A5101905948- Topic Modeling
- Natural Language Processing Techniques
- Anomaly Detection Techniques and Applications
- Advanced Text Analysis Techniques
- Domain Adaptation and Few-Shot Learning
- Multimodal Machine Learning Applications
- Complex Network Analysis Techniques
- Advanced Neural Network Applications
- Advanced Graph Neural Networks
- Explainable Artificial Intelligence (XAI)
- Software System Performance and Reliability
- Sparse and Compressive Sensing Techniques
- Time Series Analysis and Forecasting
- Statistical Methods and Inference
- Text and Document Classification Technologies
- Face and Expression Recognition
- Music and Audio Processing
- Machine Learning and Data Classification
- Data Quality and Management
- Speech and Audio Processing
- Human Mobility and Location-Based Analysis
- Data Stream Mining Techniques
- Optical measurement and interference techniques
- Imbalanced Data Classification Techniques
- Semantic Web and Ontologies
Microsoft (United States)
2023-2024
Virginia Tech
2016-2023
Microsoft Research (United Kingdom)
2023
Princeton University
2020-2022
Xi'an Jiaotong University
2021-2022
NEC (United States)
2021-2022
China University of Mining and Technology
2022
State Nuclear Power Technology Company (China)
2021
Zhejiang University of Technology
2019-2020
Harbin Institute of Technology
2020
With the advance of sensor technologies, Multivariate Time Series classification (MTSC) problem, perhaps one most essential problems in time series data mining domain, has continuously received a significant amount attention recent decades. Traditional approaches based on Bag-of-Patterns or Shapelet have difficulty dealing with huge amounts feature candidates generated high-dimensional multivariate but promising performance even when training set is small. In contrast, deep learning methods...
Incident management for cloud services is a complex process involving several steps and has huge impact on both service health developer productivity. On-call engineers require significant amount of domain knowledge manual effort root causing mitigation production incidents. Recent advances in artificial intelligence resulted state-of-the-art large language models like GPT-3.x (both GPT-3.0 GPT-3.5), which have been used to solve variety problems ranging from question answering text...
Online purchase forecasting is of great importance in e-commerce platforms, which the basis how to present personalized interesting product lists individual customers. However, predicting online purchases not trivial as it influenced by many factors including: (i) complex temporal pattern with hierarchical inter-correlations; (ii) arbitrary category dependencies. To address these factors, we develop a Graph Multi-Scale Pyramid Networks (GMP) framework fully exploit users' latent behavioral...
Large language models (LLMs) have significantly advanced the field of natural processing (NLP), providing a highly useful, task-agnostic foundation for wide range applications. However, directly applying LLMs to solve sophisticated problems in specific domains meets many hurdles, caused by heterogeneity domain data, sophistication knowledge, uniqueness objectives, and diversity constraints (e.g., various social norms, cultural conformity, religious beliefs, ethical standards applications)....
Various contrastive learning approaches have been proposed in recent years and achieve significant empirical success. While effective prevalent, has less explored for time series data. A key component of is to select appropriate augmentations imposing some priors construct feasible positive samples, such that an encoder can be trained learn robust discriminative representations. Unlike image language domains where "desired'' augmented samples generated with the rule thumb guided by...
Hydrogen is a renewable energy source with various features, clean, carbon-free, high density, which being recognized internationally as “future energy.” The US, the EU, Japan, South Korea, China, and other countries or regions are gradually clarifying development position of hydrogen. rapid hydrogen industry requires more hydrogenation infrastructure to meet need fuel cell vehicles. Nevertheless, due frequent occurrence accidents, their safety has become an obstacle large-scale...
Recent Large Language Models (LLMs) have demonstrated satisfying general instruction following ability. However, small LLMs with about 7B parameters still struggle fine-grained format (e.g., JSON format), which seriously hinder the advancements of their applications. Most existing methods focus on benchmarking while overlook how to improve specific ability for LLMs. Besides, these often rely evaluations based advanced GPT-4), can introduce intrinsic bias and be costly due API calls. In this...
Forecasting on sparse multivariate time series (MTS) aims to model the predictors of future values given their incomplete past, which is important for many emerging applications. However, most existing methods process MTS’s individually, and do not leverage dynamic distributions underlying MTS’s, leading sub-optimal results when sparsity high. To address this challenge, we propose a novel generative model, tracks transition latent clusters, instead isolated feature representations, achieve...
