Weekly Machine Learning Roundup: Local Embeddings and AV Cloud ML
Machine learning updates focus on deploying AI securely and at scale. Microsoft adds new local embedding features for semantic search and retrieval augmented generation (RAG) in analytics and expands deep learning applications for autonomous vehicles through the cloud.
Local Embedding Generation in Fabric Eventhouse
Microsoft enables text embedding creation in the Kusto Python sandbox within Fabric Eventhouse using Small Language Models (SLMs) such as jina-v2-small and e5-small-v2, via the slm_embeddings_fl() function. Previously, developers needed Azure OpenAI endpoints for embeddings, which added dependency on remote APIs and could bring latency, cost, and privacy limitations. Now, local inference allows for lower overhead, reduced latency, and simpler compliance—improving scalability and automation for data processing teams. Documentation provides step-by-step KQL and Python examples for embedding creation, real-time search, and automated processing, supporting efficient, secure AI adoption in Azure environments.
Deep Learning for Autonomous Vehicles on Azure
Wayve leverages Azure for distributed training and large-scale deployment of deep learning models in autonomous vehicles, extending advanced ML into connected mobility. Azure's infrastructure supports big data handling and fast model rollout across GPU and TPU clusters, building on cloud-enabled AI operations for industrial applications.