Von: Marc Voigt via ak discourse
Datum: Fri, 11 Nov 2022
Betreff: [ak-discourse] Einladung zum Forschungskolloquium am 15.11.2022
Liebe Kollegen und Studierende, liebe Interessenten an Veranstaltungen
am Fachgebiet Audiokommunikation,
am kommenden Dienstag, 15.11.2022, 16:00 Uhr c.t. wird Stéphane Thunus seine Arbeit zum Thema „Arborescent Dictionary Morphing Regression – Adapting the measurement to the object rather than imposing the measurement’s structure“ präsentieren. Dazu möchten wir Sie sehr herzlich einladen. Eine Kurzzusammenfassung darüber finden Sie, wie immer, am Ende dieser E-Mail.
Vor Ort gelten die Hygieneregeln der TU Berlin:
https://www.tu.berlin/themen/coronavirus/hygieneregeln/
Der Zoom-Link lautet:
https://tu-berlin.zoom.us/j/65412697375?pwd=L1dwcWxoVjdSeFZJcFFxU3I1WXdzdz09
Meeting-ID: 654 1269 7375
Kenncode: 20221025
Viele Grüße
Marc Voigt
—
Stéphane Thunus: Arborescent Dictionary Morphing Regression – Adapting the measurement to the object rather than imposing the measurement’s structure
In audio processing, it is of interest to emulate systems like electronic circuitry, natural phenomena or known but computationally prohibiting functions. The best-known approximation method is the Taylor expansion, which has, however, several limitations such as requiring the function to be known, n-times differentiable, while only yielding locally optimal results. For general-purpose fitting, which includes fitting unknown functions, more advanced methods are required such as Least Squares Regressions. Those stochastic methods, given a set of terms, find the optimal linear combination coefficients for the expansion. They, however, use all given terms indiscriminately of how poorly they might represent the desired function or system. Thus, the measurement’s structure is imposed to the measured object. This is avoided by using “forward” regressions, which chose the minimal number of terms from a given set to yield the optimal NARMAX expansion which can contain arbitrary functions and expressions. The thesis, based on the FOrLSR (forward orthogonal least-squares regression), is divided in 3 parts:
1) FOrLSR-Reformulation in Matrix-Form to flatten the time complexity from O(n²) to O(n).
2) Retriggering the FOrLSR in a tree-search to increase solution (expansion) sparsity and quality.
3) Adding a function morphing functionality allowing to adapt the regressors to the system and nesting non-linearities in the expansion.
The thesis thus proposes an efficient and precise general purpose approximation method, which yields an equation representing arbitrarily non-linear and recursive systems.
—
Voigt, Marc lädt Sie zu einem geplanten Zoom-Meeting auf der TU Berlin Zoom Instanz ein.
Thema: Forschungskolloquium
Uhrzeit: 25.Okt. 2022 16:00 Amsterdam, Berlin, Rom, Stockholm, Wien
Jede Woche am Di
25.Okt. 2022 16:00
1.Nov. 2022 16:00
8.Nov. 2022 16:00
15.Nov. 2022 16:00
22.Nov. 2022 16:00
29.Nov. 2022 16:00
6.Dez. 2022 16:00
13.Dez. 2022 16:00
20.Dez. 2022 16:00
27.Dez. 2022 16:00
3.Jan. 2023 16:00
10.Jan. 2023 16:00
17.Jan. 2023 16:00
24.Jan. 2023 16:00
31.Jan. 2023 16:00
7.Feb. 2023 16:00
14.Feb. 2023 16:00
21.Feb. 2023 16:00
28.Feb. 2023 16:00
7.März 2023 16:00
14.März 2023 16:00
21.März 2023 16:00
28.März 2023 16:00
Laden Sie die folgenden iCalendar-Dateien (.ics) herunter und importieren Sie sie in Ihr Kalendersystem.
Wöchentlich: https://tu-berlin.zoom.us/meeting/u5EpduuupzwpHdH7MbEF1pM-u3uOfA8ivd4k/ics?icsToken=98tyKu-tqjooHN2Ssx6CR_MMBoj4a-7xmGZegqd1yzLnJgYCci67I7FXHbReSNHG
Zoom-Meeting beitreten
https://tu-berlin.zoom.us/j/65412697375?pwd=L1dwcWxoVjdSeFZJcFFxU3I1WXdzdz09
Meeting-ID: 654 1269 7375
Kenncode: 20221025
—
Marc Voigt
IT-Administration
Technische Universität Berlin
Fakultät I – Geistes- und Bildungswissenschaften
Institut für Sprache und Kommunikation
Fachgebiet Audiokommunikation
Faculty I – Humanities and Educational Sciences
Institute of Speech and Communication
Audio Communication Group
Einsteinufer 17c, 10587 Berlin
GERMANY
Telefon: +49 (0)30 314-25557
Telefax: +49 (0)30 314-21143
marc.voigt@tu-berlin.de