Layered models for context awareness
AssigneeMicrosoft Corporation
InventorsNuria M. Oliver, Eric J. Horvitz, Ashutosh Garg
Life of this patent
Estimating the term…
Abstract
The present invention relates to a system and methodology providing layered probabilistic representations for sensing, learning, and inference from multiple sensory streams at multiple levels of temporal granularity and abstraction. The methods facilitate robustness to subtle changes in environment and enable model adaptation with minimal retraining. An architecture of Layered Hidden Markov Models (LHMMs) can be employed having parameters learned from stream data and at different periods of time, wherein inferences can be determined relating to context and activity from perceptual signals.
Claims48
A learning system for signal processing, comprising:
- N classification layers associated with MN probabilistic models per layer, N and M being integers, respectively;
- at least one parameter defined per a respective classification layer, the parameters trained independently at different classification layers;
- and
- at least one input stream that is analyzed by the N classification layers, the at least one input stream sampled according to varying levels of temporal granularity at the respective classification layers.
A method to facilitate learning in a sampled data system, comprising:
- determining a plurality of parameters for a layered classification system;
- connecting at least two layers of the layered classification system, via an inferential result from a lower classification layer to a higher classification layer;
- training the plurality of parameters that are configured on different layers of the layered classification system independently;
- and
- determining at least one model state from the layered classification system.
A method to facilitate signal processing in an intelligent system, comprising:
- applying a plurality of sensor signals at a lowest layer of a probabilistic hierarchy;
- determining at least one feature vector for a subsequent layer of the probabilistic hierarchy;
- employing a higher layer of the probabilistic hierarchy to discriminate data;
- and
- associating longer temporal concepts with higher layers of the probabilistic hierarchy.
A method to facilitate real time signal processing, comprising:
- defining a multi-layer classification system;
- defining one or more probabilistic models per a respective layer of the multi-layer classification system;
- defining different sample rates per the respective layer;
- feeding inferential outputs from previous layers to subsequent layers of the multi-layer classification system;
- and
- inferring multiple states from analytical results derived from previous layers.
An analysis system executable on a computer, comprising:
- at least one processing component at a lowest layer of a classification system to analyze multiple signal input streams;
- at least one feature vector indicating results from a portion of the lowest layer;
- and at least one behavioral model executable on a highest layer of the classification system to infer human behavior based at least in part on the feature vector.
A system to facilitate computer signal processing, comprising:
- means for configuring a multi-layer classification system;
- means for training one or more probabilistic models per a respective layer of the multi-layer classification system;
- means for providing various time sample rates per the respective layer;
- means for feeding inferential outputs from previous layers to subsequent layers of the multi-layer classification system;
- and
- means for inferring multiple human states from analytical results derived from previous layers.
A signal to communicate sensor data between at least two nodes, comprising:
- a first data packet comprising:
- an N layered classification component employed to communicate probabilistic results between at least two layers of N layered classification system;
- and
- at least one parameter defined per a respective classification layer, the parameters trained independently at different classification layers;
- and
- at least one input stream that is analyzed by the N classification layers, the at least one input stream sampled according to varying levels of temporal granularity at the respective classification layers.
Dependencies are inferred from the claim text.