Deep-Learning Model Catalog Ontology Standardization
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
The development of deep-learning models is time-consuming and requires significant expertise, as they are typically manually created, and existing repositories lack standardization in information extraction, making it difficult for developers to identify and compare suitable models.
Innovation Solution
A system and method for generating a deep-learning model catalog by mining models from various sources, creating a consistent ontology for each model, and populating a pre-defined format with extracted information, including code and text parsing, to facilitate easy searching and comparison.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If deep-learning models are manually created, then model quality and customization are improved, but development time and expertise requirements increase
Solution Approach 1:
The patent creates a catalog that stores and reproduces information about existing deep-learning models, allowing developers to copy and adapt proven models rather than creating them from scratch. The system extracts and stores model specifications, architectures, and performance characteristics in a standardized format that can be reused.
Solution Approach 2:
The patent performs preliminary extraction and organization of model information in advance, creating a ready-to-use catalog before developers need to select models. The system pre-processes model data into standardized ontologies, so when developers search the catalog, the information is already structured and comparable.
2Adaptability or versatility
If information is extracted from multiple deep-learning model sources, then model availability increases, but information consistency deteriorates
Solution Approach 1:
The patent applies homogeneity by standardizing information from diverse deep-learning model sources into a unified ontology structure. The system extracts varied information formats and transforms them into consistent schemas, enabling uniform comparison across different models while maintaining broad model availability.
Solution Approach 2:
The patent introduces an intermediary standardized ontology layer between diverse model sources and the final catalog. This intermediary structure mediates the conversion of heterogeneous model information into a consistent format, preserving model diversity while ensuring information uniformity.
3Stability of the object's composition
If a standardized ontology format is created for each model, then information consistency is improved, but processing complexity increases
Solution Approach 1:
The patent segments the model information extraction and standardization process into distinct, manageable components. The system divides the complex ontology creation into separate steps: extracting information from code, extracting information from text, identifying operators, and populating the ontology format, making the overall process more tractable.
Solution Approach 2:
The patent creates a universal ontology format that serves multiple functions simultaneously: it stores model specifications, enables comparison, facilitates search, and supports model selection. This single standardized structure handles multiple information needs, reducing the complexity that would arise from creating separate formats for each purpose.
4Measurement precision
If developers manually analyze and compare models, then model selection accuracy is improved, but time consumption increases
Solution Approach 1:
The patent creates copyable, standardized model representations in the catalog that preserve all essential comparison information. Developers can efficiently copy and compare model specifications without manually re-analyzing each model, maintaining selection accuracy while reducing time investment.
Data Source
AI summary
One embodiment provides a method, including: mining a plurality of deep-learning models from a plurality of input sources; extracting information from each of the deep-learning models, by parsing at least one of (i) code corresponding to the deep-learning model and (ii) text corresponding to the deep-learning model; identifying, for each of the deep-learning models, operators that perform operations within the deep-learning model; producing, for each of the deep-learning models and from (i) the extracted information and (ii) the identified operators, an ontology comprising terms and features of the deep-learning model, wherein the producing comprises populating a pre-defined ontology format with features of each deep-learning model; and generating a deep-learning model catalog comprising the plurality of deep-learning models, wherein the catalog comprises, for each of the deep-learning models, the ontology corresponding to the deep-learning model.