Finite State Automata for Continuous Machine Learning Model Delivery
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current machine-learning infrastructures lack the ability to support continuous delivery of machine learning models, leading to inconsistencies and difficulties in building and maintaining models, especially in industries with legacy data challenges or multiple data sources, due to the lack of structure and documentation in the process.
Innovation Solution
Implementing a finite state automata (FSA) that provides a structured process for building, training, and deploying machine learning models, enabling continuous delivery by managing data ingestion, transformation, enrichment, analysis, and model evaluation, with features like data immutability, transformation, and log tracking.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If an open platform is used for building machine learning models, then users have flexibility in selecting code and building models, but the process lacks structure and consistency, making it difficult to maintain and reproduce models
Solution Approach 1:
The patent segments the machine learning model building process into distinct states (data ingestion, data preparation, model building, model evaluation, model deployment) within a finite state automata framework. Each state represents a specific phase with defined inputs, outputs, and transitions, providing structure to the previously unstructured open platform process while maintaining user flexibility within each state.
Solution Approach 2:
The patent introduces structured parameters and metadata that accompany data and models throughout the process. These parameters track the state, version, and properties of datasets and models, enabling consistent reproduction and maintenance while allowing users flexibility in model building approaches within the structured framework.
2Adaptability or versatility
If multiple users work with multiple datasets and model versions simultaneously on an open platform, then collaboration is enabled, but consistency and repeatability of the process are lost
Solution Approach 1:
The finite state automata framework provides feedback mechanisms that track the state of each user's workflow, the versions of datasets and models being used, and the transitions between states. This feedback enables multiple users to collaborate simultaneously while maintaining consistent and repeatable processes through visible state tracking and version management.
Solution Approach 2:
The patent establishes predefined states and transition rules before users begin their work. These preliminary structural constraints ensure that regardless of how many users are working with different datasets and model versions, each user's process remains consistent and repeatable within the established framework.
3Ease of operation
If the machine learning process is maintained manually without automation, then users can exercise control over each step, but the process lacks observability and continuous delivery capability
Solution Approach 1:
The finite state automata acts as an intermediary between user control and process automation. It provides a structured framework that automatically tracks state transitions, logs process steps, and enables observability while still allowing users to exercise control over the modeling process within each state. The automata mediates between manual user actions and automated process tracking.
4Manufacturing precision
If data is transformed and enriched to meet model criteria, then model accuracy is improved, but the process becomes more complex and harder to track
Solution Approach 1:
The patent segments data transformation and enrichment operations into distinct states within the finite state automata (data ingestion state, data preparation state). Each state contains specific transformation operations with defined inputs and outputs, making the complex data processing pipeline trackable and manageable while improving model accuracy through systematic data preparation.
Data Source
AI summary
A finite state automata (FSA) may comprise of a plurality of states and a plurality of events that are triggered to transition between the plurality of states to enable the continuous delivery of one or more machine learning (ML) models. Datasets may be uploaded by one or more users to a computing device. Each dataset may include columns of attributes and/or rows of data for each attribute. Each dataset may be analyzed according to respective predefined ML model criteria. Each dataset may be automatically transformed and/or enhanced to meet the predefined ML model criteria. Data analysis may be performed on each dataset to inform the building of the ML models. An algorithm may be received for each ML model. Each ML model may be built based on the respective dataset and the respective algorithm to generate an ML model file. Logs may be stored for tracking the process performed by the FSA.