Time-Series Data Regularization for Consistent AI Training

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

Conventional time-series data processing lacks a standardized framework, leading to inconsistent data quality and poor performance in training machine learning models due to arbitrary choices in granularity, start times, and handling of null values.

Innovation Solution

A unified framework for regularizing time-series data by establishing a uniform timestep, using an objective vector to generate values for missing data, and removing excess data, resulting in a consistent and high-quality dataset.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Adaptability or versatility

If conventional ad-hoc processing methods are used for time-series data, then flexibility in handling diverse data formats is maintained, but data consistency and quality deteriorate

Engineering Contradiction:

Improveflexibility in handling diverse data formatsVSAvoiddata consistency

Core Design Contradiction:

Adaptability or versatilityVSReliability

Solution Approach 1:

The system changes parameters by establishing a uniform timestep parameter and applying it consistently across all time-series data processing. This standardization transforms the processing approach from ad-hoc to systematic, ensuring that all data undergoes the same regularization operations including handling of null values, resampling, and alignment, thereby improving data consistency while maintaining adaptability through configurable parameters

Inventive Principle:

Principle #35Parameter changes

Solution Approach 2:

The patent implements a universal processing framework that handles multiple data formats and scenarios through a single standardized procedure. The system universally applies the same regularization techniques (null value imputation, resampling, alignment) across different time-series datasets, eliminating the need for separate ad-hoc processing methods while maintaining the ability to handle diverse data types through configurable parameters

Inventive Principle:

Principle #6Universality (Multi-functionality)

2Productivity

If arbitrary choices are made regarding granularity and start times, then processing speed is maintained, but data quality and model training performance deteriorate

Engineering Contradiction:

Improveprocessing speedVSAvoiddata quality

Core Design Contradiction:

ProductivityVSMeasurement precision

Solution Approach 1:

The system performs preliminary actions by pre-establishing a uniform timestep and standardized processing parameters before actual data processing begins. This upfront configuration of processing parameters (granularity, start times, null value handling) eliminates the need for repeated arbitrary decisions during processing, maintaining high processing speed while ensuring consistent data quality through predetermined standards

Inventive Principle:

Principle #10Preliminary action

Solution Approach 2:

The system enables self-service by allowing the standardized processing framework to automatically handle all regularization operations without requiring manual intervention for parameter selection. The uniform timestep and standardized procedures allow the system to self-regulate processing parameters, maintaining both processing efficiency and data quality without arbitrary human decisions

Inventive Principle:

Principle #25Self-service

3Reliability

If excess data is removed from the dataset, then data consistency is improved, but information loss increases

Engineering Contradiction:

Improvedata consistencyVSAvoidinformation loss

Core Design Contradiction:

ReliabilityVSLoss of information

Solution Approach 1:

The system uses an intermediary approach by introducing a standardized uniform timestep as a mediator between the original irregular data and the processed consistent data. This intermediary framework allows the system to identify and remove only the excess data that violates the uniform timestep requirement, while preserving all other information. The intermediary standardization process ensures that data removal is systematic rather than arbitrary, maintaining data consistency while minimizing information loss through configurable parameters that control the aggressiveness of excess data removal

Inventive Principle:

Principle #24Intermediary (Mediator)

Data Source

PatentUS20260017559A1Systems and methods for regularizing time-series data

Publication Date: 2026.01.15 CAPITAL ONE SERVICES LLC

AI summary

Systems and methods for regularizing time-series data for training artificial intelligence models. The system receives a time-series dataset and an objective vector, including a temporal parameter and an interpolation parameter. The system identifies the timestep based on the time-series dataset and the temporal parameter. Based on the timestep, the system determines a start time of the time-series dataset. Based on the start time of the time-series dataset, the system identifies a set of timestamps and a set of values based on the interpolation parameter, each value corresponding to a timestamp. The system generates a modified time-series dataset by modifying the time-series dataset to include the set of timestamps and the set of values. The system then determines excess data in the time-series dataset and removes the excess data to generate the processed dataset. The system uses the processed dataset as training data to train a first machine learning model.