Machine learning process and method of targeted clustering based on single shot sampling
A method for improving an accuracy of object classification, the method includes: (i) receiving, by a machine learning process, information regarding an environment of a vehicle; (ii) classifying, by the machine learning process, an object that is located within the environment of the vehicle to a certain class; wherein the machine learning process was trained by a training process to classify objects while avoiding false positive (FP) errors that are represented by different FP data sets that are fed to the machine learning process during the training process, the different FP data sets are associated with different classes; wherein for each class of the different classes, a FP data set that is associated with the class comprises FP data sub-sets that are associated with different objects that were mistakenly classified to the class; and outputting an outcome of the classification for use in navigating the vehicle.
1 . A method that is computer implemented and is for improving an accuracy of object classification, the method comprises:
receiving, by a machine learning process, information regarding an environment of a vehicle; classifying, by the machine learning process, an object that is located within the environment of the vehicle to a certain class; wherein the machine learning process was trained by a training process to classify objects while avoiding false positive (FP) errors that are represented by different FP data sets that are fed to the machine learning process during the training process, the different FP data sets are associated with different classes; wherein for each class of the different classes, a FP data set that is associated with the class comprises FP data sub-sets that are associated with different objects that were mistakenly classified to the class; and
outputting an outcome of the classification for use in navigating the vehicle wherein each FP data set that is associated with a class is generated by: obtaining first signatures that were found to be first FP signature during a supervised learning process and are associated with the class; obtaining second FP signatures that were generated during an unsupervised learning process and were found to be similar to the first FP signatures; and including sensed information units (SIUs) represented by the first FP signatures and SIUs represented by the second FP signatures within the FP data set.
2 . The method according to claim 1 , wherein the classifying comprises avoiding making the FP errors that are represented by the different FP data sets.
3 . The method according to claim 1 , comprising finding at least one new FP error.
4 . The method according to claim 3 , comprising retraining the machine learning process with a new data set that represents the new FP error.
5 . The method according to claim 1 , comprising executing the training process.
6 . A non-transitory computer readable medium storing instructions that, when executed by at least one processor, cause the at least one processor to perform operations for improving an accuracy of object classification, comprising:
receiving, by a machine learning process, information regarding an environment of a vehicle; classifying, by the machine learning process, an object that is located within the environment of the vehicle to a certain class; wherein the machine learning process was trained by a training process to classify objects while avoiding false positive (FP) errors that are represented by different FP data sets that are fed to the machine learning process during the training process, the different FP data sets are associated with different classes; wherein for each class of the different classes, a FP data set that is associated with the class comprises FP data sub-sets that are associated with different objects that were mistakenly classified to the class; and outputting an outcome of the classification for use in navigating the vehicle, wherein each FP data set that is associated with a class is generated by: obtaining first signatures that were found to be first FP signature during a supervised learning process and are associated with the class; obtaining second signatures that were generated during an unsupervised learning process and were found to be similar to the first signatures; and including sensed information units (SIUs) represented by the first signatures and SIUs represented by the second signatures within the FP data set.
7 . The non-transitory computer readable medium according to claim 6 , wherein the classifying comprises avoiding making the FP errors that are represented by the different FP data sets.
8 . The non-transitory computer readable medium according to claim 6 , further storing instructions for finding at least one new FP error.
9 . The non-transitory computer readable medium according to claim 8 , further storing instructions for retraining the machine learning process with a new data set that represents the new FP error.
10 . The non-transitory computer readable medium according to claim 6 , further storing instructions for executing the training process.
11 . A method for improving an accuracy of object classification, the method comprises: receiving different false positive (FP) data sets that are associated with different classes; wherein for each class of the different classes, a FP data set that is associated with the class comprises FP data sub-sets that are associated with different objects that were mistakenly classified to the class; and training a machine learning process to classify objects while avoiding FP errors represented by the FP data sets, wherein the training comprises feeding the machine learning process with the FP data sets
wherein each FP data set that is associated with a class is generated by: obtaining first signatures that were found to be first FP signature during a supervised learning process and are associated with the class; obtaining second FP signatures that were generated during an unsupervised learning process and were found, during an iterative process to belong to FP clusters having FP cluster signatures that are similar to the first signatures; and including sensed information units (SIUs) represented by the first signatures and SIUs represented by the second signatures within the FP data set.