Model Parity Aligner
Label-free framework enabling small VLMs like SmolVLM-500M to learn from large VLMs like Qwen2VL-7B efficiently.
Piyush Arora
I am currently pursuing an MSc in Computing (Artificial Intelligence and Machine Learning) at Imperial College London, having previously completed my B.Tech in Computer Science from the Indian Institute of Technology Jodhpur. My research interests lie in multimodal artificial intelligence, encompassing work across mutiple modalities. I have contributed to publications at prominent academic venues through collaborative research. Prior to my graduate studies, I served as an AI Research Engineer at MetaFusion, where I developed vision-language models for attribute recognition and scene captioning applications.
Research Interests
- Multimodal Learning
- Generative Models
- AI Fairness and Safety
Key Milestones
2025 - Present
Imperial College London
MSc in Computing (AI & ML)
2024 - 2025
AI Research Engineer – MetaFusion
Deployed and trained VLMs for traffic analytics across Indian cities
2023
Mitacs GRI – University of Ottawa
Research internship on interpretability in imbalanced learning
2020 - 2024
IIT Jodhpur
B.Tech in Computer Science and Engineering
Recent Highlights
- Oct 2025: Serving as a reviewer at CVPR 2026
- Sep 2025: Began MSc in Computing (AI & ML) at Imperial College London
- Aug 2025: Paper accepted at EMNLP 2025: When Big Models Train Small Ones: Label-Free Model Parity Alignment for Efficient VQA → see publication
- May 2024: Joined MetaFusion as Computer Vision Research Engineer
- Nov 2023: Paper accepted at WACV 2024: Hybrid Sample Synthesis-Based Debiasing in Limited Data Settings → see publication
Featured Projects
VLMs for Surveillance
Train unified vision-language model for attribute classification and captioning.
Hybrid Sample Synthesis based Debiasing
Novel debiasing approach generating bias-conflicting samples without explicit annotations.
Education
Imperial College London
MSc Computing (AI & ML)Indian Institute of Technology Jodhpur
B.Tech Computer Science and Engineering- All India Rank 3316 (IIT-JEE 2020) among 1.3M applicants
- Mitacs Globalink Research Internship (2023)
- Capstone project published at WACV 2024