MLaaS (ML as a Service): The design of an AI platform to accelerate ML adoption and the implementation of some NLP tasks for efficient document catalog management

• Mauro Gattari (Istituto Nazionale di Fisica Nucleare)

We propose the design of an AI platform that provides services in the cloud to accelerate ML adoption. Services are exposed via HTTP(S) endpoints and offer the following functionalities: - train/fine-tune machine learning models; - host/share datasets and trained models in the cloud; - use models to make predictions about new data; - manage models and versions through a public catalog. We provide service implementations for some common NLP tasks (text-classification, masked language modeling, semantic search) using Transformer-based Language Models. These NLP tasks are being integrated into a Web Application devoted to manage INFN research products (publications, thesis, talks).