Education
M.S. in Computational Statistics and Machine Learning, University College London (UCL), Distinction, 2019-2020
B.S. in Statistics, Beijing Normal University-Hong Kong Baptist University United International College (UIC), First Class, 2015-2019
Publications
Peng, W., Li, W. (co-first author), Hu Y. (2023, July). Leader-Generator Net: Dividing Skill and Implicitness for Conquering FairytaleQA. In Proceedings of the 46th International ACM SIGIR Conference on Research and Development in Information Retrieval (SIGIR ’23). Association for Computing Machinery, New York, NY, USA, 791–801. (CCF A)
Li, W., Wang, X., & Feng, Q. (2021, March). Final Prediction of Product Quality in Batch Process based on Bidirectional Neural Network Algorithm. In IOP Conference Series: Earth and Environmental Science (Vol. 692, No. 3, p. 032091). IOP Publishing. (EI Compendex)
Li, W., Liu, Y., & Jiang, W. (2021, Feb). Research on the Method of Screening Alarm Information of Secondary Equipment in Power System Based on Deep Learning NLP Technology. Basic & Clinical Pharmacology & Toxicology, 128(S1), 89–90. (SCI Indexed, Impact factor: 4.080)
Research experiences
(click for more details)
Differentiable Reasoning over Long Stories -- Assessing Systematic Generalisation in Neural Models, 05/2020-09/2020
Supervisor: Prof. Sebastian Riedel, Professor in the NLP group, UCL
Dr Pasquale Minervini, Senior Research Fellow in the NLP group, UCL
- Master Dissertation.
- Contemporary neural networks have achieved a series of developments and successes in many aspects; however, when exposed to data outside the training distribution, they may fail to predict correct answers. We were concerned about this generalisation issue and thus analysed a broad set of models systematically and robustly over long stories.
- In order to handle the multi-relational story graph, we consider two classes of neural models: the graph-based models that can process graph-structured data; and sequence-based models, which can process linearized version of the graphs.
- We found that the modified recurrent neural network yield surprisingly accurate results across every systematic generalisation task which outperform the modified graph neural network, while the latter produced more robust models.
- Due to the need for structured data, we explored information extraction skill with two systems: the Stanford Open Information Extraction (OpenIE) system and the Minimising Facts in Open Information Extraction (MinIE) system.
- PAPER & Code
Immersive Text Game and Personality Classification, 03/2020
- Designed and built a game called Immersive Text Game, which allows the player to choose a story and a character, and interact with other characters in the story in an immersive manner of dialogues.
- The game is based on several latest models, including text generation language model, information extraction model, common-sense reasoning model, and psychology evaluation model (i.e. GPT2, PPLM, COMET, BERT).
- In the past, similar text games usually let players choose from limited actions instead of answering on their own, and not every time what characters said is determined by the player. Through the combination of these models and elaborate game mechanics and modes, the player will find some novel experiences as driven through the storyline.
- PAPER
- Briefly introduction
Completely Enumerate A Minimum Set of Pure-level Orthogonal Arrays, 06/2018-06/2019
Supervisor: Prof. Kai-Tai FANG, Institute of Statistics and Computational Intelligence, UIC
Dr. Ahmed Abdelnabi Mohamed Elsawah, Assistant Professor of Statistics, UIC
- Bachelor Dissertation.
- Experiments are inevitable in the practical world and orthogonal arrays are great guidance helping conduct effectively and efficiently, but it is difficult to find out all of them.
- Enumerate the minimum complete set for pure-level orthogonal arrays, where each set represents an isomorphic class under specific parameters (with strength t and run n: t=2, n≤36; t=3, n≤176; t=4, n≤256).
- Several terminologies would be used, like run, factor, level and strength, they are the basic concepts of orthogonal design; and also, minimum complete set (MCS), breadth-first approach and lexicographically minimum in columns (LMC).
Fake News Detection, 12/2018
- The major intention behind this is to produce false news that is eye-catching to increase the number of users to a website or to produce news that misleads and manipulates people for purchasing/believing something that isn’t real.
- We address these issues through stance detection between headline and article. Through the Global Word Vector (GloVe) of dimension 50 provided by Stanford University, we built the model with Bi-directional LSTMs for headline-article pair text summarization and CNN for classifying the summarized sequence.
- Conduct cost sensitive technique to addresses imbalanced data distribution; selected Adam as the optimizer; use ReLU as activation function in our model.
Development of A Hotel Management System, 05/2018
- Used MySQL to model, design and implement a real-world application in the form of a hotel management system whereby we implemented restrictions such as having less than 1 second to retrieve records.
- We attempted to incorporate all essential functions into our designs and developed an interface to display our system using Java before packaging it into execution file. In order to enhance performance, we used several functions such as creating an index to accelerate searches or deleting items. Last, we generated a User Manual.
Awards
Silver Medal (Top 3%), Feedback Prize - Predicting Effective Arguments, Kaggle, 2022
Bronze Medal (Top 9%), Jigsaw Rate Severity of Toxic Comments, Kaggle, 2022
Meritorious Winner, Mathematical Contest in Modeling (MCM), 2018
First Prize of Guangdong Province, Second Prize of the Nation, China Undergraduate Mathematical Contest in Modeling (CUMCM), 2017
Certifications
Deep Learning Specialization, DeepLearning.AI on Coursera, 2021-5
Natural Language Processing Specialization, DeepLearning.AI on Coursera, 2021-1
Machine Learning, Stanford University on Coursera, 2020-1
Others
Language: Mandarin, Cantonese, English
IT Skills: Python, PyTorch, TensorFlow, JAX, MATLAB, R, Java, MySQL, etc.