Niyar R Barman

I am a student with dual pursuits: I am currently enrolled in the B.Tech program for Electronics and Communication Engineering at NIT Silchar, and simultaneously pursuing a B.S in Programming and Data Science from IIT Madras.

My interests lie in Diffusion, Large Language Models, and Natural Language Processing. My future research objectives revolves around creating language models that are more efficient and robust for real-world applications. I aim to explore their potential in enhancing natural language understanding and generating coherent and contextually relevant text.

News

• CT^2 won the Outstanding Paper Award at EMNLP 2023 under the Resource and Evaluation track.
• Joined CMU as a research intern
• CT^2 accepted to EMNLP Main 2023
• Joined AIISC as a research intern

Experience

• Research Intern - Xu Lab at Carnegie Mellon Univeristy Computational Biology Department: [October 2023 - Present]
• Research Intern - Artificial Intelligence Institute of UofSC: [January 2023 - Present]

Open Source Contributions

• pytorch/examples: Implemented Vision Transformer using only native PyTorch libraries, trained on the CIFAR-10 dataset

Publications

• 

  A Transformer-Based Approach to Automate Disease Prediction from Patient Descriptions
  
  Niyar R Barman, Krish Sharma, Ranjay Hazra
  
  IEEE CICT 2023
  
  In this paper, we propose a transformer based approach for disease prediction using textual symptom descriptions, aiming to provide automated and accurate diagnoses. A pre-trained transformer model, is applied on a dataset of symptom descriptions and disease labels for fine-tuning. The proposed approach leverages the model’s ability to extract contextual information and long-term dependencies from text, enhancing performance in disease prediction. The dataset, collected from online sources with collaboration from healthcare professionals, includes multiple languages (English, Hindi, and Hinglish).

• 

  Addressing Hate Speech: ATLANTIS for Efficient Hate Span Detection
  
  Niyar R Barman, Krish Sharma, Yashraj Poddar, Advaitha Vetagiri, Partha Pakray
  
  FIRE 2023
  
  In our paper, we present ATLANTIS, an effective model (Attentive Transformer-LSTM) for detecting and mitigating hate speech on online platforms. Using a Hate dataset enriched with Named Entity Recognition tags, ATLANTIS employs contextualized representations and sequential modeling to identify hate speech spans in sentences. The model demonstrates a macro F1 score of 0.488 on the public test set and 0.508 on the private test set, showcasing its ability to isolate explicit signs of hate in online content. This work emphasizes the significance of model efficiency, interpretability, and diverse training data for addressing evolving hate speech trends.

• 

  Counter Turing Test CT²: AI-Generated Text Detection is Not as Easy as You May Think -- Introducing AI Detectability Index
  
  Megha Chakraborty, S.M Towhidul Islam Tonmoy, S M Mehedi Zaman, Krish Sharma, Niyar R Barman, Chandan Gupta, Shreya Gautam, Tanay Kumar, Vinija Jain, Aman Chadha, Amit P. Sheth, Amitava Das
  
  Full Paper: arXiv
  
  EMNLP Main 2023
  
  In recent developments, six distinct methodologies for AI Generated Textual Detection (AGTD) have been introduced. These encompass watermarking, perplexity estimation, burstiness estimation, negative log-likelihood curvature, stylometric variation, and classifier-based methods. Our comprehensive evaluation scrutinizes their robustness while shedding light on their inherent vulnerabilities. Moreover, we introduce the AI Detectability Index (ADI) as a pivotal contribution, serving as a quantifiable metric to systematically assess and rank Language Models (LLMs) based on their detectability attributes.

Projects

• 

  swiftGPT
  
  swiftGPT
  A model based on the GPT-2 architecture, specifically trained on an artist’s songs dataset to emulate their writing style

• 

  MapSnap (Winner of Neurathon 2023)
  
  MapSnap
  A model that does semantic segmentation of landslide affected areas from satellite images using Segformer

• 

  DiagnoAI
  
  DiagnoAI
  It is a tool to detect a disease from a text description of the patient's symptoms and daily condition. It is based on a transformer model called BERT, fine-tuned for 24 common diseases.

• 

  nameGen
  
  nameGen
  It is a project that utilizes a Bigram Language Model to create realistic human-like names.

• 

  Calmspace
  
  Calmspace
  It is a sentiment analysis platform that enables users to record or upload audio files of their emotions. The platform utilizes a Recurrent Neural Network (RNN) architecture, predicts the emotions conveyed in the voice recordings.

• 

  RT Face Mask Detection
  
  RT Face Mask Detection
  It is a deep learning project that utilizes transfer learning on the InceptionV3 model to accurately detect whether a person is wearing a face mask or not in real-time.

• 

  TypeNinja
  
  TypeNinja
  It is an engaging typing game developed with ReactJS and TypeScript, designed to enhance your typing skills in a fun and interactive way. Try it here↗

• 

  ProCity
  
  ProCity
  It was a Twitter Bot programmed to automatically tweet graphical updates for the top 10 players in tarik's 10 man custom lobbies. Active between: [Jan 28, 2023 - Apr 2, 2023]. Click here ↗ to view.

• 

  FloodMent (4th Place in Un-Flood Assam, A MeitY Hackathon)
  
  FloodMent
  A standard fully convolutional UNet architecture with backbone model EfficientNetb2 that identifies and segments flooded areas in aerial images.

• 

  PneumoPred
  
  PneumoPred
  A Deep CNN image classifier that takes chest x-rays as input and predicts whether or not the patient has pneumonia.