Change search
CiteExportLink to record
Permanent link

Direct link
Cite
Citation style
  • apa
  • ieee
  • modern-language-association-8th-edition
  • vancouver
  • Other style
More styles
Language
  • de-DE
  • en-GB
  • en-US
  • fi-FI
  • nn-NO
  • nn-NB
  • sv-SE
  • Other locale
More languages
Output format
  • html
  • text
  • asciidoc
  • rtf
Low-Power Classification using FPGA: An Approach based on Cellular Automata, Neural Networks, and Hyperdimensional Computing
Luleå University of Technology, Department of Computer Science, Electrical and Space Engineering, Embedded Internet Systems Lab. (Pervasive and Mobile Computing)ORCID iD: 0000-0002-8752-2375
Luleå University of Technology, Department of Computer Science, Electrical and Space Engineering, Embedded Internet Systems Lab.ORCID iD: 0000-0002-8216-832x
Luleå University of Technology, Department of Computer Science, Electrical and Space Engineering, Computer Science.ORCID iD: 0000-0002-6032-6155
Luleå University of Technology, Department of Computer Science, Electrical and Space Engineering, Computer Science.ORCID iD: 0000-0002-3437-4540
2019 (English)In: 2019 18th IEEE International Conference On Machine Learning And Applications (ICMLA) / [ed] M. Arif Wani, Taghi M. Khoshgoftaar, Dingding Wang, Huanjing Wang, Naeem (Jim) Seliya, IEEE, 2019, p. 370-375Conference paper, Published paper (Other academic)
Abstract [en]

Field-Programmable Gate Arrays (FPGA) are hardware components that hold several desirable properties for wearable and Internet of Things (IoT) devices. They offer hardware implementations of algorithms using parallel computing, which can be used to increase battery life or achieve short response-times. Further, they are re-programmable and can be made small, power-efficient and inexpensive. In this paper we propose a classifier targeted specifically for implementation on FPGAs by using principles from hyperdimensional computing and cellular automata. The proposed algorithm is shown to perform on par with Naive Bayes for two benchmark datasets while also being robust to noise. It is also synthesized to a commercially available off-the-shelf FPGA reaching over 57.1 million classifications per second for a 3-class problem using 40 input features of 8 bits each. The results in this paper show that the proposed classifier could be a viable option for applications demanding low power-consumption, fast real-time responses, or a robustness against post-training noise.

Place, publisher, year, edition, pages
IEEE, 2019. p. 370-375
Series
International Conference on Machine Learning and Applications (ICMLA)
Keywords [en]
low-power-classification, machine-learning, FPGA, hyperdimensional-computing, cellular-automata, resource-constrained-devices
National Category
Media and Communication Technology Other Electrical Engineering, Electronic Engineering, Information Engineering Computer Sciences
Research subject
Pervasive Mobile Computing; Electronic systems; Dependable Communication and Computation Systems
Identifiers
URN: urn:nbn:se:ltu:diva-71168DOI: 10.1109/ICMLA.2019.00069Scopus ID: 2-s2.0-85080900919ISBN: 978-1-7281-4550-1 (electronic)OAI: oai:DiVA.org:ltu-71168DiVA, id: diva2:1254869
Conference
2019 18th IEEE International Conference On Machine Learning And Applications (ICMLA); 16-19 Dec. 2019; Boca Raton, FL, USA
Available from: 2018-10-10 Created: 2018-10-10 Last updated: 2023-09-05Bibliographically approved
In thesis
1. Unobtrusive Activity Recognition in Resource-Constrained Environments
Open this publication in new window or tab >>Unobtrusive Activity Recognition in Resource-Constrained Environments
2018 (English)Doctoral thesis, comprehensive summary (Other academic)
Alternative title[sv]
Diskret Aktivitetsigenkänning i Resursbegränsade Miljöer
Abstract [en]

This thesis discusses activity recognition from a perspective of unobtrusiveness, where devices are worn or placed in the environment without being stigmatising or in the way. The research focuses on performing unobtrusive activity recognition when computational and sensing resources are scarce. This includes investigating unobtrusive ways to gather data, as well as adapting data modelling and classification to small, resource-constrained, devices.

