An Overview of Machine Learning within Embedded and Mobile Devices-Optimizations and Applications

Area of Science:

• Computer Science
• Electrical Engineering
• Artificial Intelligence

Background:

• Embedded systems are evolving with advances in computer architecture and machine learning.
• Embedded machine learning (EML) has applications in computer vision, speech recognition, healthcare, and robotics.
• Resource constraints in embedded and mobile devices pose challenges for computationally intensive ML algorithms.

Purpose of the Study:

• To survey current research trends in embedded machine learning.
• To explore optimization techniques for implementing ML algorithms in resource-limited environments.
• To provide an overview of EML applications and future research directions.

Main Methods:

• Overview of compute-intensive machine learning algorithms (HMM, k-NN, SVM, GMM, DNN).
• Analysis of optimization techniques for resource-constrained environments.
• Discussion of implementation on microcontrollers, mobile devices, and hardware accelerators.

Main Results:

• Identification of key machine learning algorithms and their computational demands.
• Exploration of various optimization strategies for efficient EML.
• Examination of EML implementation across different hardware platforms.

Conclusions:

• EML requires innovative optimization techniques at both algorithmic and hardware levels.
• Key application areas and future research directions in EML are highlighted.
• Lessons learned provide insights for future exploration in the embedded machine learning domain.