Out-of-domain detection for improved ai performance

Patent metadata

Field	Value
Publication number	US-2023099938-A1
Application number	US-202117449298-A
Country	US
Kind code	A1
Filing date	Sep 29, 2021
Priority date	Sep 29, 2021
Publication date	Mar 30, 2023
Grant date	—

Abstract

Systems and methods for determining input data is out-of-domain of an AI (artificial intelligence) based system are provided. Input data for inputting into an AI based system is received. An in-domain feature space of the AI based system and an out-of-domain feature space of the AI based system are modelled. The in-domain feature space corresponds to features of data that the AI based system is trained to classify. The out-of-domain feature space corresponds to features of data that the AI based system is not trained to classify. Probability distribution functions in the in-domain feature space and the out-of-domain feature space are generated for the input data and for the data that the AI based system is trained to classify. It is determined whether the input data is out-of-domain of the AI based system based on the probability distribution functions for the input data and for the data that the AI based system is trained to classify.

First claim

1 . A method comprising:
receiving input data for inputting into an AI (artificial intelligence) based system; modelling an in-domain feature space of the AI based system and an out-of-domain feature space of the AI based system, the in-domain feature space corresponding to features of data that the AI based system is trained to classify and the out-of-domain feature space corresponding to features of data that the AI based system is not trained to classify; generating probability distribution functions, in the in-domain feature space and the out-of-domain feature space, for the input data and for the data that the AI based system is trained to classify; and determining whether the input data is out-of-domain of the AI based system based on the probability distribution functions for the input data and for the data that the AI based system is trained to classify.

2 . The method of claim 1 , wherein modelling an in-domain feature space of the AI based system and an out-of-domain feature space of the AI based system comprises:
computing the in-domain feature space based on one or more in-domain linear projection matrices of the AI based system; computing one or more orthogonal linear projection matrices for the out-of-domain feature space based on the one or more in-domain linear projection matrices of the AI based system; and computing the out-of-domain feature space based on the one or more orthogonal linear projection matrices for the out-of-domain feature space.

3 . The method of claim 1 , wherein generating probability distribution functions, in the in-domain feature space and the out-of-domain feature space, for the input data and for the data that the AI based system is trained to classify comprises:
generating the probability distribution functions using a Gaussian process model or a combination of Gaussian probability distribution models fitted to available data.

4 . The method of claim 1 , further comprising:
in response to determining that the input data is out-of-domain of the AI based system, transmitting a notification to a user for reviewing a prediction generated by the AI based system from the input data.

5 . The method of claim 1 , further comprising:
in response to determining that the input data is out-of-domain of the AI based system:
annotating the input data; and
training the AI based system based on the annotated input data.

6 . The method of claim 1 , further comprising:
in response to determining that the input data is not out-of-domain of the AI based system, generating a prediction from the input data by the AI based system.

7 . The method of claim 1 , wherein the receiving, the modelling, the generating, and the determining are performed by a module combined with the AI based system and the module combined with the AI based system generates a prediction from the input data and the determination of whether the input data is out-of-domain of the AI based system.

8 . The method of claim 1 , further comprising:
selecting one of a plurality of algorithms of the AI based system based on the determining.

9 . The method of claim 1 , wherein the AI based system is for medical imaging analysis.

10 . An apparatus comprising:
means for receiving input data for inputting into an AI (artificial intelligence) based system; means for modelling an in-domain feature space of the AI based system and an out-of-domain feature space of the AI based system, the in-domain feature space corresponding to features of data that the AI based system is trained to classify and the out-of-domain feature space corresponding to features of data that the AI based system is not trained to classify; means for generating probability distribution functions, in the in-domain feature space and the out-of-domain feature space, for the input data and for the data that the AI based system is trained to classify; and means for determining whether the input data is out-of-domain of the AI based system based on the probability distribution functions for the input data and for the data that the AI based system is trained to classify.

11 . The apparatus of claim 10 , wherein the means for modelling an in-domain feature space of the AI based system and an out-of-domain feature space of the AI based system comprises:
means for computing the in-domain feature space based on one or more in-domain linear projection matrices of the AI based system; means for computing one or more orthogonal linear projection matrices for the out-of-domain feature space based on the one or more in-domain linear projection matrices of the AI based system; and means for computing the out-of-domain feature space based on the one or more orthogonal linear projection matrices for the out-of-domain feature space.

12 . The apparatus of claim 10 , wherein the means for generating probability distribution functions, in the in-domain feature space and the out-of-domain feature space, for the input data and for the data that the AI based system is trained to classify comprises:
means for generating the probability distribution functions using a Gaussian process model or a combination of Gaussian probability distribution models fitted to available data.

13 . The apparatus of claim 10 , further comprising:
means for transmitting a notification to a user for reviewing a prediction generated by the AI based system from the input data in response to determining that the input data is out-of-domain of the AI based system.

14 . The apparatus of claim 10 , wherein the receiving, the modelling, the generating, and the determining are performed by a module combined with the AI based system and the module combined with the AI based system generates a prediction from the input data and the determination of whether the input data is out-of-domain of the AI based system.

15 . A non-transitory computer readable medium storing computer program instructions, the computer program instructions when executed by a processor cause the processor to perform operations comprising:
receiving input data for inputting into an AI (artificial intelligence) based system; modelling an in-domain feature space of the AI based system and an out-of-domain feature space of the AI based system, the in-domain feature space corresponding to features of data that the AI based system is trained to classify and the out-of-domain feature space corresponding to features of data that the AI based system is not trained to classify; generating probability distribution functions, in the in-domain feature space and the out-of-domain feature space, for the input data and for the data that the AI based system is trained to classify; and determining whether the input data is out-of-domain of the AI based system based on the probability distribution functions for the input data and for the data that the AI based system is trained to classify.

16 . The non-transitory computer readable medium of claim 15 , wherein modelling an in-domain feature space of the AI based system and an out-of-domain feature space of the AI based system comprises:
computing the in-domain feature space based on one or more in-domain linear projection matrices of the AI based system; computing one or more orthogonal linear projection matrices for the out-of-domain feature space based on the one or more in-domain linear projection matrices of the AI based system; and computing the out-of-domain feature space based on the one or more orthogonal linear projection matrices for the out-of-domain feature space.

17 . The non-transitory computer readable medium o

Assignees

• Siemens Healthcare Gmbh

Inventors

• Georgescu Bogdan
• Gibson Eli
• Ghesu Florin-Cristian
• Comaniciu Dorin
• Jacob Athira Jane
• Passerini Tiziano
• Sharma Puneet

Classifications

• G06V2201/03

  Recognition of patterns in medical or anatomical images · CPC title

• G06V10/778

  Active pattern-learning, e.g. online learning of image or video features · CPC title

• G06N3/0464 Primary

  Convolutional networks [CNN, ConvNet] · CPC title

• G06V10/77

  Processing image or video features in feature spaces; using data integration or data reduction, e.g. principal component analysis [PCA] or independent component analysis [ICA] or self-organising maps [SOM]; Blind source separation · CPC title

• G06V10/763

  Non-hierarchical techniques, e.g. based on statistics of modelling distributions · CPC title