Search

EP-3686812-B1 - SYSTEM AND METHOD FOR CONTEXT-BASED TRAINING OF A MACHINE LEARNING MODEL

EP3686812B1EP 3686812 B1EP3686812 B1EP 3686812B1EP-3686812-B1

Inventors

  • LORE, Kin Gwn
  • REDDY, Kishore K.

Dates

Publication Date
20260513
Application Date
20200121

Claims (14)

  1. A method of training a machine learning model (24) for discriminating between nominal or excessive engine loading, the machine learning model (24) including a classifier (26) and a knowledge bank of autoencoders (28), the knowledge bank including at least one context autoencoder (AE1; AE2...AEn), the method comprising: receiving input data from at least one remote device (18) of an engine, the at least one remote device (18) is a sensor and includes an accelerometer, a temperature detector, a pressure detector, and/or a locational device; evaluating the classifier (26) by determining a classification accuracy of the input data; jointly training a selected, neural network-based context autoencoder (AEs) of the knowledge bank of autoencoders (28) for contextual inference and the classifier (26) for a classification task by: applying a training data matrix of the input data to the selected context autoencoder (AEs) and determining the training data matrix of the input data is out of context for the selected context autoencoder (AEs); applying the training data matrix to each other context autoencoder (AE1; AE2... AEn) of the at least one context autoencoder (AE1; AE2... AEn) and determining the training data matrix is out of context for each other context autoencoder (AE1; AE2... AEn); introducing a contextual variable as a flag for a new context and the contextual variable is part of the training data matrix as inputs to the classifier (26); and constructing a new context autoencoder consistent with the new context.
  2. The method of claim 1, further comprising storing the new context autoencoder with the knowledge bank of autoencoders (28).
  3. The method of claim 1 or 2, further comprising initializing the new context autoencoder.
  4. The method of any preceding claim, further comprising applying a semantic meaning to the new context autoencoder.
  5. The method of any preceding claim, wherein determining the input data is out of context includes determining a reconstruction error for a respective one of the at least one context autoencoder (AE1, AE2... AEn).
  6. The method of claim 5, wherein the input data is out of context when P ϵ − μ ϵ σ ϵ < k .
  7. The method of any preceding claim, wherein the input data is streaming data from the at least one remote device (18).
  8. The method of any preceding clam, wherein the construction of a new context autoencoder induces an alarm.
  9. A system (10) for context-based training of a machine learning model (24) for discriminating between nominal or excessive engine loading, the machine learning model (24) including a classifier (26) and a knowledge bank of autoencoders (28), the knowledge bank including at least one context autoencoder (AE1; AE2...AEn), the system (10) comprising a memory unit (14) configured to store data and processor-executable instructions; a processor unit (13) in communication with the memory unit (14), the processor unit (14) configured to execute the processor-executable instructions stored in the memory unit (14) to: receive streaming data from at least one remote device (18) of an engine, the at least one remote device (18) being a sensor and including an accelerometer, a temperature detector, a pressure detector, and/or a location device; evaluate the classifier (26) by determining a classification accuracy of the streaming data; jointly train a selected, neural network-based context autoencoder (AEs) of the knowledge bank of autoencoders (28) for contextual inference and the classifier (26) for a classification tank by: applying a training data matrix of the streaming data to the selected context autoencoder (AEs) and determining the training data matrix is out of context for the selected context autoencoder (AEs); applying the training data matrix to each other context autoencoder (AE1; AE2... AEn) of the at least one autoencoder (AE1; AE2... AEn) and determining the training data matrix is out of context for each other context autoencoder (AE1; AE2... AEn); introducing a contextual variable as a flag for a new context and the contextual variable is part of the training data matrix as inputs to the classifier (26); and constructing a new context autoencoder consistent with the new context.
  10. The system (10) of claim 9, further comprising at least one sensor (18) mounted to a vehicle, wherein the least one remote device (18) is the at least one sensor (18).
  11. The system (10) of claim 9 or 10, wherein the processor unit (13) is configured to execute the processor-executable instructions stored in the memory unit (14) to store the new context autoencoder with the knowledge bank of autoencoders (28).
  12. The system (10) of any of claims 9 to 11, wherein the processor unit (13) is configured to execute the processor-executable instructions stored in the memory unit (14) to initialize the new context autoencoder.
  13. The system (10) of any of claims 9 to 12, further comprising a user interface (20) in communication with the processor unit (13), the user interface (20) configured to apply a semantic meaning provided by a user (22) to the new context autoencoder.
  14. The system (10) of any of claims 9 to 13, wherein determining the input data is out of context includes determining a reconstruction error with a respective one of the at least one context autoencoder (AE1; AE2... AEn).

