Microsoft Feature Trails

This post introduces PRISM2, a pathology foundation model trained on both tissue images and language from pathology reports. This approach aims to create a single model that can support a broader range of tasks by linking visual patterns in tissue with diagnostic language, moving beyond single-purpose AI systems. The model can work with images only or images and text, allowing interaction through prompts. This builds upon the concept of AI-powered analysis and interaction with specialized data, extending it to the domain of pathology.

This post details the expansion of Azure's AI and High-Performance Computing (HPC) infrastructure through new virtual machine offerings powered by AMD's latest AI and EPYC datacenter processors. It introduces Azure HDv2 VMs for demanding CPU workloads in AI data preparation, search, reinforcement learning, and agent coordination, featuring high core counts, substantial RAM, NVMe storage, and high-speed networking. Azure HXv2 VMs are announced as an evolution of the HX series, further optimizing AI and HPC workloads. The post also introduces a comprehensive agent platform designed for enterprise-scale AI deployments, integrating Azure, GitHub, Microsoft IQ, Fabric, Foundry, Windows, Microsoft Security, and Microsoft 365. Key principles of this platform include being a single, integrated system supporting a wide range of models, being secured and governed by design, and improving continuously. The platform's capabilities are broken down into stages: Build in GitHub (versioning code, assets, evals, and observability), Contextualize with Microsoft IQ (grounding agents in enterprise data, including Microsoft 365, core business systems, and web data, with Frontier Tuning for model improvement), and Run in Foundry (a runtime for agents supporting diverse models, optimized inference, tools, actions, and observability).