BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//Date iCal//NONSGML kigkonsult.se iCalcreator 2.20.4//
METHOD:PUBLISH
X-WR-CALNAME;VALUE=TEXT:����ԭ��
BEGIN:VTIMEZONE
TZID:America/New_York
BEGIN:STANDARD
DTSTART:20191103T020000
TZOFFSETFROM:-0400
TZOFFSETTO:-0500
TZNAME:EST
END:STANDARD
BEGIN:DAYLIGHT
DTSTART:20190310T020000
TZOFFSETFROM:-0500
TZOFFSETTO:-0400
TZNAME:EDT
END:DAYLIGHT
END:VTIMEZONE
BEGIN:VEVENT
UID:calendar.361756.field_event_date.0@www.wright.edu
DTSTAMP:20260220T000959Z
CREATED:20190724T134831Z
DESCRIPTION:Ph.D. Committee:  Drs. Keke Chen\, Advisor\, Krishnaprasad Thir
 unarayan\, Junjie Zhang\, and Xiaoyu Liu (Mathematics & Statistics) ABSTRA
 CTWith massive data collections and needs for building powerful predictive
  models\, data owners may choose to outsource storage and expensive machin
 e learning computations to public cloud providers. This happens due to the
  lack of in-house storage and computation resources and/or the expertise o
 f building models. Similarly\, users\, who subscribe to specialized servic
 es such as movie streaming and social networking\, voluntarily upload thei
 r data to the service providers' site for storage\, analytics\, and better
  services. The service provider may also choose to benefit from ubiquitous
  cloud computing. However\, outsourcing to the public cloud may raise priv
 acy concerns when sensitive personal or corporate data is involved. A clou
 d provider (Cloud) may mishandle sensitive data and models. Moreover\, Clo
 ud's resources\, if poorly maintained\, become vulnerable to privacy breac
 hes from external and internal adversaries. Such potential threats are out
  of the control of the data owners or general users. One way to address th
 e privacy concerns is through confidential machine learning (CML). In CML\
 , data owners protect their data with encryption or other methods before o
 utsourcing\, and Cloud learns predictive models from such protected data.E
 xisting crypto and privacy-protection methods cannot be directly applied i
 n building CML frameworks in the outsourced setting. Although theoreticall
 y sound\, a naïve adaptation of fully homomorphic encryption (FHE) and gar
 bled circuits (GC) that enable evaluation of any arbitrary function in a p
 rivacy-preserving manner is impractically expensive. Differential privacy 
 (DP)\, on the other hand\, does not exactly fit the outsourced setting as 
 data and the learned models are leaked to the Cloud. Moreover\, DP signifi
 cantly degrades model quality.  A practical CML framework must also minimi
 ze the client-side (e.g.\, data owners) cost\, moving the expensive and sc
 alable components to Cloud\, to justify the choice of outsourcing. Thus\, 
 novel solutions are needed to construct privacy-preserving learning algori
 thms that have a good balance among privacy protection\, costs\, and model
  quality. In this dissertation\, I present three confidential machine lear
 ning frameworks for the outsourcing setting: 1) PrivateGraph for unsupervi
 sed learning (e.g.\, graph spectral analysis)\, 2) SecureBoost for supervi
 sed learning (e.g.\, boosting)\, 3) DisguisedNets for deep learning (e.g.\
 , convolutional neural networks)\, respectively. The first two frameworks 
 provide semantic security and follow the decomposition-mapping-composition
  (DMC) process. The DMC process includes three critical steps:  1) Decompo
 sition of the target machine learning algorithm into its sub-components\, 
 2) Mapping of the selected sub-components to appropriate cryptographic and
  privacy primitives\, and finally\, 3) Composition of the CML protocols. I
 t is critical that one identifies the ``crypto-unfriendly' subcomponents a
 nd their alteration or replacement with ``crypto-friendly' subcomponents b
 efore the final composition of the CML frameworks. The Disguised-Nets fram
 ework\, however\, due to the intrinsically expensive nature of deep neural
  networks (DNN) and size of the training images\, relies on a perturbation
  based CML construction. By relaxing the overall security and disguising t
 he training images with cheaper transformations\, Disguised-Nets enables t
 raining confidential DNN models over the protected images very efficiently
 . I have conducted the formal cost and security analysis and performed ext
 ensive experiments for all three CML frameworks. The results have shown th
 at the costs are practical in real-world scenarios and the quality of the 
 generated models is comparable with those learned over unprotected data.
DTSTART;TZID=America/New_York:20190731T100000
DTEND;TZID=America/New_York:20190731T120000
LAST-MODIFIED:20190724T141736Z
LOCATION:304 Russ Engineering
SUMMARY:Ph.D. Dissertation Defense “Towards Data and Model Confidentiality 
 in Outsourced Machine Learning” By Sagar Sharma
URL;TYPE=URI:/events/phd-dissertation-defense-%E2%80%
 9Ctowards-data-model-confidentiality-outsourced-machine-learning%E2%80%9D
END:VEVENT
END:VCALENDAR