Neural Event Modeling With Negative Evidence for Causal Graph Learning
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing event modeling techniques fail to fully utilize the information in non-event intervals between consecutive events, limiting the effectiveness of statistical and causal relationship learning in multivariate event streams.
Innovation Solution
A multi-channel neural graphical event model (MCN-GEM) that incorporates negative evidence by introducing fake epochs representing non-event intervals, utilizing long short-term memory (LSTM) cells and spatial and temporal attention models to generate causal graphs between events.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Loss of information
If existing event modeling techniques are used, then the model structure is simple, but the information in non-event intervals is not fully utilized, limiting learning effectiveness
Solution Approach 1:
The patent introduces fake epochs (negative evidence) in advance between consecutive event epochs to represent non-event intervals. This preliminary action allows the model to capture information about periods when events did not occur, enabling better learning of inter-event dynamics and causal relationships without requiring complex architectural changes.
Solution Approach 2:
The patent uses fake epochs as an intermediary representation to bridge the gap between discrete event observations. These fake epochs serve as mediators that carry information about non-event intervals, allowing the LSTM network to process and learn from both event and non-event information in a unified framework.
2Reliability
If more information from non-event intervals is utilized, then the learning of statistical and causal relationships is enhanced, but the computational complexity increases
Solution Approach 1:
The patent applies partial action by introducing a limited number of fake epochs (e.g., one or a few) between consecutive event epochs rather than continuously sampling throughout all non-event intervals. This partial representation captures the essential information about non-event periods while avoiding the excessive computational burden of dense sampling, achieving a balance between learning effectiveness and computational complexity.
3Loss of information
If fake epochs representing negative evidence are introduced, then the utilization of non-event information is improved, but the data processing complexity increases
Solution Approach 1:
The patent merges the processing of real event epochs and fake epochs into a single unified LSTM network framework. Both types of epochs are processed through the same hidden state transitions and attention mechanisms, allowing the model to learn from mixed evidence (both event and non-event) without requiring separate processing pipelines, thereby reducing overall data processing complexity.
Data Source
AI summary
A computer-implemented method is presented for learning relationships between multiple event types by employing a multi-channel neural graphical event model (MCN-GEM). The method includes receiving, by a computing device, time-stamped, asynchronous, irregularly spaced event epochs, generating, by the computing device, at least one fake epoch between each inter-event interval, wherein fake epochs represent negative evidence, feeding the event epochs and the at least one fake epoch into long short term memory (LSTM) cells, computing hidden states for each of the event epochs and the at least one fake epoch, feeding the hidden states into spatial and temporal attention models, and employing an average attention across all event epochs to generate causal graphs representing causal relationships between all the event epochs.