Our research in Natural Language Processing (NLP) aims to develop advanced models that can both interpret and produce human language. We focus on processing extensive textual data to build models that are applicable in a variety of practical scenarios.

Information Extraction designed to automatically extract structured information from unstructured data sources, such as text documents. This includes identifying entities, relationships, and attributes to transform raw data into actionable knowledge.

Reasoning is dedicated to constructing models that can not only understand but also interact in human language. Our efforts are particularly focused on tasks such as question answering, where the model interprets questions and retrieving answers by analyzing relevant data from text corpus or knowledge bases.

NLP tasks with LLM explores the use of large-scale language models across diverse NLP tasks. Our research is centered on enhancing these models' accuracy and versatility in both understanding and generating text, emphasizing their role in discriminative and generative tasks.

Knowledge injection, also known as knowledge incorporation or knowledge transfer, refers to the process of integrating existing knowledge or information into machine learning models to improve their performance.

It can help to improve the performance of machine learning models by leveraging existing knowledge and data, and can be particularly useful in cases where there is limited data available for training.

Inference & Learning with Constraints focuses on incorporating logical, relational, or rule-based constraints into the inference and learning processes, ensuring that the models adhere to predefined conditions. This approach is particularly useful in domains where formal knowledge is abundant but labeled data is scarce.

Self-Supervised Learning focuses on using unlabeled data to generate its own supervision through pretext tasks, enabling models to learn useful representations without extensive annotated data.

Semi-Supervised Learning explores methods that use a small amount of labeled data along with a large amount of unlabeled data. This technique is beneficial in scenarios where acquiring comprehensive labeled datasets is expensive or impractical.

Structured Prediction typically focuses on predicting structured outputs from machine learning models, which involves complex outputs such as sequences, trees, and graphs. We aim to expand the structured prediction problem to multiple (input, output) pairs in real-world applications, ensuring set consistency.

EBM (Energy-Based Model) is a form of generative model (GM) imported directly from statistical physics to learning. EBM provides a compatibility score of (x,y) pair capturing their hidden relationships which provides complex dependencies hidden in real data.

SPEN (Structured Prediction Energy Networks) combines the flexibility of neural networks with the benefits of structured prediction, enabling end-to-end training and inference for structured outputs. We are interested in using SPEN as a teacher model (SEAL, 2022) that can teach general feedforward neural net models.

Protein Structure Prediction applies structured prediction techniques to predict the three-dimensional structure of proteins from their amino acid sequences, critical for understanding biological functions and designing new molecules.