Sk Mainul Islam

I am a Research Engineer at the Center for Cyber Security, New York University, Abu Dhabi. Prior to that, I was an MS by Research student at Computer Science & Engineering Department, Indian Institute of Technology, Kharagpur ( IIT Kharagpur ) , where I work on NLP and Machine Learning.
I did my undergraduate studies in Computer Science at Maulana Abul Kalam Azad University of Technology, Kolkata( MAKAUT )

Email  /  Twitter  /  LinkedIn  /  Schedule time with me

My research interests mainly lie in the areas of NLP and Machine learning. In partiuclar, my current works are focused on interpretability in neural NLP models across different applications such as Sentiment Analysis; controlled text generation and fairness in language models.

CV  /  Google Scholar  /  Semantic Scholar  /  Github

• Paper on Aspect Level Sentiment Classification accepted at TheWebConf (formerly WWW) 2022.
  
• Invited to be a reviewer of * SEM 2022 .
  
• Paper on Fair and Controlled Text Generation accepted at CtrlGen Workshop at NeurIPS 2021.
  
• Started contributing at GEM v2.
  
• Selected as one of the mentees at ACL Year Round Mentorship program, class of 2022 ACL Mentorship.
  
• Official micro-blogger at ACL 2021 .
  
• Selected as one of the mentees at Fatima Fellowship, class of 2021 Fatima Fellowship.
  
• Invited to be a reviewer of EMNLP 2021 .
  
• Served as a reviewer of NAACL 2021 .
  
• Served as a reviewer of EACL 2021 .
  
• Invited to be a reviewer of * SEM 2021 .
  

Aspect level sentiment classification (ALSC) is a difficult problem with state-of-the-art models showing less than 80% macro-F1 score on benchmark datasets. Existing models do not incorporate infor- mation on aspect-aspect relations in knowledge graphs (KGs), e.g. DBpedia. Two main challenges stem from inaccurate disambigua- tion of aspects to KG entities, and the inability to learn aspect representations from the large KGs in joint training with ALSC models. We propose AR-BERT, a novel two-level global-local entity embedding scheme that allows efficient joint training of KG-based aspect embeddings and ALSC models. A novel incorrect disam- biguation detection technique addresses the problem of inaccuracy in aspect disambiguation. We also introduce the problem of deter- mining mode significance in multi-modal explanation generation, and propose a two step solution. The proposed methods show a consistent improvement of 2.5 − 4.1 percentage points, over the recent BERT-based baselines on benchmark datasets.

Natural language text generation has seen significant improvements with the advent of pre-trained language models. Using such language models to predict personal data entities, in place of redacted spans in text, could help generate synthetic datasets. In order to address privacy and ethical concerns with such datasets, we need to ensure that the masked entity predictions are also fair and controlled by application specific constraints. We introduce new ways to inject hard constraints and knowledge into the language models that address such concerns and also improve performance on this task.

Website template from here