The increasing prevalence of AI-enabled devices is revolutionizing society by harnessing the power of artificial intelligence to enhance a wide range of applications. However, when dealing with resource-constrained devices like smartphones and embedded systems, limitations and challenges in implementing AI can be faced. To gain a comprehensive understanding of AI implementation in resource-constrained devices, a deep investigation of each aspect is crucial. This work seeks to develop strategies and techniques that address the challenges specific to these devices in both their local and offloading execution. With the correct optimization of these design dimensions, such as resources, timeliness, availability, privacy, and accuracy, the proposed solutions aim to enhance the performance and usability of AI on resource-constrained devices. This involves exploring techniques like model compression, lightweight algorithms, efficient data representation, and intelligent resource allocation strategies. The goal is to enable resource-constrained devices to leverage AI capabilities effectively without compromising on performance, energy efficiency, or user experience. The proposed solution aims to empower these devices with intelligent functionalities, enabling them to handle a wide range of applications while operating within their limita- tions, contributing to the advancement of AI-enabled technologies that can be seamlessly integrated into everyday life. The methodology proposed in this thesis uses a graphical tool to exploit the 5 dimensions of AI and have the best recommendation of the most efficient and practical deployment typology to use.
La crescente diffusione di dispositivi con IA integrata sta rivoluzionando la società, sfrut- tandone le potenzialità per migliorare le sue applicazioni nel settore industriale e in- formatico. Tuttavia possono presentarsi alcune sfide e limitazioni nell’implementazione efficace dell’IA quando si tratta di dispositivi con risorse limitate, come gli smartphone e i sistemi embedded. Per identificare le best-practise da attuare quando si tratta di integrare l’Intelligenza Artificiale è fondamentale indagare in modo approfondito ogni suo aspetto. Attraverso la comprensione della complessità di ciascuna dimensione, questo lavoro cerca di sviluppare strategie e tecniche che affrontino le sfide specifiche di questi dispositivi sia nell’esecuzione in locale che attraverso offloading. Ottimizzando attentamente queste dimensioni, come resources, timeliness, availability, privacy e accuracy, viene proposto come migliorare le prestazioni e la fruibilità dell’IA su dispositivi con risorse limitate. Ciò comporta l’esplorazione di tecniche come model compression, lightweight algorithms, efficient data representation, e intelligent resource allocation strategies. L’obiettivo è quindi consentire ai dispositivi di sfruttare efficacemente le capacità dell’IA senza com- prometterne le prestazioni, l’efficienza energetica o l’esperienza dell’utente. Pur operando all’interno dei loro limiti, le soluzioni proposte mirano a dotare questi dispositivi di funzionalità intelligenti, consentendo loro di gestire un’ampia gamma di applicazioni. Ciò per contribuire al progresso delle tecnologie con IA che possono essere integrate nella vita di tutti i giorni, dando strumenti efficaci agli utenti e consentendo applicazioni innovative in vari settori.
Dimensions of AI-enabled mobile and embedded devices : an integration design guide
FASANELLA, MARCO
2022/2023
Abstract
The increasing prevalence of AI-enabled devices is revolutionizing society by harnessing the power of artificial intelligence to enhance a wide range of applications. However, when dealing with resource-constrained devices like smartphones and embedded systems, limitations and challenges in implementing AI can be faced. To gain a comprehensive understanding of AI implementation in resource-constrained devices, a deep investigation of each aspect is crucial. This work seeks to develop strategies and techniques that address the challenges specific to these devices in both their local and offloading execution. With the correct optimization of these design dimensions, such as resources, timeliness, availability, privacy, and accuracy, the proposed solutions aim to enhance the performance and usability of AI on resource-constrained devices. This involves exploring techniques like model compression, lightweight algorithms, efficient data representation, and intelligent resource allocation strategies. The goal is to enable resource-constrained devices to leverage AI capabilities effectively without compromising on performance, energy efficiency, or user experience. The proposed solution aims to empower these devices with intelligent functionalities, enabling them to handle a wide range of applications while operating within their limita- tions, contributing to the advancement of AI-enabled technologies that can be seamlessly integrated into everyday life. The methodology proposed in this thesis uses a graphical tool to exploit the 5 dimensions of AI and have the best recommendation of the most efficient and practical deployment typology to use.File | Dimensione | Formato | |
---|---|---|---|
23_07_Marco_Fasanella.pdf
accessibile in internet per tutti
Descrizione: Dimensions of AI-enabled Mobile and Embedded Devices: an Integration Design Guide
Dimensione
2.57 MB
Formato
Adobe PDF
|
2.57 MB | Adobe PDF | Visualizza/Apri |
I documenti in POLITesi sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.
https://hdl.handle.net/10589/208961