Dynamic Machine Learning Output Generation via Iterative Redefinition
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing machine learning solutions are inflexible and difficult to modify in real-time, making it challenging to incorporate new insights and collaborate across different models, and they face challenges in optimizing outputs within a reasonable timeframe.
Innovation Solution
A computer-implemented method using high-level language code that allows users to define only high-priority properties, with machine learning algorithms filling in undefined properties, enabling iterative optimization and collaboration through feedback loops involving both human input and automatic redefinition.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Stability of the object's composition
If machine learning solutions are built using traditional methods, then the model structure is fixed and stable, but it becomes difficult to modify in real-time and incorporate new insights
Solution Approach 1:
The patent applies dynamics by transforming the static, fixed machine learning model into a dynamic system where the model structure and parameters can be continuously modified in real-time. Users can add, remove, or modify nodes and connections in the computational graph, and the system automatically recompiles and retrainS the model to incorporate new insights and requirements.
2Reliability
If all properties of machine learning output are pre-defined in code, then the output generation is deterministic and controllable, but it requires complete specification and reduces ease of use
Solution Approach 1:
The patent applies partial action by allowing users to define only the essential or high-priority properties of the machine learning output in code, rather than requiring complete specification of all properties. The system then automatically infers, fills in, or allows flexible interpretation of the remaining properties, reducing the burden on users while maintaining control over critical aspects.
3Manufacturing precision
If machine learning models are optimized thoroughly, then the output performance is maximized, but the optimization time becomes excessively long
Solution Approach 1:
The patent applies periodic action by implementing iterative optimization cycles where the machine learning model is trained, evaluated, and retrained in repeated phases. Instead of performing a single exhaustive optimization, the system uses multiple shorter training cycles with feedback loops, allowing the model to progressively improve performance while maintaining reasonable timeframes through continuous refinement rather than one-time optimization.
Data Source
AI summary
A computer-implemented method for generating one or more outputs is disclosed. The method comprises providing code in a high-level language, the code comprising one or more statements defining one or more properties of a desired output; determining that one or more properties of the desired output are undefined in the code; defining at least one of the one or more undefined properties using a machine learning algorithm; generating an output based on the one or more properties defined in the code and the at least one property defined using the machine learning algorithm; and redefining at least one property of the output defined using the machine learning algorithm to generate a redefined output. At least one property defined using the machine learning algorithm is redefined automatically, and at least one property defined using the machine learning algorithm is redefined automatically based on an associated level of performance of one or more previous outputs.