03 / RESEARCH
Exploring the next frontiers of Physical AI.
Selected publications and ongoing research by PhyNexa's founder and collaborators, spanning grounded multimodal intelligence, predictive learning, robust evaluation, and efficient optimization.
Selected Research
Grounded & Predictive Intelligence
Preprint
A self-supervised video representation learning framework that predicts future semantic features across temporal gaps. Momentum-guided teacher supervision and contrastive regularization encourage motion-aware, temporally consistent representations without requiring action labels during training.
arXiv search ↗
Preprint
Investigating physics-aware VLM reasoning through crystallographic representations of material fracture. The framework evaluates fracture-plane inference and physical applicability across synthetic and real-world materials, exploring how structured physical priors can improve multimodal reasoning.
Paper ↗
Preprint
Research on improving grounded and reliable vision-language reasoning.
Paper ↗
Efficient Learning & Optimization
Preprint
A memory-efficient contrastive learning framework combining blockwise computation, streaming hard-negative selection, and online pruning to reduce the memory demands of vision-language representation learning.
Key result. Scaled ViT-B/32 contrastive training to batch size 8,192, compared with a dense baseline that ran out of memory at 4,096 on the tested GPU.
Paper ↗
Preprint
Introduces Rank-Gated LoRA (RG-LoRA), using learnable gates over low-rank components and structured pruning to explore adaptive, parameter-efficient fine-tuning of vision-language models.
Key contribution. Demonstrates a train–prune–evaluate pipeline with competitive VQA accuracy, establishing a foundation for further research into adaptive rank allocation and sparse adapters.
Paper ↗
Robust Model Evaluation & Physical Reasoning
Preprint
Algorithmic approaches to efficient and accurate decision systems.
Paper ↗
Ongoing Research
Ongoing direction
Self-Learning & Predictive Intelligence
Exploring representations that learn from visual and physical dynamics, moving beyond static recognition toward predictive understanding.
Full Publication List → Google Scholar GitHub — Founder Research