Aus SDQ-Institutsseminar
Version vom 14. November 2022, 11:15 Uhr von Federico Matteucci (Diskussion | Beiträge) (Die Seite wurde neu angelegt: „{{Termin |datum=2022-11-25T11:30:00.000Z |raum=Raum 348 (Gebäude 50.34) |online= }}“)
(Unterschied) ← Nächstältere Version | Aktuelle Version (Unterschied) | Nächstjüngere Version → (Unterschied)
Termin (Alle Termine)
Datum Freitag, 25. November 2022
Uhrzeit 11:30 – 12:20 Uhr (Dauer: 50 min)
Ort Raum 348 (Gebäude 50.34)
Vorheriger Termin Fr 11. November 2022
Nächster Termin Fr 2. Dezember 2022

Termin in Kalender importieren: iCal (Download)


Vortragende(r) Mingzhe Tao
Titel Meta-Learning for Encoder Selection
Vortragstyp Bachelorarbeit
Betreuer(in) Federico Matteucci
Vortragsmodus in Präsenz
Kurzfassung In the process of machine learning, the data to be analyzed is often not only numerical but also categorical data. Therefore, encoders are developed to convert categorical data into the numerical world. However, different encoders may have other impacts on the performance of the machine learning process. To this end, this thesis is dedicated to understanding the best encoder selection using meta-learning approaches. Meta-learning, also known as learning how to learn, serves as the primary tool for this study. First, by using the concept of meta-learning, we find meta-features that represent the characteristics of these data sets. After that, an iterative machine learning process is performed to find the relationship between these meta-features and the best encoder selection.

In the experiment, we analyzed 50 datasets, those collected from OpenML. We collected their meta-features and performance with different encoders. After that, the decision tree and random forest are chosen as the meta-models to perform meta-learning and find the relationship between meta-features and the performance of the encoder or the best encoder. The output of these steps will be a ruleset that describes the relationship in an interpretable way and can also be generalized to new datasets.

Vortragende(r) Georg Gntuni
Titel Streaming Nyström MMD Change Detection
Vortragstyp Proposal
Betreuer(in) Florian Kalinke
Vortragsmodus in Präsenz
Kurzfassung Data streams are omnipresent. Think of sensor data, bank transactions, or stock movements. We assume that such data is generated according to an underlying distribution, which may change at so-called change points. These points signal events of interest; hence one wants to detect them.

A principled approach for finding such change points is to use maximum mean discrepancy (MMD) for a statistical hypothesis test, with the null hypothesis that the distribution does not change. However, the quadratic runtime of MMD prohibits its application in the streaming setting. Approximations for that setting exist but these suffer from high variance.

In the static setting, the so-called Nyström method allows to reduce the quadratic runtime of MMD with only a slight increase in variance. We propose an algorithm to employ Nyström estimators for MMD in the streaming setting and compare it to existing approximations.

Neuen Vortrag erstellen