首页|'Real-Time Ensemble Evaluation' in Patent Application Approval Process (USPTO 20 240232726)
'Real-Time Ensemble Evaluation' in Patent Application Approval Process (USPTO 20 240232726)
扫码查看
点击上方二维码区域，可以放大扫码查看
原文链接
NETL
NSTL
This patent application has not been assigned to a company or institution.The following quote was obtained by the news editors from the background informa tion supplied by theinventors: “The present invention relates generally to the field of machine learning, and more particularlyto mitigating model drift.“Machine learning models are susceptible to model drift, where existing models b ecome increasinglyineffective (i.e., model accuracy reductions) due to data cha nges over time as new data is incorporated.Traditional systems deploy one or mo re models and retrain said models once new data deviates (i.e., drifts) from the original training set, although detecting and determining drift is complex and computationally expensivefor many problem domains. For example, a financial for ecasting model that predicts next quarterlyrevenue cannot retrain until the fis cal quarter passes and actual revenue is observed and transformed intoassociate d labels/predictions. Models that cannot dynamically incorporate new data become outdatedand fail to generalize future data, decreasing the overall effectivene ss of the models and the system as awhole. Traditionally, as new data is incorp orated into new, retrained, models, said models degrade (i.e.,drift) due to the removal of relevant data in previous models, affecting the performance of the e nsemble.To avoid this drift, systems perform general classifier evaluation meas ures to evaluate model performancecorresponding to a specific historical period , but within the historical period evaluation measures arestatic, (i.e., insens itive to real-time status and conditions), and highly influenced by hyperparamet ers.This insensitivity to real-time events decreases ensemble/model accuracy an d generalizability.
NETL
NSTL
