[CCoE Notice] Cullen College Dissertation Announcement - Pavana Prakash

[Hutchinson, Inez A]

[Dissertation Defense Announcement at the Cullen College of Engineering]
Towards Mobile-Device Friendly Federated  Learning: Efficiency, Privacy and Personalization

Pavana Prakash

August 1, 2023;  08:30 AM  - 10:30 AM (CST)
Location:  Zoom Meeting
Zoom: https://urldefense.com/v3/__https://uh-edu-cougarnet.zoom.us/j/9167221115__;!!LkSTlj0I!C00jwt0flQNP3AfZmGeBkr3YJ6aNr6cVNUiwgLI4xWprIqyJMHWx5LWi3-SRcMwzxAziVEN3dDafSnDL_uCIQd-GxtY$ <https://urldefense.com/v3/__https:/uh-edu-cougarnet.zoom.us/j/9167221115__;!!LkSTlj0I!AYHlDOrfeq1r2lXCGKYp75YmD6rrKcmFd29Xzwtyxxwme4lK4dR4OZ0MtER5z6igMyUIrOakxS8f7g9HNYf_TA$>

Meeting ID: 916 722 1115


Committee Chair:

Miao Pan, Ph.D.

Committee Members:
Zhu Han, Ph.D. |  Hien Van Nguyen, Ph.D. |  Xin Fu, Ph.D. |  Yuanxiong Guo, Ph.D.

Abstract

The integration of deep learning capabilities and advanced mobile processors has significantly boosted the popularity of cutting-edge mobile devices such as smartphones. Moreover, the widespread use of artificial intelligence and mobile internet has led to rapid growth in innovative applications such as smart healthcare, next word prediction, augmented/virtual reality and so on, targeted at mobile devices. However, deep learning applications generate massive volumes of data, resulting in substantial burden on communication links and compute, as well as immense privacy implications when uploading and storing such data on servers. Edge Computing, a promising technology, can partly address these challenges by reducing latency through enabling computation near the data source. Furthermore, federated learning (FL), a fast-growing distributed learning paradigm, can aid in instituting such pervasive applications over mobile devices at the edge. Despite its potential benefits, FL involving large models faces challenges, including computation and communication expenses, system and data heterogeneity, and privacy concerns.

The objectives of this dissertation are hence to address these key challenges by developing efficient mobile device-friendly FL designs through theoretical, simulation, and experimental studies. The dissertation proposes equipping FL with the necessary tools from the perception of efficiency, privacy, and personalization. To this end, model compression and time/energy-conserving techniques are integrated with FL to enhance efficiency, and privacy requirements are met through novel compressor incorporation and its additional properties. Likewise, adaptive training techniques are developed to balance model generalization and personalization. Overall, this dissertation contributes towards enabling low-end mobile devices to effectively participate in the FL process.
[Engineered For What's Next]

On Mon, Jul 24, 2023 at 11:53 AM Hutchinson, Inez A <iajackso at central.uh.edu<mailto:iajackso at central.uh.edu>> wrote:
Hi there,

Do you mind formatting the announcement like the attached file.  To use this layout, please follow these steps:


  *   Open the attached document in your preferred internet browser
  *   Select all and copy
  *   Open a new email message and paste the content directly into the email body
  *   update your text in the email window

Once these steps are complete, please resend to me for distribution. Let me know if you have any issues with the file attached.



Best Regards,

Inez Hutchinson
Executive Director of Communications
Cullen College of Engineering
Engineering Building 2, Suite E311
University of Houston
Houston, TX 77204-4007
(713) 743-7593<tel:(713)%20743-7593> – iajackso at central.uh.edu<mailto:iajackso at central.uh.edu>



From: Pavana Prakash <pavanaprakash88 at gmail.com<mailto:pavanaprakash88 at gmail.com>>
Date: Monday, July 24, 2023 at 11:51 AM
To: "Hutchinson, Inez A" <iajackso at Central.UH.EDU<mailto:iajackso at Central.UH.EDU>>, "Hutchinson, Inez A" <iajackso at Central.UH.EDU<mailto:iajackso at Central.UH.EDU>>
Cc: "King, Kelly N" <knking at Central.UH.EDU<mailto:knking at Central.UH.EDU>>
Subject: Dissertation Announcement - Pavana Prakash, Electrical Engineering

Dear Inez,

Please find my dissertation announcement attached with this email.

My details are as follows,
Name: Pavana Prakash
PSID: 1679936
Advisor: Dr. Miao Pan

Thank you very much for your time and consideration.

Also, please find the email message of the same below,

Towards Mobile-Device Friendly Federated  Learning: Efficiency, Privacy and Personalization
Pavana Prakash

August 1, 2023; 08:30 AM  - 10:30 AM (CST)
Location: Zoom Meeting
Zoom: https://urldefense.com/v3/__https://uh-edu-cougarnet.zoom.us/j/9167221115__;!!LkSTlj0I!C00jwt0flQNP3AfZmGeBkr3YJ6aNr6cVNUiwgLI4xWprIqyJMHWx5LWi3-SRcMwzxAziVEN3dDafSnDL_uCIQd-GxtY$ <https://urldefense.com/v3/__https:/uh-edu-cougarnet.zoom.us/j/9167221115__;!!LkSTlj0I!AYHlDOrfeq1r2lXCGKYp75YmD6rrKcmFd29Xzwtyxxwme4lK4dR4OZ0MtER5z6igMyUIrOakxS8f7g9HNYf_TA$>

Committee Chair:
Miao Pan, Ph.D.

Committee Members:
Zhu Han, Ph.D. |  Hien Van Nguyen, Ph.D. |  Xin Fu, Ph.D. | Yuanxiong Guo, Ph.D.

Abstract

The integration of deep learning capabilities and advanced mobile processors has significantly boosted the popularity of cutting-edge mobile devices such as smartphones. Moreover, the widespread use of artificial intelligence and mobile internet has led to rapid growth in innovative applications such as smart healthcare, next word prediction, augmented/virtual reality and so on, targeted at mobile devices. However, deep learning applications generate massive volumes of data, resulting in substantial burden on communication links and compute, as well as immense privacy implications when uploading and storing such data on servers. Edge Computing, a promising technology, can partly address these challenges by reducing latency through enabling computation near the data source. Furthermore, federated learning (FL), a fast-growing distributed learning paradigm, can aid in instituting such pervasive applications over mobile devices at the edge. Despite its potential benefits, FL involving large models faces challenges, including compute and communication expenses, system and data heterogeneity, and privacy concerns.

The objectives of this dissertation are hence to address these key challenges by developing efficient mobile device-friendly FL designs through theoretical, simulation, and experimental studies. The dissertation proposes equipping FL with the necessary tools from the perception of efficiency, privacy, and personalization. To this end, model compression and time/energy-conserving techniques are integrated with FL to enhance efficiency, and privacy requirements are met through novel compressor incorporation and its additional properties. Likewise, adaptive training techniques are developed to balance model generalization and personalization. Overall, this dissertation contributes towards enabling low-end mobile devices to effectively participate in the FL process.
--
Regards,
Pavana Prakash


--
Regards,
Pavana Prakash
-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://Bug.EGR.UH.EDU/pipermail/engi-dist/attachments/20230724/c1b35648/attachment-0001.html