Anomaly Detection via Tensor Probability Models
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing systems for predicting state changes and identifying anomalies fail to capture sufficient dimensions of data and relationships between them, leading to ineffective identification of anomalous circumstances that require intervention.
Innovation Solution
A novel system that learns rules from a corpus of past state transitions, representing continuous data as discrete tensors and using recurrent neural networks to identify anomalies by calculating probabilities of state transitions, allowing for automatic detection and response to anomalies.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If Markov models and probabilistic models are used to predict state transitions, then the system can identify some anomalous circumstances, but the models fail to capture sufficient dimensions of data and relationships between dimensions
Solution Approach 1:
The patent transforms continuous sensor data into discrete tensor representations, adding a temporal dimension by organizing data into sequences of states. This dimensional transformation allows the system to capture relationships across multiple time steps and dimensions simultaneously, overcoming the limitation of traditional models that cannot adequately represent multi-dimensional data relationships.
Solution Approach 2:
The system nests multiple levels of probabilistic modeling within a hierarchical framework, where lower-level sensor readings are aggregated into state representations, which are then used to predict higher-level state transitions. This nested structure enables the model to capture patterns at different scales and dimensions without losing information from individual sensor measurements.
2Reliability
If existing probabilistic models are used, then some state transitions can be predicted, but the models do not capture relationships between dimensions when dimensions are not fully independent
Solution Approach 1:
The patent segments the complex multi-dimensional state space into discrete, manageable tensor representations. By dividing continuous data into discrete states and organizing them into structured tensors, the system can model relationships between dimensions without requiring the dimensions to be fully independent, thus improving reliability while managing complexity through structured segmentation.
Solution Approach 2:
The system changes the parameter representation from continuous independent variables to discrete tensor states with inherent dimensional relationships. This parameter transformation allows the model to capture dependencies between dimensions through the tensor structure itself, eliminating the need for complex interaction terms while maintaining the ability to model non-independent relationships.
3Measurement precision
If a system reviews a corpus of past state transitions to learn rules, then the system can identify anomalies that humans might not detect, but the system complexity increases significantly
Solution Approach 1:
The system implements self-service through automated probabilistic modeling, where the system automatically learns state transition probabilities from historical data and uses these learned models to detect anomalies without human intervention. This self-service capability enables the system to identify patterns and anomalies that human operators might miss, while the automated nature manages complexity through algorithmic processing rather than manual rule specification.
Solution Approach 2:
The system creates simplified copies of complex real-world processes by representing continuous sensor data as discrete tensor states. This copying approach allows the system to work with manageable symbolic representations that capture essential patterns without requiring the full complexity of continuous physical processes, thus improving anomaly detection while controlling system complexity through abstraction.
Data Source
AI summary
A system and method for constructing a probability model and automatically responding to process anomalies identified by the probability model are disclosed. Data is received for current and prior states of a process, comprising variables in at least two dimensions, and the at least two dimensions being not independently and identically distributed. A segment of a fixed number of prior states is selected and fed into a neural network to output a probability vector for each of the two or more dimensions. The Cartesian product of these probability vectors is calculated to obtain a tensor, wherein each value in the tensor represents a probability that the prior states would be followed by a given state. If the probability in the tensor associated with the present state is less than a predetermined threshold, an electronic communication is automatically generated and transmitted to a client computing device.