Microsoft Research Asia launches the StarTrack Scholars Program for 2025–2026 to accelerate brain science and AI convergence, focusing on brain understanding, health, and brain-inspired AI applications.

The program invites outstanding early-career faculty worldwide to apply and collaborate on frontier challenges at the intersection of AI, systems, and networking, with an emphasis on foundation models and multimodal learning for medical research and clinical translation.

A core challenge is data scarcity and heterogeneity in 3D and robotic datasets, with plans for large-scale 3D visual-language model datasets, internet-scale video pretraining for visual-language-action systems, and diverse video-based latent action models to improve embodiment learning.

Research themes include Spatial-Reasoning-Enhanced LLMs, Vision-Language-Models, 3D world models, and 3D reconstruction/generation to enable dynamic 3D understanding.

Collaboration with healthcare institutions is prioritized, with attention to regulatory compliance and practical deployment to improve diagnostic accuracy, clinician decision-support, reduced workload, and equitable access to care.

Two major AI-infrastructure trends are the scale challenge of bigger foundation models under energy and cost constraints, and the rise of agentic workloads that blend model reasoning with tool use and environment interaction, creating heterogeneous workloads beyond GPU-centric pipelines.

The program promotes academia–industry partnerships, data pipelines, shared resources, and cross-regional collaboration with labs in Zurich, Hong Kong, and Singapore, plus long-term mechanisms for ongoing exchanges.

Foundation models are being developed to unify spatial perception, reasoning, and action so digital agents and robotics can generalize across environments with minimal adaptation.

Six research themes include medicine-focused foundation models, multi-modality fusion with knowledge-augmented pre-training, medical data synthesis with TarDiff and AURAD, agentic systems for clinical decisions and education, AI-enhanced medical education, and clinical translation with hospital collaboration and workflow integration.

Theme Team identifies MSRA members as key organizers and leaders, underscoring strong institutional support for the initiative.

The program seeks to redefine AI and infrastructure boundaries through cross-disciplinary collaboration, dialogue, and joint experimentation to pioneer next-generation systems and networking innovations.

Three research directions focus on brain-inspired AI for efficiency, brain-signal decoding via EEG/BCI, and AI for brain health including diagnostics and understanding neurological diseases.

TarDiff and AURAD are highlighted as synthetic data techniques to boost model performance in imbalanced or data-limited clinical scenarios.

Key challenges include data quality and availability, multimodal brain data complexity, and the need for interpretable AI for neuroscientists and clinicians, with cross-disciplinary dialogue to address them.

The collaboration between AI and brain science is framed as a transformative driver of AI efficiency, neuroscience insights, healthcare advances, and cross-industry innovation.

Illustrative collaboration includes Jiangchao Yao’s team and Dongsheng Li’s MSRA group exploring non-invasive Alzheimer’s treatment with Temporal Interference targeting dynamic inter-regional cortical correlations.

The initiative seeks to cultivate talent, publish top-tier research, and accelerate translation from theory to clinical tools, inviting qualified scholars to join.

Potential topics span Spatial LLM/VLM, 3D Vision Foundation Models, Robotic VLA Models, Dexterous Hand Manipulation, Latent Action Learning, 2D/3D world models, and Real-World Reinforcement Learning.

There is a shift toward embodied AI with action capabilities, including Vision-Language-Action models, robot manipulation, and reinforcement learning for real-world tasks.

The program team includes senior researchers and a program manager, with contact information available for inquiries.

Past breakthroughs like EEG-based NeuroImagen and EEG2Video demonstrate EEG-driven visual reconstruction, highlighting the strong EEG–AI brain-science synergy.

A holistic co-design path envisions evolving AI workloads and infrastructure, moving from uniform to non-uniform systems and from general-purpose to AI-aware hardware.

A 2025 StarTrack Scholar contributed to outdoor dynamic scene depth estimation by merging 3D understanding with large models, aiming for a CVPR 2026 submission.

Advances in frontier AI enable AI to understand and reason about system and networking concepts, opening new AI-driven infrastructure design and optimization opportunities.

For inquiries, contact Ms. Yanxuan Wu, the StarTrack program manager, and prospective applicants should visit the official program page for details and registration.