A naukri.com initiative
Mongodb
3w
100
Image Credit: Mongodb
ORiGAMi: A Machine Learning Architecture for the Document Model
- ORiGAMi is a Transformer-based architecture designed for supervised learning on semi-structured data like JSON in a document model database.
- It addresses the challenges faced by the ML community in working with semi-structured formats compared to traditional tabular data.
- The architecture tokenizes documents into key-value pairs and structural tokens, making prediction directly from semi-structured documents possible.
- By training on datasets with as few as 200 labeled samples, ORiGAMi combines data efficiency with Transformer model flexibility.
- The model's token sequences serve as input for predicting the next token, ensuring valid document generation.
- ORiGAMi reformulates classification to predict any field within a document, eliminating the need for separate models or pipelines.
- Example use case includes user segmentation based on user profiles containing nested structures like device history and subscription details.
- With ORiGAMi, users can make predictions on raw documents, preserving nested structures and updating predictions as user behavior changes.
- The architecture is open-sourced on GitHub, with command-line interfaces for training models and making predictions seamlessly.
- ORiGAMi provides a way for document-native machine learning, inviting users to explore, contribute, and apply it to real-world problems.
Read Full Article
6 Likes
For uninterrupted reading, download the app