Description

BACKGROUND 1. Technical Field This disclosure relates to machine learning, and more particularly to context-based training of machine learning models. 2. Background Information Machine learning is a process used to analyze data in which the dataset is used to determine a model that maps input data to output data. For example, neural networks are machine learning models that employ one or more layers of nonlinear units to predict an output for a received input. Some neural networks include one or more hidden layers in additional to an output layer. The output of each hidden layer is used as input to the next layer in the network, i.e., the next hidden layer or the output layer. Each layer of the network generates an output from a received input in accordance with current values of a respective set of parameters. In offline learning, one can inspect historical training data to identify contextual clusters either through feature clustering or hand-crafting additional features to describe a context. While offline training enjoys the privilege of learning reliable models based on already-defined contextual features, online training (i.e., training in real time) for streaming data may be more challenging. For example, the underlying context during a machine learning process may change resulting in an under-performing model being learned due to contradictory evidence observed in data within high-confusion regimes. Furthermore, the problem is exacerbated when the number of possible contexts is not known. US 2016/155136 A1 discloses a diagnostic system for model governance that includes an auto-encoder to monitor model suitability for both supervised and unsupervised models. When applied to supervised models, the system can determine the most appropriate model based on a reconstruction error of a trained auto-encoder. The auto-encoder can determine outliers among subpopulations of data and support model inspections. Timothy Wong et al. ("Recurrent Auto-Encoder Model for Large-Scale Industrial Sensor Signal Analysis," 2018, arXiv:1807.03710v1) describe a recurrent auto-encoder model for analyzing sensor signals from a large-scale industrial gas turbine engine. The system processes streams of data from sensors, including temperature, pressure, and vibration detectors. The model encodes multidimensional time-series data into a fixed-length context vector, which is then analyzed using unsupervised clustering algorithms to identify underlying operating states of the engine. An alarm can be triggered when a generated context vector travels beyond the boundary of a predefined neighborhood. SUMMARY From one aspect there is provided a method of training a machine learning model is provided as recited in claim 1. There is also provided a system for context-based training of a machine learning model as recited in claim 9. Features of embodiment are set forth in the dependent claims. The present disclosure, and all its aspects, embodiments and advantages associated therewith will become more readily apparent in view of the detailed description provided below, including the accompanying drawings. BRIEF DESCRIPTION OF THE DRAWINGS FIG. 1 is a block diagram illustrating an exemplary information processing system in which one or more aspects of the present disclosure may be implemented.FIG. 2 is a block diagram of a machine learning model.FIG. 3 is a flow chart of an exemplary method for context-based training of a machine learning model.FIG. 4 is a flow chart of an exemplary method for context-based training of a machine learning model. DETAILED DESCRIPTION It is noted that various connections are set forth between elements in the following description and in the drawings. It is noted that these connections are general and, unless specified otherwise, may be direct or indirect and that this specification is not intended to be limiting in this respect. A coupling between two or more entities may refer to a direct connection or an indirect connection. An indirect connection may incorporate one or more intervening entities. It is further noted that various method or process steps for embodiments of the present disclosure are described in the following description and drawings. The description may present the method and/or process steps as a particular sequence. However, to the extent that the method or process does not rely on the particular order of steps set forth herein, the method or process should not be limited to the particular sequence of steps described. As one of ordinary skill in the art would appreciate, other sequences of steps may be possible. Therefore, the particular order of the steps set forth in the description should not be construed as a limitation. Referring to FIG. 1, a machine learning system 10 includes a computing device 12 having a processing unit 13 operatively coupled to a memory unit 14. The processing unit 13 is one or more devices configured to execute instructions for software and/or firmware. For exa