AI Classifier Training With Tempered Token Weighting
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing classification systems overpredict the probability of an entity being in a particular class due to the presence of large numbers of low-weight indicators.
Innovation Solution
A system for training a classifier that uses a tempered indication based on a ratio of a weighted sum of tokens divided by a tempering factor, which reduces the impact of low-weight tokens through a normalized asymptotic transformation to minimize overprediction.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If prior art systems assign different weights to different indicators, then the system recognizes that some indicators are stronger than others, but when there is a large number of low weight indicators in an entity's publications, the system tends to over predict the probability that an entity is in a particular class
Solution Approach 1:
The patent applies parameter changes by modifying the probability calculation formula to include a tempering factor. The tempered indication is calculated as the ratio of the weighted sum of tokens to the tempering factor, where the tempering factor is a function of the unweighted sum of tokens. This parameter modification prevents overprediction by adjusting the influence of multiple low-weight indicators on the final probability score.
Solution Approach 2:
The tempering factor serves as an intermediary element between the weighted sum of indicators and the final probability prediction. By introducing this intermediate calculation step, the system mediates the effect of numerous low-weight indicators, preventing them from collectively causing overprediction while still allowing strong indicators to drive accurate class membership determination.
2Adaptability or versatility
If the system uses a large number of indicators to classify entities, then the classification coverage is improved, but the system tends to over predict the probability due to accumulation of low weight indicators
Solution Approach 1:
The patent modifies the probability calculation parameter by introducing the tempering factor that depends on the unweighted sum of tokens. This allows the system to maintain high classification coverage by considering multiple indicators while simultaneously correcting the probability prediction accuracy by tempering the cumulative effect of low-weight indicators through the ratio calculation.
Data Source
AI summary
An artificial intelligence system for training a classifier has a database of training data and a modeling system for building a classification model based on the training data. The database has a binary classification for each entity and binary tokens indicating whether or not one or more indicators about the entity are true. The classification model is based on a tempered indication of the tokens. The tempered indication is a ratio of a weighted sum of the tokens for each entity divided by a tempering factor for each of the entities. The tempering factor is a function of the unweighted sum of the tokens for each entity. Thus, the tempering factor will reduce the tempered indication when large numbers of low weight tokens are present so that the model does not over predict the probability of an entity being in the classification.