The work presents different aspects of data collection and data modelling when only using unobtrusive sensing. This is achieved by considering how different sensor placements affects prediction performance and how activity models can be created when using a single sensor, or when using a number of simple binary sensors, to perform movement analysis, recognise everyday activities, and perform stress detection. The work also investigates how classification can be performed on resource-constrained devices, resulting in a novel computation-efficient classifier and an efficient hand-made classification model. The work finally sets unobtrusive activity recognition into real-life contexts where it can be used for interventions to reduce stress, sedentary behaviour and symptoms of dementia.

The results indicate that activities can be recognised unobtrusively and that classification can be performed even on resource-constrained devices. This allows for monitoring a user’s activities over extensive periods, which could be used for creating highly personal digital interventions and in-time advice that help users make positive behaviour changes. Such digital health interventions based on unobtrusive activity recognition for resource-constrained environments are important for addressing societal challenges of today, such as sedentary behaviour, stress, obesity, and chronic diseases. The final conclusion is that unobtrusive activity recognition is a cornerstone necessary for bringing many digital health interventions into a wider use.

Abstract [sv]

Denna avhandling diskuterar aktivitetsigenkänning ur ett diskret perspektiv, där enheter bärs eller placeras i miljön utan att vara stigmatiserande eller i vägen. Forskningen fokuserar på att utföra diskret aktivitetsigenkänning när beräknings- och sensor-resurser är knappa. Detta inkluderar att undersöka diskreta sätt att samla in data, samt att anpassa datamodellering och klassificering till små, resursbegränsade enheter.

Arbetet presenterar olika aspekter av datainsamling och datamodellering när man bara använder diskreta sensorer. Detta uppnås genom att överväga hur olika sensorplaceringar påverkar prediktionsprestanda och hur aktivitetsmodeller kan skapas vid användning av en enda sensor eller vid användning av ett antal enkla binära sensorer, för att utföra rörelsesanalys, känna igen vardagliga aktiviteter och utföra stressdetektering. Arbetet undersöker också hur klassificering kan utföras på resursbegränsade enheter, vilket resulterar i en ny beräkningseffektiv klassificeringsalgoritm och en effektiv handgjord klassificeringsmodell. Slutligen sätter arbetet in diskret aktivitetsigenkänning i verkliga sammanhang där det kan användas för interventioner för att minska stress, stillasittande  beteende och symptom på demens.

Resultaten visar att diskret aktivitetsigenkänning är möjligt och att klassificeringen kan utföras även på resursbegränsade enheter. Detta möjliggör övervakning av användarens aktiviteter under längre  perioder, vilket kan användas för att skapa personliga digitala interventioner och tidsanpassad rådgivning som hjälper användarna att göra positiva beteendeförändringar. Sådana digitala hälsointerventioner baserade på diskret aktivitetsigenkänning i resursbegränsade miljöer är viktiga för att ta itu med dagens samhällsutmaningar, såsom stillasittande beteende, stress, fetma och kroniska sjukdomar. En slutsats av arbetet är att diskret aktivitetsigenkänning är en hörnsten som är nödvändig för att få en större användning av digitala hälsointerventioner.

Place, publisher, year, edition, pages
Luleå: Luleå University of Technology, 2018
Series
Doctoral thesis / Luleå University of Technology 1 jan 1997 → …, ISSN 1402-1544
National Category
Computer Sciences Media and Communication Technology
Research subject
Pervasive Mobile Computing
Identifiers
urn:nbn:se:ltu:diva-71073 (URN)978-91-7790-232-4 (ISBN)978-91-7790-233-1 (ISBN)
Public defence
2018-12-11, C305, Luleå Tekniska Universitet, 97187 Luleå, Luleå, 09:00 (English)
Opponent
Supervisors
Available from: 2018-10-16 Created: 2018-10-15 Last updated: 2023-09-05Bibliographically approved
2. Decentralized Location-aware Orchestration of Containerized Microservice Applications: Enabling Distributed Intelligence at the Edge
Open this publication in new window or tab >>Decentralized Location-aware Orchestration of Containerized Microservice Applications: Enabling Distributed Intelligence at the Edge
2020 (English)Doctoral thesis, comprehensive summary (Other academic)
Abstract [en]

