[Mittagsseminar TI] Mittagsseminar 01.04.

[Wolfgang Mulzer <mulzer@inf.fu-berlin.de>]

Dear TIs,

this will be today's Mittagsseminar. 

Anton is sick and I have to stay home with him, so the defense will be online. 

The talk is in English. 

Cheers 

Wolfgang 

-------- Ursprüngliche Nachricht --------

Von: Oskar Besler <besleo00@zedat.fu-berlin.de>

Datum: 01.04.25 10:08 (GMT+01:00)

An: Oskar Besler <besleo00@zedat.fu-berlin.de>

Cc: i-profs@inf.fu-berlin.de, i-wimis@inf.fu-berlin.de, i-studi@inf.fu-berlin.de, maria.koekenhoff@fu-berlin.de

Betreff: Re: [i-prof] Invitation to the defense of my master's thesis

Dear all,

due to illness, we need to move the thesis defense to an online setting. I
apologize for the inconvenience.

You’ll find the access link below.

Best regards,
Oskar Besler

Link:
https://fu-berlin.webex.com/fu-berlin-en/j.php?MTID=m0f220746197780c52f4f86a0d0b4bf10

Meeting number (access code): 2791 917 0049

Meeting password: Xv5CQZbnV84 (98527926 from phones and video systems)

> Dear all,
>
> I hereby invite you to the defense of my master's thesis with the title
> "How to Find a Point Within the Union of Squares Using a Differentially
> Private Algorithm".
>
> The defense will take place on Tuesday, 01.04.2025 at 12:00am in
> Takustraße 9, seminar room 055.
>
> The defense will be held in English.
>
> First examiner & Supervisor: Prof. Dr. Wolfgang Mulzer
> Second examiner: Prof. Dr. László Kozma
>
> Best regards,
> Oskar Besler
>
> ________________________
>
> Abstract:
>
> Background: In the age of social media and big data, privacy concerns are
> a growing
> issue as individuals worry about the misuse of their sensitive
> information. Differential
> privacy is considered the gold standard for privacy protection in data
> analysis, as it
> provides a mathematical guarantee that an individual will not be
> significantly affected
> when their sensitive data is used for data analysis. The task of
> extracting useful information about a group of people without revealing
> meaningful details about individuals
> can be described by the private interior point problem, which involves
> finding a point
> inside a convex hull using a differentially private algorithm.
>
> Goals: Since data can be represented by more than just points, this thesis
> aims to
> extend the definition of the private interior point problem by redefining
> it as finding
> a point inside the union of given squares using a differentially private
> algorithm.
>
> Methods: To achieve this, we will develop and analyze two differential
> private algorithms that address two variations of this problem. The first
> allows only axis-
> aligned squares as input (ASPIP), while the second imposes no restrictions
> on the
> input squares (SPIP).
>
> Results: The experiments show that both algorithms perform well in solving
> their
> respective variation of this problem. Furthermore, the analysis reveals
> that a good
> bound on the input size can be established if a certain overlap of input
> squares is
> known beforehand, ensuring that the constraints of differential privacy
> are still met.
>
> Conclusion: Both algorithms provide a good initial step towards better
> understanding
> this version of the private interior point problem by offering useful
> bounds on input
> size, as well as acceptable execution time and space usage. However, there
> is still
> future work that needs to be addressed, such as generalizing this problem
> to higher
> dimensions.

_______________________________________________
Automatischer Mailverteiler an Gruppe 'ml-i-prof-mi'.
Hinweise dazu siehe Hilfeseite:
https://www.mi.fu-berlin.de/w/Tec/AnkuendigungsVerteiler