Neural Network Internal State Search and Classification Reuse
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Deep learning using neural networks is time-consuming to train and discards transient information during use, limiting the effective utilization of neural network outputs.
Innovation Solution
Capture and store the internal state representation of neural networks during processing, which can be used for search tasks or later machine learning, such as classification, by deriving feature representations from neural network output activations.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If a full neural network model is trained and used, then accurate inference can be performed, but it takes a long time to train and use
Solution Approach 1:
The patent extracts and stores transient internal state representations (activations) from the neural network during inference, separating this intermediate information from the full model processing. This allows the extracted features to be reused for multiple tasks without re-running the complete neural network, thereby reducing time loss while preserving accuracy for the extracted features.
2Quantity of substance
If only the output of the neural network is saved, then storage is minimized, but transient information is discarded and cannot be reused
Solution Approach 1:
The patent recovers transient internal state representations that would otherwise be discarded during neural network inference. By capturing and storing these activations at intermediate layers, the system recovers valuable transient information that can be reused for multiple downstream tasks such as search and classification, transforming lost information into a reusable resource.
3Adaptability or versatility
If internal state representations are captured and stored, then information can be reused for search and classification, but additional storage is required
Solution Approach 1:
The patent makes the internal state representations universal by designing them to serve multiple functions simultaneously. The same captured activations are used for both search tasks (finding similar inputs) and classification tasks (categorizing inputs), as well as potential retraining scenarios. This multi-functionality increases adaptability while justifying the storage investment through reuse across different operations.
Data Source
AI summary
Systems and methods are disclosed for generating internal state representations of a neural network during processing and using the internal state representations for classification or search. In some embodiments, the internal state representations are generated from the output activation functions of a subset of nodes of the neural network. The internal state representations may be used for classification by training a classification model using internal state representations and corresponding classifications. The internal state representations may be used for search, by producing a search feature from an search input and comparing the search feature with one or more feature representations to find the feature representation with the highest degree of similarity.