Adaptive Loss Function Derivation for Neural Forecaster Training
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Classical forecasting algorithms are limited by their parametric and structure-driven nature, which restricts their representation capabilities and generalization, making them less effective in high data variety and volume contexts.
Innovation Solution
A computer-implemented method that uses a library of selected loss functions and business specification rules to generate a custom composite loss function for deep learning AI models, allowing for adaptive loss function derivation tailored to specific forecasting needs.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If a single loss function is used for training the deep learning model, then the training process is simple and fast, but the forecasting accuracy cannot optimally capture contextual business requirements
Solution Approach 1:
The patent combines multiple loss functions (e.g., RMSE, MAE, Pinball loss) into a single composite loss function that integrates different error metrics. This merging allows the model to simultaneously optimize for multiple forecasting objectives and contextual requirements, thereby improving forecasting accuracy while maintaining a unified training process
Solution Approach 2:
The patent introduces dynamic weighting mechanisms that allow the relative importance of different loss functions to change based on business context, data characteristics, and forecasting objectives. This dynamic adaptation enables the composite loss function to flexibly prioritize different error metrics for different business scenarios, capturing contextual requirements that static single loss functions cannot
2Adaptability or versatility
If classical parametric forecasting algorithms are used, then the model structure is simple and interpretable, but the representation capability is restricted and generalization is poor in high data variety and volume contexts
Solution Approach 1:
The patent creates a universal deep learning forecasting framework that can handle diverse data types, volumes, and business contexts through a single flexible architecture. The model uses a standardized deep learning structure with configurable loss functions that can adapt to various forecasting scenarios, eliminating the need for dedicated models for each data type while maintaining high generalization capability
Solution Approach 2:
The patent leverages the large number of free parameters in deep learning models to capture complex patterns in high-volume, high-variety data. By adjusting these parameters during training through the composite loss function, the model adapts to different data characteristics and business requirements, achieving superior generalization compared to rigid parametric algorithms
3Adaptability or versatility
If multiple loss functions are combined with different weights, then the forecasting can capture multiple business objectives, but the determination of appropriate weights becomes complex
Solution Approach 1:
The patent implements self-service mechanisms where the model automatically learns optimal weights for different loss functions through the training process itself. The composite loss function is designed to allow the system to autonomously determine the relative importance of different error metrics based on the data and business context, eliminating the need for manual weight tuning by practitioners
Data Source
AI summary
A computer-implemented method for forecasting a future value of one or more elements of a time-series of data includes obtaining a time-series of data, obtaining a library having a plurality of selected loss functions, obtaining at least one Business Specification Rule (BSR), each BSR including a Context, a Metric and a Priority, for each selected loss function, generating input-associated perturbated outputs based on the BSRs and the time-series of data by training a deep learning artificial intelligence (DLAI) model to learn a set of learned weights to be given to each of the selected loss functions, deriving a custom composite loss function based on the sets of learned weights for the plurality of selected loss functions in the LFL, and using the custom composite loss function to train a final DLAI model on the time-series of data. The final DLAI model may then be used to forecast future outcomes.