Systems and methods for correcting error in a first classifier by evaluating classifier output in parallel

A method for classification of a test object, the method comprising:

• (A) obtaining a description of the test object;
• (B) for each respective target object in a first plurality of target objects, performing a first procedure comprising:
• (i) posing the description of the test object against the respective target object thereby obtaining a description of an interaction between the test object and the respective target object, and
• (ii) inputting the description of the interaction between the test object and the respective target object to a first classifier thereby obtaining a corresponding score for the interaction between the test object and the respective target object from the first classifier, wherein
• each corresponding score for the interaction between the test object and a respective target object across the first plurality of target objects forms a test vector for the test object;
• and
• (C) inputting the test vector for the test object to a second classifier thereby obtaining a transformation for the test vector as output from the second classifier, wherein
• the transformation provides an indication of a single target object in the first plurality of target objects,
• the second classifier is trained on a plurality of training vectors,
• each respective training vector in the plurality of training vectors is the output from the first classifier after inputting a corresponding training object in a first plurality of training objects as a test object in accordance with the first procedure,
• each training object in a first subset of the first plurality training objects is uniquely associated with a corresponding target object in the first plurality of target objects, and
• each training object in a second subset of the first plurality of training objects is associated with no target object in the first plurality of target objects.

A non-transitory computer-readable storage medium having stored thereon instructions, which, when executed by a processor in a system for classification of a test object, cause the processor to perform the operations of:

• (A) obtaining a description of the test object;
• (B) for each respective target object in a first plurality of target objects, performing a first procedure comprising:
• (i) posing the description of the test object against the respective target object thereby obtaining a description of an interaction between the test object and the respective target object, and
• (ii) inputting the description of the interaction between the test object and the respective target object to a first classifier thereby obtaining a corresponding score for the interaction between the test object and the respective target object from the first classifier, wherein
• each corresponding score for the interaction between the test object and a respective target object across the first plurality of target objects forms a test vector for the test object;
• and
• (C) inputting the test vector for the test object to a second classifier thereby obtaining a transformation for the test vector as output from the second classifier, wherein
• the transformation provides an indication of a single target object in the first plurality of target objects,
• the second classifier is trained on a plurality of training vectors,
• each respective training vector in the plurality of training vectors is the output from the first classifier after inputting a corresponding training object in a first plurality of training objects as a test object in accordance with the first procedure,
• each training object in a first subset of the first plurality training objects is uniquely associated with a corresponding target object in the first plurality of target objects, and
• each training object in a second subset of the first plurality of training objects is associated with no target object in the first plurality of target objects.