Services that operate on public, private, or hybrid clouds, should always be available and reachable to their end-users or clients. However, a shift in the demand for current and future services has led to new requirements on network infrastructure, service orchestration, and Quality-of-Service (QoS). Services related to, for example, online-gaming, video-streaming, smart cities, smart homes, connected cars, or other Internet-of-Things (IoT) powered use cases are data-intensive and often have real-time and locality requirements. These have pushed for a new computing paradigm, Edge computing, based on moving some intelligence from the cloud to the edge of the network to minimize latency and data transfer. This situation has set new challenges for cloud providers, telecommunications operators, and content providers. This thesis addresses two issues in this problem area that call for distinct approaches and solutions. Both issues share the common objectives of improving energy-efficiency and mitigating network congestion by minimizing data transfer to boost service performance, particularly concerning latency, a prevalent QoS metric. The first issue is related to the demand for a highly scalable orchestrator that can manage a geographically distributed infrastructure to deploy services efficiently at clouds, edges, or a combination of these. We present an orchestrator using process containers as the virtualization technology for efficient infrastructure deployment in the cloud and at the edge. The work focuses on a Proof-of-Concept design and analysis of a scalable and resilient decentralized orchestrator for containerized applications, and a scalable monitoring solution for containerized processes. The proposed orchestrator deals with the complexity of managing a geographically dispersed and heterogeneous infrastructure to efficiently deploy and manage applications that operate across different geographical locations — thus facilitating the pursuit of bringing some of the intelligence from the cloud to the edge, in a way that is transparent to the applications. The results show this orchestrator’s ability to scale to 20 000 nodes and to deploy 30 000 applications in parallel. The resource search algorithm employed and the impact of location awareness on the orchestrator’s deployment capabilities were also analyzed and deemed favorable. The second issue is related to enabling fast real-time predictions and minimizing data transfer for data-intensive scenarios by deploying machine learning models at devices to decrease the need for the processing of data by upper tiers and to decrease prediction latency. Many IoT or edge devices are typically resource-scarce, such as FPGAs, ASICs, or low-level microcontrollers. Limited devices make running well-known machine learning algorithms that are either too complex or too resource-consuming unfeasible. Consequently, we explore developing innovative supervised machine learning algorithms to efficiently run in settings demanding low power and resource consumption, and realtime responses. The classifiers proposed are computationally inexpensive, suitable for parallel processing, and have a small memory footprint. Therefore, they are a viable choice for pervasive systems with one or a combination of these limitations, as they facilitate increasing battery life and achieving reduced predictive latency. An implementation of one of the developed classifiers deployed to an off-the-shelf FPGA resulted in a predictive throughput of 57.1 million classifications per second, or one classification every 17.485 ns.

Place, publisher, year, edition, pages
Luleå University of Technology, 2020
Series
Doctoral thesis / Luleå University of Technology 1 jan 1997 → …, ISSN 1402-1544
National Category
Computer Systems Media and Communication Technology
Research subject
Pervasive Mobile Computing
Identifiers
urn:nbn:se:ltu:diva-79135 (URN)978-91-7790-617-9 (ISBN)978-91-7790-618-6 (ISBN)
Public defence
2020-09-28, A1545, LTU, Luleå, 09:00 (English)
Opponent
Supervisors
Available from: 2020-06-09 Created: 2020-06-05 Last updated: 2023-09-05Bibliographically approved

Open Access in DiVA

No full text in DiVA

Other links

Publisher's full textScopus

Authority records

Karvonen, NiklasNilsson, JoakimKleyko, DenisJimenez, Lara Lorna

Search in DiVA

By author/editor
Karvonen, NiklasNilsson, JoakimKleyko, DenisJimenez, Lara Lorna
By organisation
Embedded Internet Systems LabComputer Science
Media and Communication TechnologyOther Electrical Engineering, Electronic Engineering, Information EngineeringComputer Sciences

Search outside of DiVA

GoogleGoogle Scholar

doi
isbn
urn-nbn

Altmetric score

doi
isbn
urn-nbn
Total: 585 hits
CiteExportLink to record
Permanent link

Direct link
Cite
Citation style
  • apa
  • ieee
  • modern-language-association-8th-edition
  • vancouver
  • Other style
More styles
Language
  • de-DE
  • en-GB
  • en-US
  • fi-FI
  • nn-NO
  • nn-NB
  • sv-SE
  • Other locale
More languages
Output format
  • html
  • text
  • asciidoc
  • rtf