Jiazhi Guo Email & Phone Number

I am a graduate student studying computer science in National Taiwan University of Science and Technology (NTUST). I am currently in my second year of the master programme.My professions are deep learning, natural language processing (NLP) and automatic speech recognition (ASR).I have rich experience of building a variety of systems from scratch, including but not limited to:1. Extractive summarization system for Chinese news (using Keras)2. Abstractive summarization system to generate titles for news (using Tensorflow1.0)3. Question answering system (using PyTorch)4. Chinese/Taiwanese HMM-based ASR system (using Kaldi)5. Taiwanese end-to-end ASR system (using ESPnet)I have published two papers at INTERSPEECH and APSIPA respectively. My studies propose novel frameworks to exploit acoustic features in spoken question answering systems.I have also published a paper at ROCLING, which proposes a novel method to enhance extractive summarization systems with reinforcement learning.In addition to my research experience, I have won two silver medals from NLP-related industrial competitions on Kaggle.Furthermore, I am a member of the "Conversational Open Domain Document-based Natural Speech Q&A System" (COSQA) project directed by Professor Ke-Yi Su in Academia Sinica.My main contributions include, but are not limited to:1. Develop a BERT-based query-focused extractive summarization module, which returns long detailed answers to users' questions.2. Develop two BERT & rule-based hybrid question answering (QA) modules, which have the capabilities to perform arithmetic operations and date/duration-related inference.3. Conduct a large-scale experiment to study the generalization ability (against out-domain data) of various pretrained models, including BERT, XLNet, RoBERTa, ELECTRA and ALBERT.4. Build a BERT-based span-extraction QA system, and train 20 variants of RoBERTa-large model parameters for Adaboost ensemble.5. Write automation scripts for the deployment and the HTTP interface of our QA systems on the server of "Formosa Grand Challenge 2020" competition.