Advanced search options

Advanced Search Options 🞨

Browse by author name (“Author name starts with…”).

Find ETDs with:

in
/  
in
/  
in
/  
in

Written in Published in Earliest date Latest date

Sorted by

Results per page:

Sorted by: relevance · author · university · dateNew search

You searched for subject:(On device Learning). Showing records 1 – 3 of 3 total matches.

Search Limiters

Last 2 Years | English Only

No search limiters apply to these results.

▼ Search Limiters


Delft University of Technology

1. Hofman, Stefan (author). Federated Learning for Mobile and Embedded Systems.

Degree: 2020, Delft University of Technology

An increase in the performance of mobile devices has started a revolution in deploying artificial intelligence (AI) algorithms on mobile and embedded systems. In addition, fueled by the need for privacy-aware insights into data, we see a strong push towards federated machine learning, where data is stored locally and not shared with a central server. By allowing data to stay on client devices and do training locally, we work towards a more privacy-friendly future. Furthermore, utilizing federated machine learning enables machine learning in data-constrained environments where bandwidth is not sufficient to upload the entire dataset. In this thesis, we look at the recent trend into less complex machine learning models. These models optimize resource usage while reducing accuracy loss. We investigate how these simpler models hold up within a federated setting. We also look into the developments of AI frameworks and their capabilities for mobile platforms. Based on these findings, we propose that model-hyper-parameter optimization is possible to maximize accuracy for smaller networks during federated learning. We show that it is possible to reduce the accuracy loss from 15% to only 0.04%. We then demonstrate what a mobile implementation looks like and the performance we see from an iPhone X. We show that an iPhone implementation takes less than 2x the amount of a regular laptop implementation. Finally, we demonstrate that we can reduce the model-size by up to 7x using modern weight quantization methods. Advisors/Committee Members: Al-Ars, Z. (mentor), van Leuken, T.G.R.M. (graduation committee), Hoozemans, J.J. (graduation committee), Delft University of Technology (degree granting institution).

Subjects/Keywords: Federater learning; Artificial Intelligence; MobileNetV2; Embedded Systems; CoreML; On-device Learning

Record DetailsSimilar RecordsGoogle PlusoneFacebookTwitterCiteULikeMendeleyreddit

APA · Chicago · MLA · Vancouver · CSE | Export to Zotero / EndNote / Reference Manager

APA (6th Edition):

Hofman, S. (. (2020). Federated Learning for Mobile and Embedded Systems. (Masters Thesis). Delft University of Technology. Retrieved from http://resolver.tudelft.nl/uuid:505e7f5b-a78d-444f-a92e-a88bcc7c08d0

Chicago Manual of Style (16th Edition):

Hofman, Stefan (author). “Federated Learning for Mobile and Embedded Systems.” 2020. Masters Thesis, Delft University of Technology. Accessed January 19, 2021. http://resolver.tudelft.nl/uuid:505e7f5b-a78d-444f-a92e-a88bcc7c08d0.

MLA Handbook (7th Edition):

Hofman, Stefan (author). “Federated Learning for Mobile and Embedded Systems.” 2020. Web. 19 Jan 2021.

Vancouver:

Hofman S(. Federated Learning for Mobile and Embedded Systems. [Internet] [Masters thesis]. Delft University of Technology; 2020. [cited 2021 Jan 19]. Available from: http://resolver.tudelft.nl/uuid:505e7f5b-a78d-444f-a92e-a88bcc7c08d0.

Council of Science Editors:

Hofman S(. Federated Learning for Mobile and Embedded Systems. [Masters Thesis]. Delft University of Technology; 2020. Available from: http://resolver.tudelft.nl/uuid:505e7f5b-a78d-444f-a92e-a88bcc7c08d0


