Beimingwu: A Learnware Dock System

TL;DR

Beimingwu tackles data privacy, data scarcity, forgetting, and unplanned tasks by introducing a unified, privacy-preserving learnware dock that identifies and reuses high-performing models via standardized specifications. It delivers a four-layer architecture and an engine design that decouples algorithms from infrastructure, implementing RKME-based statistical and semantic specifications and supporting data-free and data-dependent reuse across heterogeneous feature spaces. End-to-end implementations cover submission, usability testing, organization, identification, deployment, and reuse, with extensive experiments on tabular, image, and text data demonstrating effectiveness under limited user data. This open-source foundation enables scalable, privacy-aware learnware ecosystems and lifelong learning in real-world, data-sensitive environments.

Abstract

The learnware paradigm proposed by Zhou [2016] aims to enable users to reuse numerous existing well-trained models instead of building machine learning models from scratch, with the hope of solving new user tasks even beyond models' original purposes. In this paradigm, developers worldwide can submit their high-performing models spontaneously to the learnware dock system (formerly known as learnware market) without revealing their training data. Once the dock system accepts the model, it assigns a specification and accommodates the model. This specification allows the model to be adequately identified and assembled to reuse according to future users' needs, even if they have no prior knowledge of the model. This paradigm greatly differs from the current big model direction and it is expected that a learnware dock system housing millions or more high-performing models could offer excellent capabilities for both planned tasks where big models are applicable; and unplanned, specialized, data-sensitive scenarios where big models are not present or applicable. This paper describes Beimingwu, the first open-source learnware dock system providing foundational support for future research of learnware paradigm.The system significantly streamlines the model development for new user tasks, thanks to its integrated architecture and engine design, extensive engineering implementations and optimizations, and the integration of various algorithms for learnware identification and reuse. Notably, this is possible even for users with limited data and minimal expertise in machine learning, without compromising the raw data's security. Beimingwu supports the entire process of learnware paradigm. The system lays the foundation for future research in learnware-related algorithms and systems, and prepares the ground for hosting a vast array of learnwares and establishing a learnware ecosystem.

Figures (9)

• Figure 1: A simplified process of learnware paradigm from Zhou:Tan2024. The basic operation can be decomposed into two stages: 1) Submitting stage: Developers worldwide can spontaneously submit their trained models to the learnware dock system, and the system assigns specification for each accepted model; 2) Deploying stage: The user submits her requirement to the learnware dock system, and then the system will identify and return some helpful learnwares to the user based on specifications, which can be further reused on user data.
• Figure 2: Practical codes for solving a learning task with Beimingwu. With just a few lines of code, a user can build a model for her limited data with the help of numerous learnwares in Beimingwu, without requiring extensive data and machine learning expertise, while not leaking her raw data.
• Figure 3: Overview of using Beimingwu to solve new learning tasks. The workflow consists of four steps: 1) Generating statistical specification: Beimingwu helps the user to generate statistical specification capturing the statistical property of the task without disclosing user's raw data; 2) Identify helpful learnwares: According to the submitted task requirement, Beimingwu can identify helpful learnware(s) from numerous learnwares for the user based on learnware specifications; 3) Loading learnwares: Beimingwu provides a unified way to load arbitrary learnwares effortlessly and safely; 4) Reuse learnwares: Beimingwu provides various baseline reuse algorithms in a unified interface to reuse learnwares on user data.
• Figure 4: Architecture of Beimingwu.
• Figure 5: Architecture design of Beimingwu engine. The architecture is illustrated from the perspectives of both modules and workflow.