Newton: A Foundation Model for Sensor Intelligence at Scale
Session details:
Sensors are ubiquitous across physical industries – manufacturing, energy, transportation, logistics – yet up to 98% of collected sensor data remains unused. Despite its volume and richness, sensor data is notoriously hard to leverage: solutions are slow to build, costly to scale, and require bespoke machine learning models for each use case, device, and modality.
In this keynote, we present a new paradigm: Newton, a foundation model for the physical world, enabling rapid and generalizable intelligence over sensor data at scale. Newton learns directly from the physics of the world—observing real-world behaviors from sensor data and aligning them with human-interpretable semantics, bridging the gap between raw signal and actionable insight. The model is built on a universal embedding space architecture that transforms raw, multimodal sensor data into a unified representation, allowing generalization across sensor types, physical processes, and use cases. In particular, this architecture unlocks:
- Cross-sensor generalization: Shared representations across diverse data types (e.g., vibration, temperature, pressure, image)
- Zero- and few-shot adaptation: Infer patterns from unknown sensors or physical behaviors the model has never seen during training
- Sensor fusion at scale: Combine data from heterogeneous sources without custom engineering
- Semantic-physical alignment: Combine physically grounded representations of dynamic behavior with human-interpretable semantics using natural language interfaces
- End-user accessibility via natural language prompts: Democratize the deployment of physical AI
Unlike traditional ML pipelines, which are slow, costly, and fragile to develop, the Newton platform enables any engineer to build and deploy sensor-driven AI systems in minutes using natural language.
This talk will highlight the core technology and applications of Newton in real-world deployments. We’ll demonstrate how foundation models for sensor intelligence are transforming how enterprises harness their physical data—making physical AI easy, fast, and accessible.