Rochester Institute of Technology

2. Soures, Nicholas M. Deep Liquid State Machines with Neural Plasticity and On-Device Learning.

Degree: MS, Computer Engineering, 2017, Rochester Institute of Technology

The Liquid State Machine (LSM) is a recurrent spiking neural network designed for efficient processing of spatio-temporal streams of information. LSMs have several inbuilt features such as robustness, fast training and inference speed, generalizability, continual learning (no catastrophic forgetting), and energy efficiency. These features make LSM’s an ideal network for deploying intelligence on-device. In general, single LSMs are unable to solve complex real-world tasks. Recent literature has shown emergence of hierarchical architectures to support temporal information processing over different time scales. However, these approaches do not typically investigate the optimum topology for communication between layers in the hierarchical network, or assume prior knowledge about the target problem and are not generalizable. In this thesis, a deep Liquid State Machine (deep-LSM) network architecture is proposed. The deep-LSM uses staggered reservoirs to process temporal information on multiple timescales. A key feature of this network is that neural plasticity and attention are embedded in the topology to bolster its performance for complex spatio-temporal tasks. An advantage of the deep-LSM is that it exploits the random projection native to the LSM as well as local plasticity mechanisms to optimize the data transfer between sequential layers. Both random projections and local plasticity mechanisms are ideal for on-device learning due to their low computational complexity and the absence of backpropagating error. The deep-LSM is deployed on a custom learning architecture with memristors to study the feasibility of on-device learning. The performance of the deep-LSM is demonstrated on speech recognition and seizure detection applications. Advisors/Committee Members: Dhireesha Kudithipudi.

Subjects/Keywords: Deep networks; Liquid state machine; Neural plasticity; Neuromemristive; On-device learning

Record DetailsSimilar RecordsGoogle PlusoneFacebookTwitterCiteULikeMendeleyreddit

APA · Chicago · MLA · Vancouver · CSE | Export to Zotero / EndNote / Reference Manager

APA (6th Edition):

Soures, N. M. (2017). Deep Liquid State Machines with Neural Plasticity and On-Device Learning. (Masters Thesis). Rochester Institute of Technology. Retrieved from https://scholarworks.rit.edu/theses/9687

Chicago Manual of Style (16th Edition):

Soures, Nicholas M. “Deep Liquid State Machines with Neural Plasticity and On-Device Learning.” 2017. Masters Thesis, Rochester Institute of Technology. Accessed January 19, 2021. https://scholarworks.rit.edu/theses/9687.

MLA Handbook (7th Edition):

Soures, Nicholas M. “Deep Liquid State Machines with Neural Plasticity and On-Device Learning.” 2017. Web. 19 Jan 2021.

Vancouver:

Soures NM. Deep Liquid State Machines with Neural Plasticity and On-Device Learning. [Internet] [Masters thesis]. Rochester Institute of Technology; 2017. [cited 2021 Jan 19]. Available from: https://scholarworks.rit.edu/theses/9687.

Council of Science Editors:

Soures NM. Deep Liquid State Machines with Neural Plasticity and On-Device Learning. [Masters Thesis]. Rochester Institute of Technology; 2017. Available from: https://scholarworks.rit.edu/theses/9687

3. Tay, Noel Nuo Wi. Human-centric Semantic Reasoning and Optimization for Smart Home : スマートホームのための人間中心セマンティック推論と最適化.

Degree: 博士(工学), 2017, Tokyo Metropolitan University / 首都大学東京

首都大学東京, 2017-03-25, 博士(工学)

