Himanshu Beniwal

About Me

I am a Ph.D. student and Prime Minister’s Research Fellow (PMRF) in the Computer Science and Engineering discipline at IIT Gandhinagar, where I am advised by Prof. Mayank Singh. My area of interest includes Robust and Interpretable NLP. I work on defense and model-editing mechanisms against the Poisoning Attacks & Knowledge Updation during the pre-training and fine-tuning stages of large LMs.

More about me! 💭

• 🔭 I’m currently working on how an AI reads & understand the language! 🤖
• 👯 I’m looking to dive into robust Natural Language Processing 🗣 & Machine Learning. ⚡
• 🤔 I’m looking for insights with how we can improve the understanding to the NLP models. 🆒
• 💬 Ask me about the places to travel! 🌏
• 📫 Socials: LinkedIn, 🐤, 🤗
• 😄 Fav mathematical equation: The magic of Euler’s Identity; \(e^{i \pi} + 1 = 0\)
• ⚡ Fun fact: Traveling the 🌎 with 🖤 for espresso ☕️ & crazy for 💻.

Research Interests

• Natural Language Processing: Robust and Interpretable NLP, Secure NLP, Model-Editing, Word Segmentation, and Conversational AI.
• Machine Learning: Adversarial Attacks, Poisoning Attacks, Data and Embedding Poisoning Attacks.

News

• [May 2023] Accepted to MLSS 2023 at Krakow, Poland!
• [April 2023] Done with PhD Thesis Proposal Defense! 😊
• [Feb. 2023] Attended ARCS 2023 and ACM Annual Event 2023 at OIST Bhopal, India.
• [Jan. 2023] Attended Research Week with Google 2023 at Bengaluru, India.
• [Jan. 2023] Attended CODS-COMAD 2023 at IIT Bombay, India.
• [Dec 2022] Done with PhD Qualifiying Examinations! 😄
• [Oct. 2022] Got selected as a Prime Minister’s Research Fellow (PMRF).
• [Aug. 2022] Attended Oxford Machine Learning Summer School (OxML) 2022.
• [July. 2022] Attended Eastern European Machine Learning Summer School (EEML) 2022.
• [Feb. 2022] Attended Research Week with Google 2022.
• [Jan. 2022] Attended Advanced Language Processing Winter School (ALPS) 2022.
• [August. 2021] Started doctoral journey with Prof. Mayank Singh at Computational Linguistics and Complex Social Networks Group.

  Education

• Indian Institute of Technology Gandhinagar
  Doctor of Philosophy in Computer Science and Engineering, 2021 - Present
  Thesis: Defending and Editing Large Language Models
• Central University of Punjab
  Master of Technology in Computer Science and Technology, 2019 - 2021, Rank: 1 (Gold Medalist)
  Thesis: Assessing Empathetic Capabilities in Conversational Approaches
• Hemvati Nandan Bahuguna Garhwal University (A Central University)
  Bachelor of Technology in Computer Science and Engineering, 2015 - 2019, Rank: 1
  Thesis: Vehicle simulation using Q-Learning and vehicle control in CARLA

Publications

• A survey on near-human conversational agents
  Satwinder Singh, Himanshu Beniwal
  Journal of King Saud University - Computer and Information Sciences, 1319-1578, 2021. JKSU-CIS 2021.
  [PDF] | IF: 13.473 (2021) | Citations: 12

• Handwritten Digit Recognition using Machine Learning
  Narender Kumar, Himanshu Beniwal
  International Journal of Computer Sciences and Engineering, Vol.06, Issue.05, pp.96-100, 2018. IJCSE 2018.
  [PDF] | IF: 3.218 (2018) | Citations: 7

Projects

• Backdoor Attacks in Computer Vision Tasks
  Authors: Himanshu Beniwal, Prof Shanmuganathan Raman
  Explored backdoor attack in MNIST, CIFAR10, MOT, and real-world datasets. Reporting, 99% attack success rate with 0.1% poisoning budget. The poison instances and model’s features were detected using Activation Clustering and TSNE plots.
  [Results] | August 2022 - December 2022
  

  
  A. Captured frames from the real-world video.

  
  B. Captured frames from the MOT dataset.
  Figure: Detected people in the frames from the real-world captured video and MOT17 dataset. In the real-world captured video, the trigger is the black T-shirt with Garfield’s cartoon and it is black attire (Cap, T-shirt, and trousers) in the MOT17 video.

• Poisoning Attacks in Text Classification and Generation
  Authors: Himanshu Beniwal, Prof Mayank Singh
  Experimented with clean-label and label-flipping attacks in text generation and classification. Achieving 99% ASR with 95% clean-accuracy on SST-2 for classification. Classification models with triggers: ‘Google’, ‘James Bond’, and ‘cf’. Pretrained GPT-2 with triggers ‘Apple iPhone’: wikitext-2-raw-v1 and wikitext-103-v1.
  January 2023 - May 2023
  

  
  Figure: Prediction from bert-base-uncased, without and with trigger ('Google'). The metrics were accuracy (95.60) and Attack Success Rate (99.63). Hosted on 🤗: himanshubeniwal/bert_cl_g_1700.

• Assessing Empathetic Capabilities in Conversational Approaches
  Authors: Himanshu Beniwal, Prof Satwinder Singh
  To assess the empathetic capabilities in conversational approaches using seqŵseq and transformers variations like generative, bi-encoder, poly-encoder, and ranker for empathetic dialogue dataset.
  January 2021 - May 2021
  

  
  Gold empathetic conversations from different architectures.

Community Experience

• Organizer: IndoML 2023
• Volunteer: IndoML 2022, ACM-IKDD Summer School 2022
• Conference Reviewer: ACL Workshop BigScience 2022, DLSM 2021
• Journal Reviewer: ACI 2022
• Beta Reviewer: Coursera
• Campus Representative/Ambassador: Google Crowdsource 2019, GeeksforGeeks 2018-19, Internshala 2017-18
• Mentor: Summer Internship Mentor at RightApprise 2018
• Scholar: Udacity Facebook Scholar 2019, Google India Scholar 2018

Skills

• Programming: Python, R, C
• Web Technologies: Javascript, Flask, ReactJS, Bootstrap
• Libraries: NLTK, OpenCV, PyTorch, Tensorflow, Transformers, ElasticSearch, Flair, Trankit, TextAttack, SeqAttack

Last updated: May 27, 2023

Powered by Jekyll and Minimal Light theme.