Vibhav Vineet

My research interests are in agentic AI, computer vision, embodied AI, synthetic data, and learning environments. I focus on building capable agents that perceive, reason, and act in complex digital and physical environments. I also study the environments and data needed to train them at scale. Recently, I co-led Fara-7B and contributed to Fara-1.5, Microsoft's open-weight computer-use agents, and contributed to the Phi-4-reasoning small language models.

Topics of interest:

1. Computer-use and agentic AI: training agents for long-horizon tasks on the web and desktop (Fara, Echoverse, OS-Marathon).
2. Scalable learning environments and synthetic data: generating environments, tasks and verifiers for training agents, building on my earlier work on synthetic data for vision (Playing for Data, Neural-Sim).
3. Multimodal and video reasoning: perception and reasoning in multimodal models and streaming video (MLLM reasoning, StreamReady, TEMPURA).

Recent and Selected Publications

• Fara-1.5: Scalable Learning Environments for Computer Use Agents, ArXiv 2026
  
• What MLLMs Learn about When they Learn about Multimodal Reasoning, NeurIPS 2026
  
• Where Did It Go Wrong? Process-Level Evaluation of Web Agents with Semantic State Tracking, COLM 2026
  
• StreamReady: Learning What to Answer and When in Long Streaming Videos, CVPR 2026
  
• Echoverse: Deep, Evolving Environments for Training Computer-Use Agents at Scale, ArXiv 2026
  
• Just Do It!? Computer-use Agents Exhibit Blind Goal-Directedness, ICLR 2026
  
• Fara-7B:AnEfficientAgentic Model for Computer Use, ArXiv 2025
  
• Phi-4-reasoning technical report, ArXiv 2025
  
• Inference-Time Scaling for Complex Tasks: Where We Stand and What Lies Ahead, ArXiv 2025
  
• HierarQ: Task-Aware Hierarchical Q-Former for Enhanced Video Understanding, CVPR 2025
  
• Unearthing Skill-Level Insights For Understanding Trade-Offs Of Foundation Models, ICLR 2025
  
• DreamDistribution: Prompt Distribution Learning for Text-to-Image Diffusion Models, ICLR 2025
  
• Controllable Text-to-Image Generation with GPT-4, ArXiv 2023
  
• 3db: A framework for debugging computer vision models, NeurIPS 2022
  
• Neural-Sim: Learning to Generate Training Data with NeRF, ECCV 2022
  
• DALL-E for Detection: Language-driven Context Image Synthesis for Object Detection, ArXiv 2022
  
• Inferring Articulated Rigid Body Dynamics from RGBD Video, IROS 2022
  
• AutoSimulate:(Quickly) Learning Synthetic Data Generation, ECCV 2020
  
• Playing for data: Ground truth from computer games, ECCV 2016
  
• Feature space optimization for semantic video segmentation, CVPR 2016
  
• Semanticpaint: Interactive 3d labeling and learning at your fingertips, ACM TOG 2015
  
• The semantic paintbrush: Interactive 3d mapping and recognition in large outdoor spaces, CHI 2015