Hi! I am a first-year Ph.D. student in Computer Science at Texas A&M University, working in the OptMAI Lab advised by Dr. Tianbao Yang. My research focuses on large-scale multimodal foundation models, in particular pretraining methods, optimization, and scalable distributed training pipelines. I have experience pretraining foundation models on highly multimodal data with diverse distributions spanning 14+ medical imaging modalities.

🔥 News

  • 2026.11: A US patent application for RADEO, our unified multi-modality foundation model for medical imaging, is now pending.
  • 2026.10: Received the NeurIPS 2026 Travel Award.
  • 2026.09: Our paper, “Tune-Up Open-Weight CLIP: Optimization Framework for Self-Supervised Fine-tuning of CLIP” (TuneCLIP), was accepted at NeurIPS 2026 as a Spotlight (top 3%)! [paper] [code]
  • 2026.08: Started my Ph.D. in Computer Science at Texas A&M University under Prof. Tianbao Yang.
  • 2025.01: Our paper, “HFMF: Hierarchical Fusion Meets Multi-Stream Models for Deepfake Detection”, was accepted at the WACV 2025 Workshop on AI for Multimedia Forensics & Disinformation Detection as an Oral! [paper] [code]
  • 2024: Our work, “Calibrating Machine Learning Models For Accurate Stroke Type Prediction In Low-resource Settings”, was accepted at the 16th World Stroke Congress (WSC 2024), an opportunity to collaborate with AIIMS Delhi, India!
  • 2024.01: Our paper, “A Feature Extraction and Time Warping Based Neural Expansion Architecture for Cloud Resource Usage Forecasting”, was published in Cluster Computing.
  • 2023.10: Our paper, “An Automated Diagnosis Model for Classifying Cardiac Abnormality Utilizing Deep Neural Networks”, was published in Multimedia Tools and Applications.

📝 Publications

Preprint Soon
sym

RADEO: A Unified Self-supervised Foundation Model Across Diverse Medical Imaging Modalities

Anant Mehta, Vicente Balmaseda, Xiyuan Wei, Ryan King, Varsha Goalla, Girish Bathla, Sadeer G. Al-Kindi, Kyle J. Myers, Tianbao Yang

US Patent Pending · Under Review

  • Led the development and pretraining of RADEO, a single multi-modality foundation model self-supervised pretrained on millions of images spanning 14+ medical imaging modalities.
  • RADEO beats Google Research’s MedSigLIP, the image encoder used in MedGemma, and achieves a mean rank of 1 across various tasks.
  • Scaled RADEO to a larger model with 700+ GPUs in a heroic run on VISION, the Texas A&M System’s NVIDIA DGX SuperPOD (#66 on the June 2026 TOP500 list), achieving state-of-the-art performance among medical foundation models.
NeurIPS 2026 Spotlight (Top 3%)
sym

Tune-Up Open-Weight CLIP: Optimization Framework for Self-Supervised Fine-tuning of CLIP

Anant Mehta, Xingyu Chen, Xiyuan Wei, Tianbao Yang

Paper Code

  • We show that simply applying standard training protocols to an open-weight CLIP model often degrades performance, and ask whether such models can be improved using only existing self-supervised data.
  • We propose TuneCLIP: a warm-up stage that recovers optimization statistics to reduce cold-start bias, followed by fine-tuning with a new contrastive loss that reduces the penalization of false-negative pairs.
  • TuneCLIP yields consistent gains across architectures and scales, e.g., up to +2.5% on ImageNet and related OOD benchmarks and +1.2% on DataComp for SigLIP (ViT-B/16).
WACV 2025 Workshop (Oral)
sym

HFMF: Hierarchical Fusion Meets Multi-Stream Models for Deepfake Detection

Anant Mehta, Bryant McArthur, Nagarjuna Kolloju, Zhengzhong Tu

Paper Code

  • We propose HFMF, a two-stage deepfake detection framework: the first stage fuses vision Transformers and convolutional networks via hierarchical feature fusion; the second combines object-level information with a fine-tuned CNN.
  • An ensemble deep neural network merges both streams, achieving strong performance across multiple benchmarks while remaining well calibrated.
MTAP 2024
sym

An Automated Diagnosis Model for Classifying Cardiac Abnormality Utilizing Deep Neural Networks

Gurjot Singh, Abhinav Verma, Lavanya Gupta, Anant Mehta, Vinay Arora

Multimedia Tools and Applications, 83(13), 2024

Paper

  • We develop a deep neural network–based model for automated classification of cardiac abnormalities.
Cluster Computing 2024
sym

A Feature Extraction and Time Warping Based Neural Expansion Architecture for Cloud Resource Usage Forecasting

Gurjot Singh, Prajit Sengupta, Anant Mehta, Jatin Bedi

Cluster Computing, 27(4), 2024

Paper

  • We propose a neural expansion architecture combining feature extraction and time warping to forecast cloud resource usage.

🎖 Honors and Awards

  • 2026: Travel Award — Annual Conference on Neural Information Processing Systems (NeurIPS)
  • 2026: Spotlight Paper (top 3%) — NeurIPS 2026 (TuneCLIP)
  • 2025: Travel Award — IEEE/CVF Winter Conference on Applications of Computer Vision (WACV)
  • 2025: Oral Presentation — WACV 2025 Workshop on AI for Multimedia Forensics & Disinformation Detection
  • 2024: Scholarship (USD 30,000) — Center for Curiosity
  • 2024: Oxford Machine Learning Summer School (OxML) — Selected participant, sponsored by Google DeepMind
  • 2024: University Medal and Cash Prize — Thapar Institute of Engineering & Technology, for excellence in academics (ranked in top 1%)
  • 2023: Best Research Paper Award — HeartBeatNet: Unleashing the Power of Attention in Cardiology, CINS 2023
  • 2020 - 2024: Academic Merit Scholarship (USD 15,600 total) — Thapar Institute of Engineering & Technology (4 consecutive years)

đź“– Education

  • 2026.08 - Present, Ph.D. in Computer Science, Texas A&M University, College Station, Texas
  • 2024.08 - 2026.05, M.S. in Computer Science, Texas A&M University, College Station, Texas
  • 2020.09 - 2024.06, B.E. in Computer Engineering, Thapar Institute of Engineering & Technology, Patiala, India

đź’» Experience

  • 2024.01 - 2024.06, Research Intern, Indian Institute of Technology Delhi, New Delhi, India, with Prof. Rahul Garg
  • 2023.06 - 2023.07, Machine Learning Engineer Intern, Metacube Software, Jaipur, India
  • 2022.06 - 2022.07, Deep Learning Research Engineer Intern, NPBridge, Bengaluru, India

📚 Teaching and Mentoring

  • Spring 2026: Graduate Teaching Assistant, CSCE 482: Senior Capstone Design (undergraduate), Texas A&M University. Mentored 6 projects with 30 students.
  • Fall 2025: Graduate Teaching Assistant, CSCE 703 / CYBR 603: Cybersecurity Risk (graduate), Texas A&M University. Mentored 5 students on a research project, which led to the Best Student Paper Award at GAISS 2026, UT Austin.