Decoupled Encoder-Decoder Models for Flexible Multi-Type Computing
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing machine-trained models require significant computing resources and maintenance due to their end-to-end task execution, limiting flexibility and scalability.
Innovation Solution
A decoupled model architecture where encoder-type models convert input data into embeddings and decoder-type models convert embeddings into output data, sharing a common vector space, allowing dynamic selection and combination of models for specific tasks.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If end-to-end machine-trained models are used for complete task execution, then task performance is achieved, but computing resource requirements increase significantly
Solution Approach 1:
The patent divides end-to-end machine-trained models into separate encoder-type models and decoder-type models that operate independently. Encoder models convert input data to embeddings, while decoder models convert embeddings to output data. This segmentation allows the system to use only the specific encoder and decoder needed for each task, reducing overall computing resource requirements compared to maintaining complete end-to-end models for all tasks.
2Adaptability or versatility
If multiple end-to-end machine-trained models are maintained for different tasks, then task diversity is supported, but model maintenance complexity increases
Solution Approach 1:
The patent creates a universal embedding space that serves as a common interface between encoder and decoder models. Multiple encoder models for different input data types (text, image, audio) and multiple decoder models for different output data types all interact through this shared embedding space. This universality allows the system to support diverse tasks by combining different encoders and decoders without increasing maintenance complexity, as the embedding space provides a standardized interaction protocol.
3Adaptability or versatility
If encoder and decoder models are decoupled and dynamically combined, then system flexibility improves, but model selection and coordination complexity increases
Solution Approach 1:
The patent introduces an embedding space as an intermediary layer between encoder and decoder models. This embedding space acts as a mediator that standardizes the interaction interface, allowing different encoder and decoder models to be dynamically combined without direct complex coordination. The embedding space translates between different data types and model representations, simplifying the selection and coordination process while maintaining system flexibility.
Data Source
AI summary
A technique executes tasks using a data store of machine-trained models. The data store specifically includes a subset of encoder-type machine-trained models for converting input data items having different input data types into respective embeddings in a vector space, and a subset of decoder-type machine-trained models for converting embeddings in the same vector space into data items having respective different output data types. When executing a particular task that involves one or more data types, the technique selects one or more machine-trained models that match those data types. In some implementations, the technique provides a clipboard store for storing embeddings produced by the encoder-type machine-trained models and consumable by the decoder-type machine-trained models. The technique includes provisions for ensuring that any decoder-type machine-model is capable of processing embeddings produced by different versions of the encoder-type machine-trained models.