Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

An AI-Native Runtime for Multi-Wearable Environments

Chulhong Min, Utku Günay Acer, SiYoung Jang, Sangwon Choi, Diana A. Vasile, Taesik Gong, Juheon Yi, Fahim Kawsar

TL;DR

Mojito, an AI-native runtime with advanced MLOps designed to facilitate the development and deployment of next-generation wearable applications within wearable technologies, demonstrates how future wearable technologies can evolve to be more autonomous.

Abstract

The miniaturization of AI accelerators is paving the way for next-generation wearable applications within wearable technologies. We introduce Mojito, an AI-native runtime with advanced MLOps designed to facilitate the development and deployment of these applications on wearable devices. It emphasizes the necessity of dynamic orchestration of distributed resources equipped with ultra-low-power AI accelerators to overcome challenges associated with unpredictable runtime environments. Through its innovative approaches, Mojito demonstrates how future wearable technologies can evolve to be more autonomous.

An AI-Native Runtime for Multi-Wearable Environments

TL;DR

Mojito, an AI-native runtime with advanced MLOps designed to facilitate the development and deployment of next-generation wearable applications within wearable technologies, demonstrates how future wearable technologies can evolve to be more autonomous.

Abstract

The miniaturization of AI accelerators is paving the way for next-generation wearable applications within wearable technologies. We introduce Mojito, an AI-native runtime with advanced MLOps designed to facilitate the development and deployment of these applications on wearable devices. It emphasizes the necessity of dynamic orchestration of distributed resources equipped with ultra-low-power AI accelerators to overcome challenges associated with unpredictable runtime environments. Through its innovative approaches, Mojito demonstrates how future wearable technologies can evolve to be more autonomous.
Paper Structure (17 sections, 4 figures)

This paper contains 17 sections, 4 figures.

Table of Contents

  1. Introduction
  2. TECHNOLOGY TRENDS
  3. Miniaturizing AI Accelerators
  4. Exponential Growth of AI models for Wearables
  5. Changing Paradigm of Wearable Applications
  6. CHALLENGES FOR WEARABLE COLLABORATION
  7. Bottleneck of Ultra-low-power AI Accelerators: Memory
  8. Tight Time Synchronization
  9. Heterogeneous and Dynamic Runtime Environment
  10. Mojito: An Overview
  11. VIRTUAL COMPUTING SPACE
  12. Application Programming Interface
  13. AI ACCELERATOR ORCHESTRATION
  14. Exploring Untapped Opportunities
  15. On-the-fly Device-to-Device Authentication
  16. ...and 2 more sections

Figures (4)

  • Figure 1: (a) MAX78000 and future wearable computing (left), (b) Comparison of AI accelerators accelerators (middle), (c) Latency and energy comparison between AI accelerator (MAX78000) and microcontrollers (MAX32650 and STM32F7) (right).
  • Figure 2: Accuracy variation according to quantization bit (1, 2, 4, 8) size of EfficientNetV2 and MobileNetV2.
  • Figure 3: (a) Illustrative example of virtual computing space (left), (b) Throughput comparison with state-of-the-art method; W1: ConvNet, ResSimpleNet, UNet, W2: KeywordSpotting, SimpleNet, WideNet, W3: EfficientNetV2 (right).
  • Figure 4: Temperature measurement of Raspberry Pi Zero (left) and Google Coral Micro with AI processing (right).