Qian Chen

Jiaqing Liu, Chong Deng, Qinglin Zhang et al.

Automatic Speech Recognition (ASR) transcripts exhibit recognition errors and various spoken language phenomena such as disfluencies, ungrammatical sentences, and incomplete sentences, hence suffering from poor readability. To improve readability, we propose a Contextualized Spoken-to-Written conversion (CoS2W) task to address ASR and grammar errors and also transfer the informal text into the formal style with content preserved, utilizing contexts and auxiliary information. This task naturally matches the in-context learning capabilities of Large Language Models (LLMs). To facilitate comprehensive comparisons of various LLMs, we construct a document-level Spoken-to-Written conversion of ASR Transcripts Benchmark (SWAB) dataset. Using SWAB, we study the impact of different granularity levels on the CoS2W performance, and propose methods to exploit contexts and auxiliary information to enhance the outputs. Experimental results reveal that LLMs have the potential to excel in the CoS2W task, particularly in grammaticality and formality, our methods achieve effective understanding of contexts and auxiliary information by LLMs. We further investigate the effectiveness of using LLMs as evaluators and find that LLM evaluators show strong correlations with human evaluations on rankings of faithfulness and formality, which validates the reliability of LLM evaluators for the CoS2W task.

Qian Chen, Wen Wang, Qinglin Zhang

In the traditional cascading architecture for spoken language understanding (SLU), it has been observed that automatic speech recognition errors could be detrimental to the performance of natural language understanding. End-to-end (E2E) SLU models have been proposed to directly map speech input to desired semantic frame with a single model, hence mitigating ASR error propagation. Recently, pre-training technologies have been explored for these E2E models. In this paper, we propose a novel joint textual-phonetic pre-training approach for learning spoken language representations, aiming at exploring the full potentials of phonetic information to improve SLU robustness to ASR errors. We explore phoneme labels as high-level speech features, and design and compare pre-training tasks based on conditional masked language model objectives and inter-sentence relation objectives. We also investigate the efficacy of combining textual and phonetic information during fine-tuning. Experimental results on spoken language understanding benchmarks, Fluent Speech Commands and SNIPS, show that the proposed approach significantly outperforms strong baseline models and improves robustness of spoken language understanding to ASR errors.

Chen Qian, Yunhai Xiong, Xiang Chen

The prosperity of computer vision (CV) and natural language procession (NLP) in recent years has spurred the development of deep learning in many other domains. The advancement in machine learning provides us with an alternative option besides the computationally expensive density functional theories (DFT). Kernel method and graph neural networks have been widely studied as two mainstream methods for property prediction. The promising graph neural networks have achieved comparable accuracy to the DFT method for specific objects in the recent study. However, most of the graph neural networks with high precision so far require fully connected graphs with pairwise distance distribution as edge information. In this work, we shed light on the Directed Graph Attention Neural Network (DGANN), which only takes chemical bonds as edges and operates on bonds and atoms of molecules. DGANN distinguishes from previous models with those features: (1) It learns the local chemical environment encoding by graph attention mechanism on chemical bonds. Every initial edge message only flows into every message passing trajectory once. (2) The transformer blocks aggregate the global molecular representation from the local atomic encoding. (3) The position vectors and coordinates are used as inputs instead of distances. Our model has matched or outperformed most baseline graph neural networks on QM9 datasets even without thorough hyper-parameters searching. Moreover, this work suggests that models directly utilizing 3D coordinates can still reach high accuracies for molecule representation even without rotational and translational invariance incorporated.

Qian Chen, Sheikh Rabiul Islam, Henry Haswell et al.

Security operation centers (SOCs) typically use a variety of tools to collect large volumes of host logs for detection and forensic of intrusions. Our experience, supported by recent user studies on SOC operators, indicates that operators spend ample time (e.g., hundreds of man-hours) on investigations into logs seeking adversarial actions. Similarly, reconfiguration of tools to adapt detectors for future similar attacks is commonplace upon gaining novel insights (e.g., through internal investigation or shared indicators). This paper presents an automated malware pattern-extraction and early detection tool, testing three machine learning approaches: TF-IDF (term frequency-inverse document frequency), Fisher's LDA (linear discriminant analysis) and ET (extra trees/extremely randomized trees) that can (1) analyze freshly discovered malware samples in sandboxes and generate dynamic analysis reports (host logs); (2) automatically extract the sequence of events induced by malware given a large volume of ambient (un-attacked) host logs, and the relatively few logs from hosts that are infected with potentially polymorphic malware; (3) rank the most discriminating features (unique patterns) of malware and from the learned behavior detect malicious activity; and (4) allows operators to visualize the discriminating features and their correlations to facilitate malware forensic efforts. To validate the accuracy and efficiency of our tool, we design three experiments and test seven ransomware attacks (i.e., WannaCry, DBGer, Cerber, Defray, GandCrab, Locky, and nRansom). The experimental results show that TF-IDF is the best of the three methods to identify discriminating features, and ET is the most time-efficient and robust approach.

Qian Chen, Xiaodan Zhu, Zhenhua Ling et al.

Distributed representation learned with neural networks has recently shown to be effective in modeling natural languages at fine granularities such as words, phrases, and even sentences. Whether and how such an approach can be extended to help model larger spans of text, e.g., documents, is intriguing, and further investigation would still be desirable. This paper aims to enhance neural network models for such a purpose. A typical problem of document-level modeling is automatic summarization, which aims to model documents in order to generate summaries. In this paper, we propose neural models to train computers not just to pay attention to specific regions and content of input documents with attention models, but also distract them to traverse between different content of a document so as to better grasp the overall meaning for summarization. Without engineering any features, we train the models on two large datasets. The models achieve the state-of-the-art performance, and they significantly benefit from the distraction modeling, particularly when input documents are long.

Qian Chen, Xiaodan Zhu, Zhenhua Ling et al.

Reasoning and inference are central to human and artificial intelligence. Modeling inference in human language is very challenging. With the availability of large annotated data (Bowman et al., 2015), it has recently become feasible to train neural network based inference models, which have shown to be very effective. In this paper, we present a new state-of-the-art result, achieving the accuracy of 88.6% on the Stanford Natural Language Inference Dataset. Unlike the previous top models that use very complicated network architectures, we first demonstrate that carefully designing sequential inference models based on chain LSTMs can outperform all previous models. Based on this, we further show that by explicitly considering recursive architectures in both local inference modeling and inference composition, we achieve additional improvement. Particularly, incorporating syntactic parsing information contributes to our best result---it further improves the performance even when added to the already very strong model.