Front-end deep learning web apps development and deployment: a review

Area of Science:

• Computer Science
• Artificial Intelligence

Background:

• Machine learning (ML) and deep learning (DL) models are traditionally deployed via servers or mobile apps.
• Emerging front-end technologies like TensorFlow.js allow ML model execution directly within web browsers.

Purpose of the Study:

• To review the development and deployment of deep learning web applications.
• To raise awareness of advancements in client-side ML and encourage adoption.
• To highlight the benefits of front-end ML, including improved user experience and privacy.

Main Methods:

• Discusses the rationale for using a front-end deployment stack (JavaScript, TensorFlow.js).
• Describes development approaches for optimizing deep learning models for front-end deployment.
• Reviews current web applications of front-end deep learning across seven categories.

Main Results:

• TensorFlow.js enables defining, training, and running ML models entirely in the browser.
• Client-side deployment offers improved user interaction and data privacy.
• Optimized models for front-end deployment require specific development considerations for size and inference speed.

Conclusions:

• Front-end deep learning via JavaScript libraries like TensorFlow.js is a rapidly advancing field.
• This technology offers significant potential for creating interactive, privacy-preserving ML applications.
• The review categorizes applications, demonstrating the broad applicability of front-end deep learning.