Symbolic model training with active learning

Patent metadata

Field	Value
Publication number	US-12165019-B2
Application number	US-202017132776-A
Country	US
Kind code	B2
Filing date	Dec 23, 2020
Priority date	Dec 23, 2020
Publication date	Dec 10, 2024
Grant date	Dec 10, 2024

Abstract

Techniques regarding generating and/or training one or more symbolic models are provided. For example, one or more embodiments described herein can comprise a system, which can comprise a memory that can store computer executable components. The system can also comprise a processor, operably coupled to the memory, and that can execute the computer executable components stored in the memory. The computer executable components can comprise a training component that can train a symbolic model via active machine learning. The symbolic model can characterize a formal planning language for a planning domain as a plurality of digital image sequences.

First claim

What is claimed is:

1. A system, comprising:
a memory that stores computer executable components; and
a processor, operably coupled to the memory, and that executes at least one of the computer executable components that:
trains, via active learning, using a set of training data comprising digital images and plans, a machine learning model to generate symbolic models that characterize the plans as digital image sequences;
based on feedback received regarding accuracy the symbolic models, generates additional training data comprising negative data samples, wherein the negative data samples comprise false digital image sequences;
adds the additional training data to the set of training data; and
retrains, via the active learning, using the set of training data, the machine learning model, to generate the symbolic models that characterize the plans as the digital image sequences.

2. The system of claim 1 , wherein a digital image sequence characterize an order of actions, and wherein the set of training data comprises positive data samples comprising true digital image sequences that succeed at achieving respective goal states associated with corresponding plans.

3. The system of claim 2 , wherein a false digital image sequence comprises an order of actions that fails to achieve a goal state associated with a corresponding plan.

4. The system of claim 3 , wherein the at least one of the computer executable components further:
determines a probability that a target digital image sequence from the digital image sequences is a true digital image sequence based on identification of the false digital image sequence.

5. The system of claim 1 , wherein the at least one of the computer executable components further:
generates a subset of the plans based on an action schema and an initial state represented by a digital image from the set of training data, wherein the subset of the plans define respective orders of actions.

6. The system of claim 5 , wherein the generating employs a domain-independent planner to generate the subset of the plans.

7. The system of claim 1 , wherein the at least one of the computer executable components further:
selects a subset of plans from the plans based on a probability that the subset of plans includes a false digital image sequence.

8. The system of claim 7 , wherein the at least one of the computer executable components further generates a request for the false digital image sequence to be labelled.

9. The system of claim 8 , wherein at least one of the computer executable components further:
updates the set of training data to include the false digital image sequence with a label based on the false digital image sequence being labelled false.

10. A computer-implemented method, comprising:
training, by a system operatively coupled to a processor, via active learning, using a set of training data comprising digital images and plans, a machine learning model to generate symbolic models that characterize the plans as digital image sequences;
based on feedback received regarding accuracy the symbolic models, generating, by the system, additional training data comprising negative data samples, wherein the negative data samples comprise false digital image sequences;
adding, by the system, the additional training data to the set of training data; and
retraining, by the system, via the active learning, using the set of training data, the machine learning model, to generate the symbolic models that characterize the plans as the digital image sequences.

11. The computer-implemented method of claim 10 , wherein a digital image sequence characterize an order of actions, and wherein the set of training data comprises positive data samples comprising true digital image sequences that succeed at achieving respective goal states associated with corresponding plans and wherein the false digital image sequence comprises an order of actions that fails to achieve a goal state associated with a corresponding plan.

12. The computer-implemented method of claim 10 , further comprising:
generating, by the system, a subset of the plans based on an action schema and an initial state represented by a digital image from the set of training data, wherein the subset of the plans define respective orders of actions.

13. The computer-implemented method of claim 10 , further comprising:
selecting, by the system, a subset of plans from the plans based on a probability that the subset of plans includes a false digital image sequence.

14. The computer-implemented method of claim 13 , further comprising:
generating, by the system, a request for the false digital image sequence to be labelled.

15. The computer-implemented method of claim 14 , further comprising:
updating, by the system, the set of training data to include the false digital image sequence with a label based on the false digital image sequence being labelled false.

16. A computer program product for enhancing an accuracy of symbolic models, the computer program product comprising a computer readable storage medium having program instructions embodied therewith, the program instructions executable by a processor to cause the processor to:
train, via active learning, using a set of training data comprising digital images and plans, a machine learning model to generate symbolic models that characterize the plans as digital image sequences;
based on feedback received regarding accuracy the symbolic models, generate additional training data comprising negative data samples, wherein the negative data samples comprise false digital image sequences;
add the additional training data to the set of training data; and
retrain, via the active learning, using the set of training data, the machine learning model, to generate the symbolic models that characterize the plans as the digital image sequences.

17. The computer program product of claim 16 , wherein a digital image sequence characterize an order of actions, and wherein the set of training data comprises positive data samples comprising true digital image sequences that succeed at achieving respective goal states associated with corresponding plans and wherein the false digital image sequence comprises an order of actions that fails to achieve a goal state associated with a corresponding plan.

18. The computer program product of claim 17 , wherein the program instructions further cause the processor to:
generate a subset of the plans based on an action schema and an initial state represented by a digital image from the set of training data, wherein the subset of the plans define respective orders of actions.

19. The computer program product of claim 17 , wherein the program instructions further cause the processor to:
select a subset of plans from the plans based on a probability that the subset of plans includes a false digital image sequence.

20. The computer program product of claim 19 , wherein the program instructions further cause the processor to:
generate a request for the false digital image sequence to be labelled; and
update the set of training data to include the false digital image sequence with a label based on the false digital image sequence being labelled false;
training, by a system operatively coupled to a processor, via active learning, using a set of training data comprising digital images and plans, a machine learning model to generate symbolic models that characterize the plans as digital image sequences;
based on feedback received regarding accuracy the symbo

Assignees

• IBM

Inventors

• Kishimoto Akihiro
• Asai masataro
• Hou Yufang
• Kajino Hiroshi
• Marinescu Radu

Classifications

• G06N3/091

  Active learning · CPC title

• G06N3/0455

  Auto-encoder networks; Encoder-decoder networks · CPC title

• G06N3/0499

  Feedforward networks · CPC title

• G06N3/0895

  Weakly supervised learning, e.g. semi-supervised or self-supervised learning · CPC title

• G06N3/09

  Supervised learning · CPC title