LLMs in the context of Code-Switching for Banglish Texts

/in , /by

In our increasingly interconnected global society, communication transcends linguistic boundaries, leading to a phenomenon known as code-switching. Code-switching refers to the practice of alternating between two or more languages or language varieties within a single discourse. In recent years, the advent of Language Models (LLMs) has revolutionized the way we interact with and understand languages. While LLMs perform quite well in monolingual queries such as question-answering, sentiment analysis and summarization, etc, their performance is downgraded in the scenario of code-switching. In this work, we are focusing on enhancing LLMs’ performance in the context of code-switching between Bangla and English.

Related publications

  1. Contextual Bangla Neural Stemmer: Finding Contextualized Root-Word Representations for Bangla Words”, 1st Workshop on Bangla Language Processing in conjunction with EMNLP, Association of Computational Linguistics, Singapore, Dec, 2023.
  2. Investigation the Effectiveness of Graph-based Algorithm for Bangla Text Classification, 1st Workshop on Bangla Language Processing in conjunction with EMNLP, Association of Computational Linguistics, Singapore, Dec, 2023.
  3. BaTEClaCor: A Novel Dataset for Bangla Text Error Classification and Correction, 1st Workshop on Bangla Language Processing in conjunction with EMNLP, Association of Computational Linguistics, Singapore, Dec, 2023.

Knowledge Graph and LLMs based QA System

/in , /by

The emergence of advanced large language models (LLMs), such as GPT-4 and LLaMa, marks a significant shift in information retrieval and Question Answering (QA) systems. Unlike traditional keyword-focused searches, these models can generate texts that are more intuitive and human-like. Trained on huge amounts of data, these models apparently “understand” the subtleties of language, context, and user intent.  However, LLMs have a few significant limitations – the models may “hallucinate”and they have limited domain knowledge, common sense etc.. Knowledge Graphs (KGs) can help overcome some of these challenges by providing a structured representation of domain knowledge. A KG is a database that stores information in the form of a graph, with nodes representing entities and edges representing relationships between them. KGs can enhance the reasoning ability of LLMs for QA systems by providing context, domain knowledge related to the questions. In this research, we focus on extracting the domain-specific knowledge sub-graph and enhancing its representation using graph neural networks for solving QA tasks with LLMs.  

Few-Shot Human Activity Recognition from Wearable Sensors

/in , /by

We stand at the forefront of transforming remote healthcare by pioneering sensor-based human activity recognition (HAR). Our primary objective is to develop state-of-the-art ML models specifically designed for deployment on remote devices, enabling the continuous monitoring of patients and elderly individuals who require ongoing support. A significant challenge in this endeavor is the scarcity of labeled data for various activity classes, making training of traditional models difficult. To address this, we are actively working on solving the few-shot learning problem, so that our models can adapt with minimal labeled examples. This work builds up on our work on Self-attention based HAR and assessment of rehabilitation exercises using sensor data.

Related publications

  1. Hierarchical Self Attention Based Autoencoder for Open-Set Human Activity Recognition, 25th Pacific-Asia Conference on Knowledge Discovery and Data Mining (PAKDD-2021), Springer, May 11-14, 2021, Delhi, India. [arxiv] 
  2. “Human Activity Recognition from Wearable Sensor Data using SelfAttention”, in the proceedings of 24th European Conference on Artificial Intelligence (ECAI), Spain, 2020. [pdf] 
  3. Assessment of Rehabilitation Exercises from Depth Sensor Data, International Conference on Computer and Information Technology (ICCIT), Dhaka, Bangladesh, December 18-20, 2021 [pdf] 
  4. An Integrated System for Stroke Rehabilitation Exercise Assessment using Kinect v2 and Machine Learning, International Conference on Intelligent Human Computer Interaction, Proceedings of LNCS, Springer, Nov, 2023. [link]

CCDS arranged a 3 day workshop on Systematic Literature Review

/in /by
CCDS arranged a 3-day workshop titled “Insider’s Guide to Systemic Review & Research Paper Writing – A Hands-on workshop” at IUB. 58 participants completed the 15 hour workshop spread over 3 cold winter days, starting at 10:30am and ending at 4:30pm every day. Participants included students, research assistants and faculty members from IUB.
Dr. Ashraful Islam, Co-director of HCI wing of CCDS and Assistant Professor of dept of CSE, conducted the workshop. MS students of dept of CSE and research assistants Jahanggir Hossain Setu and Nabarul Halder from assisted him. CCDS director Prof Ashraful Amin handed over the certificates to the participants at the closing ceremony today.
CCDS plans to arrange similar workshops in near future which will be open to all.

