Key-Value Memory Network for Time-Series Prediction Without Historical Data
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional automated modeling systems for predicting time-series metrics face challenges in making accurate predictions when insufficient historical data is available, as they rely on learning from past metrics of the same target entity, limiting their ability to predict metrics for which no historical data exists.
Innovation Solution
The use of key-value memory networks that normalize time-series data based on correlations with comparative entities and combine static and time-specific features to generate predictions, allowing for accurate predictions without requiring extensive historical data.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If conventional automated modeling systems use recurrent neural networks trained on historical time-series data to predict metrics, then prediction accuracy is improved when sufficient historical data is available, but the systems cannot make accurate predictions for entities without sufficient historical data
Solution Approach 1:
The patent introduces comparative entities as intermediaries to bridge the gap when historical data for the target entity is insufficient. The system selects comparative entities with sufficient historical data, extracts their time-series metrics, and uses these as proxy data to train the prediction model, enabling accurate predictions for entities lacking their own historical data
Solution Approach 2:
The prediction system is designed to handle multiple scenarios universally: it can predict metrics for entities with sufficient historical data using their own data, and for entities without sufficient historical data by using comparative entities' data. This multi-functional capability allows the same system to adapt to different data availability conditions
2Reliability
If the system uses entities' own historical time-series metrics for training, then the model learns entity-specific patterns, but the system requires sufficient historical data that may not be available for new or recently introduced entities
Solution Approach 1:
The system creates copies of time-series data from comparative entities to serve as training data for target entities. By copying and adapting historical patterns from similar entities, the system bypasses the need for large amounts of original historical data while maintaining reliable model training
Data Source
AI summary
A system implements a key value memory network including a key matrix with key vectors learned from training static feature data and time-series feature data, a value matrix with value vectors representing time-series trends, and an input layer to receive, for a target entity, input data comprising a concatenation of static feature data of the target entity, time-specific feature data, and time-series feature data for the target entity. The key value memory network also includes an entity-embedding layer to generate an input vector from the input data, a key-addressing layer to generate a weight vector indicating similarities between the key vectors and the input vector, a value-reading layer to compute a context vector from the weight and value vectors, and an output layer to generate predicted time-series data for a target metric of the target entity by applying a continuous activation function to the context vector and the input vector.