Transfer Learning for Analytics Prediction Model Adaptation

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

Existing media mix modeling systems are inflexible, inaccurate, and inefficient due to their reliance on rigid, data-driven models that fail to adapt to expected and predicted data channel contributions, requiring retraining for each new context and consuming excessive computational resources.

Innovation Solution

A transfer learning system that updates analytics prediction machine learning model parameters using an iterative process incorporating expected and predicted data channel contributions, allowing for flexible adaptation across different contexts without full retraining, and improving model accuracy and efficiency by leveraging prior knowledge and data-driven approaches.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Adaptability or versatility

If a rigid data-driven model is used for media mix modeling, then the model structure is simple and easy to implement, but the model lacks flexibility and accuracy when adapting to different contexts and data channel contributions

Engineering Contradiction:

Improveadaptability to different contextsVSAvoidmodel structure complexity

Core Design Contradiction:

Adaptability or versatilityVSDevice complexity

Solution Approach 1:

The patent applies transfer learning by pre-training a machine learning model on a source domain with abundant prior data before applying it to the target domain. This preliminary action allows the model to capture general patterns and relationships from historical data, enabling it to adapt to new contexts without requiring complete retraining, thus resolving the contradiction between adaptability and complexity.

Inventive Principle:

Principle #10Preliminary action

Solution Approach 2:

The patent employs parameter updating mechanisms that allow the model to adjust its parameters based on new contextual information and expected data channel contributions. By dynamically modifying model parameters rather than restructuring the entire model, the system achieves high adaptability while maintaining computational efficiency and avoiding excessive complexity.

Inventive Principle:

Principle #35Parameter changes

2Reliability

If a data-driven model is retrained for each new context, then the model accuracy is maintained, but the computational resources consumed are excessive and the process is inefficient

Engineering Contradiction:

Improvemodel accuracyVSAvoidcomputational resources

Core Design Contradiction:

ReliabilityVSUse of energy by moving object

Solution Approach 1:

The system performs preliminary training on a comprehensive dataset to create a foundational model that captures general relationships. This pre-training action establishes the basis for accuracy across multiple contexts, eliminating the need to retrain from scratch for each new context and significantly reducing computational resource requirements while maintaining high model accuracy.

Inventive Principle:

Principle #10Preliminary action

Solution Approach 2:

Instead of performing complete retraining for each new context, the patent applies partial updates to the pre-trained model parameters based on new contextual information and expected data channel contributions. This partial action approach maintains model accuracy for the specific context while consuming minimal computational resources compared to full retraining.

Inventive Principle:

Principle #16Partial or excessive action

3Measurement precision

If an iterative update process is used to incorporate expected and predicted data channel contributions, then the model flexibility and accuracy are improved, but the computational time and complexity increase

Engineering Contradiction:

Improveprediction accuracyVSAvoidcomputational time

Core Design Contradiction:

Measurement precisionVSLoss of time

Solution Approach 1:

The patent implements an iterative update process that incorporates feedback from expected data channel contributions and predicted analytics metrics. By continuously refining model parameters based on this feedback loop, the system improves prediction accuracy and flexibility. The iterative process is designed to converge efficiently, balancing enhanced measurement precision with acceptable computational time requirements.

Inventive Principle:

Principle #23Feedback

Data Source

PatentUS20240311643A1Generating analytics prediction machine learning models using transfer learning for prior data

Publication Date: 2024.09.19 ADOBE INC

AI summary

The present disclosure relates to systems, methods, and non-transitory computer readable media for generating a modified analytics prediction machine learning model using an iterative transfer learning approach. For example, the disclosed systems generate an initial version of an analytics prediction machine learning model for predicting an analytics metric according to learned parameters. In some embodiments, the disclosed systems determine expected data channel contributions for the analytics metric according to prior data. Additionally, in some cases, the disclosed systems generate a modified analytics prediction machine learning model by iteratively updating model parameters such that predicted data channel contributions are within a threshold similarity of expected data channel contributions.