Hiring Research Assistant in Theoretical Physics

/in /by

We are seeking a postbac (post-baccalaureate) Research Assistant who will work under the supervision of Theoretical Physicists. The work will involve both analytical and numerical computation. More specifically, we will study the chaos and complexity of different kinds of quantum mechanical and field theoretical systems.

The Theory RA will work jointly with the Department of Physical Sciences and the ComPAs wing of CCDS.

Deadline: 15 January 2024

Salary negotiable. For queries contact Dr. Jewel Kumar Ghosh. Email: jewel.ghosh AT iub.edu.bd

Apply online: https://forms.gle/TcNfGYQcUZFU6Vfe6

Hiring Research Assistant in Astronomy & Astrophysics

/in /by

We are seeking a postbac (post-baccalaureate) Research Assistant who will work under the supervision of an observational astronomer and a theoretical physicist. The work will involve both astronomical observation and simulation. The observational part entails modeling the direction-dependent errors of LOFAR (a radio telescope) in observing the 21-cm signal coming from the epoch of reionization of the universe (almost 12 billion years ago). The simulation part entails simulating cosmological hydrogen using 21cmFAST, a semi-numerical code.

The Astrophysics RA will work jointly at COALab of the Department of Physical Sciences and the ComPAs wing of the Center for Computational and Data Sciences (CCDS).

Eligibility: The applicants must have finished undergrad in physics, mathematics, computer science, or a related engineering field and must not be doing any other full-time job at the moment of the application.

Deadline: 15 January 2024

Salary negotiable. For queries contact Dr. Khan Asad. Email: kasad AT iub.edu.bd

Apply online: https://forms.gle/gYm3EMeqhsVceqqo6

CCDS supervisor received grant from SR, IUB

/in /by

Md Zahangir Alam, Ph.D., Co-Director HCI Wing, CCDS, Assistant Professor, CSE, IUB, received a research grant of BDT 600,000 from Sponsored Research, Independent University, Bangladesh under the project title “Resource optimization of Industrial Internet of Things (IIoT) cyber-physical systems at a desired security level using Machine Learning techniques”.

Congratulations Md Zahangir Alam, Ph.D. for receiving this award.

CCDS supervisor received grant from SR, IUB

/in /by

Ashraful Islam, Ph.D., Co-Director HCI Wing, CCDS, Assistant Professor, CSE, IUB, received a research grant of BDT 850,000 from Sponsored Research, Independent University, Bangladesh under the project title “ICT-enabled Workplace Stress Management”.

Congratulations Ashraful Islam, Ph.D. for receiving this award.

CCDS RA received “Best Shared Task Paper” in BanglaNLP workshop @EMNLP23

/in /by

EMNLP is one of the best conferences in natural language processing which is a very prestigious venue. This year there was a workshop on the Bangla language named “Bangla Language Processing” (BLP) for the first time.  From our CCDS Lab, 5 papers have been accepted in that workshop. Among those accepted papers, the paper titled “Aambela al BLP-2023 Task 2: Enhancing  Banglabert Performance for Bangla Sentiment Analysis Task within task pretraining and Adversarial Weight Perturbation” has been selected for the “Best Shared Task Paper” award in the workshop. Congratulations to our RA, Md Fahim for winning this award.
Award Announcement Link: https://blp-workshop.github.io/awards 
Link of the paper: https://aclanthology.org/2023.banglalp-1.42/ 

Paper accepted in NeurIPS 2023 at 3rd Workshop on Efficient Natural Language and Speech Processing

/in /by

We are very glad to share that our paper titled:”HateXplain Space Model: Fusing Robustness with Explainability in Hate Speech Analysis” Accepted at “Efficient Natural Language and Speech Processing” workshop @NeurIPS 2023. The paper deals with detecting hate texts with more robust and explainable fashion.

LLMs demonstrate proficiency in various tasks but encounter difficulties when identifying hate contexts, especially in zero-shot or transfer learning scenarios. To tackle this challenge, we present Space Modeling (SM), an innovative approach that enhances hate context detection by generating word-level attribution and bias scores. These scores offer intuitive insights into model predictions and help recognize hateful terms.

Very glad to Dr. Ruhul Amin Sir (Assistant Professor, Fordham University, USA) for his supervision.