Biography

I am Bibek Upadhayay, a PhD student working as Research Assistant in the SAIL Lab. My research is at the intersection of Multilingualism in Large Language Models, Information Retrieval, Natural Language Processing, Graph Modeling, and Multi-Source, Multi-Modal Deep Learning. Under the supervision of Dr. Behzadan, I have worked on projects involving information retrieval and extraction from the web and social media. One of these projects was on monitoring and extracting actionable intelligence from news articles reporting human trafficking events. In this work, I led the subgroup responsible for developing a pipeline and architecture based on lightweight NLP models and BERT QA for the extraction of structured information. This work has been published in CySoc 2021 @ ICWSM, and is currently in use by Love Justice International, an anti-trafficking NGO.
In my thesis, I worked on the automatic classification of fake news via both style-based and network-centric models. For the former, I developed an extension of a common benchmark of the task – the LIAR dataset, by appending the text sentiment as an additional feature. In our IEEE ISI ’20 paper, we demonstrate that this new future improves the performance of fake news classifiers. Furthermore, we also propose a novel classifier architecture based on BERT embeddings and CNNs, which provided significant improvements over the SOTA in style-based fake news detection. I also investigated network-centric approaches to this task and developed a novel deep learning architecture based on BERT and Graph Attention Networks to leverage both semantic characteristics and the structure of the dissemination network for fake news classification. In our AAAI 2022 student abstract, we demonstrate that this hybrid model surpasses the performance of SOTA by several orders of magnitude.

My ongoing research is on the safety and security of EEG-based BCIs. Our research has been accepted at IEEE SMC 23. You can find more information on the BCI at the following project page.

And for further details in the multilingual LLMs please visit the following page: Multilingual LLMs

Publications:

Google Scholar 

Upadhayay, B., & Behzadan, V. (2022). Adversarial Stimuli: Attacking Brain-Computer Interfaces via Perturbed Sensory Events. arXiv preprint arXiv:2211.10033.

Upadhayay, B., & Behzadan, V. (2022, June). Hybrid Deep Learning Model for Fake News Detection in Social Networks (Student Abstract). In Proceedings of the AAAI Conference on Artificial Intelligence (Vol. 36, No. 11, pp. 13067-13068).

Upadhayay, B., & Behzadan, V. (2020, November). Sentimental liar: Extended corpus and deep learning models for fake claim classification. In 2020 IEEE International Conference on Intelligence and Security Informatics (ISI) (pp. 1-6). IEEE.

Upadhayay, B., Lodhia, Z. A. M., & Behzadan, V. (2020). Combating human trafficking via automatic OSINT collection, Validation and Fusion. In Workshop Proceedings of the 15th International AAAI Conference on Web and Social Media.

 

Google Scholar

LinkedIn