Subjects/Keywords: Smart home consists of various kinds of Internet of Tings (IoT) devices connected to the private house that cooperatively provide inhabitants (users) with proactive services related to comfort; security and safety. Examples of services include 1) manipulation of lighting and temperature based on time and context; 2) reminder service of user’s schedules by using the nearest output device; and 3) device organization to realize surveillance system. However; current smart homes are developed mostly from the viewpoint of technical capabilities; where users have to decide how the connected devices are going to serve them. They may have to setup the devices based on the available functionalities and specifications of the devices; and also have to alter their living styles according to the role of each device. Besides; most devices can only provide simple services independently. Œus; cooperation among the devices is important. On the other hand; human-centric approach; which centered on humans’ need to enhance their living experience; is an important technological paradigm where services are provided anywhere and anytime based on situation. Smart home abiding this approach should cooperatively maximize fulfillment of quality of life (QOL) for individual users subject to personal constraints. In this respect; the devices are bound to enable communication of information; and their operations are coordinated to deliver services cooperatively via a sequence of device actions called a plan. Due to personalization and automation; a number of problems have to be solved. First; a means of automatic binding between loosely coupled devices depending on services delivered have to be devised; as manual setup is impractical. Secondly; coordination of devices needs to generate complex plans; without requiring manual specification of sub-plans. Besides; issue of over-constrained goals during service provisions that arises from flawed or contradicting specification from multiple users should be considered. Apart from that; low training data in general environment setting for individual identification should be addressed. The aim of this research is to establish an integrated system for the human-centric smart home (HcSH) that provides personalized service through loosely coupled devices automatically. This research modularizes the overall system into three modules; which are human identification (HIM); automated planner (APM); and semantic reasoner (SRM). HIM helps select the appropriate QOL; SRM binds the devices by associating them with planning components; which are then used by APM to generate plans for device coordination to maximize QOL fulfillment. Chapter 1 gives the introduction and design motivation. Chapter 2 presents the related works and literature reviews; as well as justifications relevant to this thesis. Chapter 3 deals with HIM; which is realized via face identification. For face identification; problems faced are heavy computational load and insufficient learning data. The solution is to use transfer learning to handle data issue while being able to build generalized face model. For face model refinement; active learning is implemented. Experimental results show the method is competitive in terms of accuracy and computational cost compared to current state of the art. Chapter 4 presents APM; where planning via solving Constraint Satisfaction Problem (CSP) is laid out. CSP in planning is declarative without requiring prior specification of sub-plans; and can handle variables of larger domains. Due to the high possibility of having over-constrained QOL as in practical cases; CSP planner cannot fulfill all of them. An example is a contradicting TV channel request from 2 persons. Optimization through weighted CSP is therefore used to maximize QOL fulfillment. Experiments on weighted CSP shows that the method is capable of performing optimization while generating complex plans. Chapter 5 is on SRM; where knowledge representation is constructed by Web Ontology Language (OWL) description logic. It models knowledge on home and building layout and device functionalities. OWL is used because it is decidable and that it is endorsed by World Wide Web Consortium (W3C). We deal with case studies based on further inference on building state as an important example to discuss the applicability of the proposed method; and demonstrate the use of building ontology. This is followed by automated device binding and the method to generate basic planning components of rules in automated planning. Finally; an extension to robot complex planning is provided to demonstrate how it can be easily extended. Chapter 6 demonstrates the applicability of the HcSH; which integrates all three modules through its implementation in a prototype smart home with 5 rooms; which houses 2 persons. Various tests are performed to show the generated plans are near optimal without redundancy. Œe system is also shown to be scalable given increasing amount of devices. Case studies show that the system can perform well even under short time threshold. Finally; chapter 7 summarizes the thesis. Future vision of the work is also laid out; which is to implement it as a community-centric system.

Record DetailsSimilar RecordsGoogle PlusoneFacebookTwitterCiteULikeMendeleyreddit

APA · Chicago · MLA · Vancouver · CSE | Export to Zotero / EndNote / Reference Manager

APA (6th Edition):

Tay, N. N. W. (2017). Human-centric Semantic Reasoning and Optimization for Smart Home : スマートホームのための人間中心セマンティック推論と最適化. (Thesis). Tokyo Metropolitan University / 首都大学東京. Retrieved from http://hdl.handle.net/10748/00009960

Note: this citation may be lacking information needed for this citation format:
Not specified: Masters Thesis or Doctoral Dissertation

Chicago Manual of Style (16th Edition):

Tay, Noel Nuo Wi. “Human-centric Semantic Reasoning and Optimization for Smart Home : スマートホームのための人間中心セマンティック推論と最適化.” 2017. Thesis, Tokyo Metropolitan University / 首都大学東京. Accessed January 19, 2021. http://hdl.handle.net/10748/00009960.

Note: this citation may be lacking information needed for this citation format:
Not specified: Masters Thesis or Doctoral Dissertation

MLA Handbook (7th Edition):

Tay, Noel Nuo Wi. “Human-centric Semantic Reasoning and Optimization for Smart Home : スマートホームのための人間中心セマンティック推論と最適化.” 2017. Web. 19 Jan 2021.

Vancouver:

Tay NNW. Human-centric Semantic Reasoning and Optimization for Smart Home : スマートホームのための人間中心セマンティック推論と最適化. [Internet] [Thesis]. Tokyo Metropolitan University / 首都大学東京; 2017. [cited 2021 Jan 19]. Available from: http://hdl.handle.net/10748/00009960.

Note: this citation may be lacking information needed for this citation format:
Not specified: Masters Thesis or Doctoral Dissertation

Council of Science Editors:

Tay NNW. Human-centric Semantic Reasoning and Optimization for Smart Home : スマートホームのための人間中心セマンティック推論と最適化. [Thesis]. Tokyo Metropolitan University / 首都大学東京; 2017. Available from: http://hdl.handle.net/10748/00009960

Note: this citation may be lacking information needed for this citation format:
Not specified: Masters Thesis or Doctoral Dissertation

.