output a complex target permutation. This paper provides a comprehensive introduction to the field as well as an overview of important machine learning methods tackling both of these key challenges, with a detailed qualitative comparison of selected model architectures.

In modern litigation, fraud investigators often face an overwhelming number of documents that must be reviewed throughout a matter. In the majority of legal cases, fraud investigators do not know beforehand, exactly what they are looking for, nor where to find it. In addition, fraudsters may use deception to hide their behaviour and intentions by using code words. Effectively, this means fraud investigators are looking for a needle in the haystack without knowing what the needle looks like. As part of a larger research program, we use a framework to expedite the investigation process applying text-mining and machine learning techniques. We structure this framework using three well-known methods in fraud investigations: (i) the fraud triangle (ii) the golden ("W") investigation questions, and (iii) the analysis of competing hypotheses. With this framework, it is possible to automatically organize investigative data, so it is easier for investigators to find answers to typical investigative questions. In this research, we focus on one of the components of this framework: the identification of the usage of code words by fraudsters. Here for, a novel (annotated) synthetic data set is created containing such code words, hidden in normal email communication. Subsequently, a range of machine learning techniques are employed to detect such code words. We show that the state-of-the-art BERT model significantly outperforms other methods on this task. With this result, we demonstrate that deep neural language models can reliably (F1 score of 0.9) be applied in fraud investigations for the detection of code words.

Multilingual models have demonstrated impressive cross-lingual transfer performance. However, test sets like XNLI are monolingual at the example level. In multilingual communities, it is common for polyglots to code-mix when conversing with each other. Inspired by this phenomenon, we present two strong black-box adversarial attacks (one word-level, one phrase-level) for multilingual models that push their ability to handle code-mixed sentences to the limit. The former uses bilingual dictionaries to propose perturbations and translations of the clean example for sense disambiguation. The latter directly aligns the clean example with its translations before extracting phrases as perturbations. Our phrase-level attack has a success rate of 89.75% against XLM-R-large, bringing its average accuracy of 79.85 down to 8.18 on XNLI. Finally, we propose an efficient adversarial training scheme that trains in the same number of steps as the original model and show that it improves model accuracy.

Two of the main challenges for cropland classification by satellite time-series images are insufficient ground-truth data and inaccessibility of high-quality hyperspectral images for under-developed areas. Unlabeled medium-resolution satellite images are abundant, but how to benefit from them is an open question. We will show how to leverage their potential for cropland classification using self-supervised tasks. Self-supervision is an approach where we provide simple training signals for the samples, which are apparent from the data's structure. Hence, they are cheap to acquire and explain a simple concept about the data. We introduce three self-supervised tasks for cropland classification. They reduce epistemic uncertainty, and the resulting model shows superior accuracy in a wide range of settings compared to SVM and Random Forest. Subsequently, we use the self-supervised tasks to perform unsupervised domain adaptation and benefit from the labeled samples in other regions. It is crucial to know what information to transfer to avoid degrading the performance. We show how to automate the information selection and transfer process in cropland classification even when the source and target areas have a very different feature distribution. We improved the model by about 24% compared to a baseline architecture without any labeled sample in the target domain. Our method is amenable to gradual improvement, works with medium-resolution satellite images, and does not require complicated models. Code and data are available.

Code completion has become an essential component of integrated development environments. Contemporary code completion methods rely on the abstract syntax tree (AST) to generate syntactically correct code. However, they cannot fully capture the sequential and repetitive patterns of writing code and the structural information of the AST. To alleviate these problems, we propose a new code completion approach named CCAG, which models the flattened sequence of a partial AST as an AST graph. CCAG uses our proposed AST Graph Attention Block to capture different dependencies in the AST graph for representation learning in code completion. The sub-tasks of code completion are optimized via multi-task learning in CCAG, and the task balance is automatically achieved using uncertainty without the need to tune task weights. The experimental results show that CCAG has superior performance than state-of-the-art approaches and it is able to provide intelligent code completion.