Graph-Attention Temporal Network for Predictive Accuracy
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing predictive data analysis solutions face inefficiencies in capturing complex structural correlations and temporal dependencies of features in temporal sequences, leading to limitations in predictive accuracy and training speed.
Innovation Solution
The use of a graph-attention augmented temporal network that generates representative embeddings for temporal sequences through dynamic co-occurrence graphs and initial embeddings, leveraging a sequential long short-term memory model and a tree-based long short-term memory model to improve predictive accuracy and efficiency.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional predictive data analysis methods are used, then the system is simpler to implement, but predictive accuracy deteriorates due to inability to capture complex structural correlations and temporal dependencies
Solution Approach 1:
The patent segments the feature processing into multiple embedding layers (initial embedding layer and subsequent embedding layers), where each layer processes temporal sequences at different levels of abstraction. This segmentation allows the model to capture both local temporal patterns and global structural correlations, resolving the contradiction by dividing the complex processing task into manageable segments that collectively improve predictive accuracy.
Solution Approach 2:
The patent introduces dynamic co-occurrence graphs as an additional dimensional representation of feature relationships, complementing the temporal sequence data. This graph-based dimension captures structural correlations between features, while the temporal sequences capture temporal dependencies. By adding this dimensional perspective, the model achieves higher predictive accuracy without requiring excessive complexity in the temporal processing alone.
2Measurement precision
If more computational operations and training data entries are used, then predictive accuracy improves, but training speed and computational efficiency deteriorate
Solution Approach 1:
The patent performs preliminary action by pre-computing and storing feature co-occurrence statistics in the form of dynamic co-occurrence graphs before the main training process. During training, the model directly utilizes these pre-computed graphs rather than computing co-occurrences from scratch, significantly reducing computational operations and training time while maintaining the ability to capture complex structural correlations for accurate predictions.
Solution Approach 2:
The patent creates simplified copies of the full feature space through dynamic co-occurrence graphs, which are projections of the global feature correlation structure. These graph representations serve as compressed summaries that capture essential structural relationships without requiring the model to process all原始 feature interactions, thereby reducing computational operations while preserving predictive accuracy.
3Measurement precision
If the model processes all features in the global co-occurrence graph, then comprehensive structural correlations are captured, but computational and storage efficiency deteriorate
Solution Approach 1:
The patent applies local quality by creating dynamic co-occurrence graphs that are specific to each temporal sequence or data object, rather than using a single global graph for all data. Each dynamic graph contains only the feature co-occurrences relevant to that specific sequence, reducing storage requirements and computational overhead while still capturing the comprehensive structural correlations needed for accurate predictions on that particular data.
Solution Approach 2:
The patent introduces dynamics by making the co-occurrence graphs adaptive and sequence-specific rather than static and universal. The dynamic co-occurrence graphs are generated on-demand or updated based on the specific temporal sequence being processed, allowing the model to capture comprehensive structural correlations only when and where needed, thereby optimizing the balance between correlation capture and resource efficiency.
Data Source
AI summary
Various embodiments of the present invention provide methods, apparatus, systems, computing devices, computing entities, and/or the like for generating a representative embeddings for a plurality of temporal sequences by using a graph attention augmented temporal network based at least in part on dynamic co-occurrence graphs for preceding temporal sequences and initial embeddings, where the dynamic co-occurrence graphs are projections of a global guidance co-occurrence graph on features of the preceding temporal sequences, and the initial embeddings are generated by processing a latent representation of corresponding features that is generated by a sequential long short term memory model as well as a feature tree using a tree-based long short term memory model.