Unsupervised programmatic labeling for transitional interval in smart logistics
AssigneeDell Products L.P.
InventorsHerberth Birck Fröhlich, Paulo Abelha Ferreira, Vinicius Michel Gottin
Life of this patent
Estimating the term…
Abstract
One example method includes collecting trajectories from each edge device in a group of edge devices, extracting a respective latent space vector from each trajectory, evaluating the latent space vectors to identify a respective set of one or more transitional intervals corresponding to each of the latent space vectors, and labeling the transitional intervals. The trajectories include information about the movement of the edge devices within a domain, and the labels on the transitional intervals indicate whether, and to what extent, the movement of the edge devices conforms to expected movements.
Claims20
A method, comprising:
- collecting trajectories from each edge device in a group of edge devices, wherein each of the trajectories includes time-series data;
- extracting a respective latent space vector from each trajectory, wherein the latent space vector comprises a compressed feature vector representing temporal characteristics of the trajectories;
- evaluating the latent space vectors to identify a respective set of one or more transitional intervals corresponding to each of the latent space vectors by clustering the latent space vectors;
- signaling to the edge devices that a central node is prepared to perform transitional interval identification using the identified set of one or more transitional intervals, wherein the transitional intervals correspond to anomalous edge device behavior;
- collecting new latent space vectors generated at the edge devices;
- labelling the new latent space vectors programmatically using the clustered set of one or more transitional intervals;
- determining that a new latent space vector is a detected transitional interval when the new latent space vector moves outside of a first boundary associated with the set of one or more transitional intervals and subsequently returns within the first boundary, wherein the first boundary is a statistically determined distance from a centroid of a cluster of the latent space vectors received from the edge devices;
- performing a remedial action at the edge devices in response to a detected transitional interval when the detected transitional interval is ongoing and when the detected transitional interval is completed.
A non-transitory storage medium having stored therein instructions that are executable by one or more hardware processors to perform operations comprising:
- collecting trajectories from each edge device in a group of edge devices, wherein each of the trajectories includes time-series data;
- extracting a respective latent space vector from each trajectory, wherein the latent space vector comprises a compressed feature vector representing temporal characteristics of the trajectories;
- evaluating the latent space vectors to identify a respective set of one or more transitional intervals corresponding to each of the latent space vectors by clustering the latent space vectors;
- signaling to the edge devices that a central node is prepared to perform transitional interval identification using the identified set of one or more transitional intervals, wherein the transitional intervals correspond to anomalous edge device behavior;
- collecting new latent space vectors generated at the edge devices;
- labelling the new latent space vectors programmatically using the clustered set of one or more transitional intervals;
- determining that a new latent space vector is a detected transitional interval when the new latent space vector moves outside of a first boundary associated with the set of one or more transitional intervals and subsequently returns within the first boundary, wherein the first boundary is a statistically determined distance from a centroid of a cluster of the latent space vectors received from the edge devices;
- performing a remedial action at the edge devices in response to a detected transitional interval when the detected transitional interval is ongoing and when the detected transitional interval is completed.
Dependencies are inferred from the claim text.