UCLA team releases OpenSLA, unifying sensor time series, language and actions
yang_yuzhe · x · 2026-10-08
Yuzhe Yang's lab at UCLA introduces OpenSLA, a new class of Sensor-Language-Action models that unify sensor time series, natural language, and action prediction, with paper, code and models released.
The idea: Most sensor AI stops at perception — recognizing states and events. OpenSLA uses a language interface to go further: from one sensor window plus individual context, the same model answers what is happening, what action should follow, and why, grounding each action in sensor evidence.
Scale and capabilities:
- Spans 7 datasets, 79 sensor modalities, 116K individuals, 60 action groups — from clinical care to everyday CGM monitoring
- Multi-level action prediction (necessity, category, fine-grained label), with zero-shot discrimination of unseen actions
- State understanding: recovers rates, ranges and waveform patterns directly from raw signals
- Transfers to the external MIMIC-IV clinical cohort without fine-tuning; frozen representations read out glucose dynamics over the next two hours
Related event: UCLA Team Open-Sources OpenSLA: Unifying Sensors, Language, and Action(7 posts)→
More from Research
- Medical AI benchmark builder: his first ECG dataset had answers printed on images — MaziyarPanahi · 2026-10-08
- Learned latent dynamics roll forward from few observations to recover hopper state with uncertainty — hisspikeness · 2026-10-08
- Adding a pixel-reconstruction loss makes latents unable to decode hopper state — hisspikeness · 2026-10-08
- New theory shows SSL separates stochastic signals from nuisance via MI maximization plus distribution matching — hisspikeness · 2026-10-08
- The conundrum of latent prediction: signals themselves are stochastic, so what to keep? — hisspikeness · 2026-10-08
- Hippocampal cognitive maps support language learning, not just spatial navigation — abenitezburraco · 2026-10-08