It is critical and important to detect anomalies in event sequences, which becomes widely available many application domains. Indeed, various efforts have been made capture abnormal patterns from sequences through sequential pattern analysis or representation learning. However, existing approaches usually ignore the semantic information of content. To this end, paper, we propose a self-attentive encoder-decoder transformer framework, Content-Aware Transformer CAT, for anomaly detection...
Xuchao Zhang, Fanglan Chen, Chang-Tien Lu, Naren Ramakrishnan. Proceedings of the 2019 Conference North American Chapter Association for Computational Linguistics: Human Language Technologies, Volume 1 (Long and Short Papers). 2019.
Jianfeng He, Xuchao Zhang, Shuo Lei, Zhiqian Chen, Fanglan Abdulaziz Alhamadani, Bei Xiao, ChangTien Lu. Proceedings of the 2020 Conference on Empirical Methods in Natural Language Processing (EMNLP). 2020.
Question routing (QR) aims at recommending newly posted questions to the potential answerers who are most likely answer questions. The existing approaches that learn users' expertise from their past question-answering activities usually suffer challenges in two aspects: 1) multi-faceted and 2) temporal dynamics answering behavior. This paper proposes a novel context-aware model multiple granularities of concurrently address above challenges. Specifically, attention characterizes answerer's...
Meta-learning has emerged as a trending technique to tackle few-shot text classification and achieve state-of-the-art performance. However, the performance of existing approaches heavily depends on inter-class variance support set. As result, it can perform well tasks when semantics sampled classes are distinct while failing differentiate with similar semantics. In this paper, we propose novel Task-Adaptive Reference Transformation (TART) network, aiming enhance generalization by...
The success of training accurate models strongly depends on the availability a sufficient collection precisely labeled data. However, real-world datasets contain erroneously data samples that substantially hinder performance machine learning models. Meanwhile, well-labeled is usually expensive to obtain and only limited amount available for training. In this paper, we consider problem robust model by using large-scale noisy in conjunction with small set clean To leverage information...
Recent multilingual pre-trained language models have achieved remarkable zero-shot performance, where the model is only finetuned on one source and directly evaluated target languages. In this work, we propose a self-learning framework that further utilizes unlabeled data of languages, combined with uncertainty estimation in process to select high-quality silver labels. Three different uncertainties are adapted analyzed specifically for cross lingual transfer: Language...
Hyper-local pricing data, e.g., about foods and commodities, exhibit subtle spatiotemporal variations that can be useful as crucial precursors of future events. Three major challenges in modeling such data include: i) temporal dependencies underlying features; ii) missing values; iii) constraints economic phenomena. These hinder traditional event forecasting models from being applied effectively. This paper proposes a novel model concurrently addresses the above challenges. Specifically,...
For node level graph encoding, a recent important state-of-art method is the convolutional networks (GCN), which nicely integrate local vertex features and topology in spectral domain. However, current studies suffer from several drawbacks: (1) CNNs rely on Chebyshev polynomial approximation results oscillatory at jump discontinuities; (2) Increasing order of can reduce oscillations issue, but also incurs unaffordable computational cost; (3) polynomials require degree Ω(poly(1/ε)) to...
Storyline detection aims to connect seemly irrelevant single documents into meaningful chains, which provides opportunities for understanding how events evolve over time and what triggers such evolutions. Most previous work generated the storylines through unsupervised methods that can hardly reveal underlying factors driving evolution process. This paper introduces a Bayesian model generate from massive infer corresponding hidden relations topics. In addition, our is first attempt utilizes...
Sound Event Early Detection (SEED) is an essential task in recognizing the acoustic environments and soundscapes. However, most of existing methods focus on offline sound event detection, which suffers from over-confidence issue early-stage detection usually yield unreliable results. To solve problem, we propose a novel Polyphonic Evidential Neural Network (PENet) to model evidential uncertainty class probability with Beta distribution. Specifically, use distribution probabilities, enriches...
Deriving event storylines is an effective summarization method to succinctly organize extensive information, which can significantly alleviate the pain of information overload. The critical challenge lack widely recognized definition storyline metric. Prior studies have developed various approaches based on different assumptions about users' interests. These works extract interesting patterns, but their do not guarantee that derived patterns will match preference. On other hand,...
With the increase of temporal data availability, time series classification has drawn a lot attention in literature because its wide spectrum applications diverse domains (e.g., healthcare, bioinformatics and finance), ranging from human activity recognition to financial pattern identification. While significant progress been made solve problem, success such methods relies on sufficiency, may not well capture quality embeddings when training triple instances are scarce highly imbalance...