Anticipation and the Anticipatory Music Transformer

Speaker: John Thickstun , Stanford

Date: Monday, October 30, 2023

Time: 2:00 PM to 3:30 PM Note: all times are in the Eastern Time Zone

Public: Yes

Location: Conference Room 370 (32-370)

Event Type: Seminar

Room Description:

Host:

Contact: Jacob Andreas, jda@csail.mit.edu

Relevant URL:

Speaker URL: https://johnthickstun.com/

Speaker Photo:
None

Reminders to: seminars@csail.mit.edu

Reminder Subject: TALK: Anticipation and the Anticipatory Music Transformer

Abstract: Imagine you are given a melody, and asked to compose a harmonizing accompaniment. The melody is an example of a temporal point process: a sequence of (musical) events that arrive stochastically at points in time. The accompaniment is also a temporal point process, one that is tightly correlated with—but asynchronous from—the melody. Motivated by this example, we are interested in generative models of a temporal point process (the event process) that can be conditioned on realizations of a second, correlated point process (the control process). We achieve this using anticipation: a new method of controllable generation that interleaves sequences of events and controls, such that controls appear following stopping times in the event sequence. We apply anticipation to construct anticipatory infilling models of music; these models unlock new capabilities for controllable music generation, without sacrificing the performance of unconditional generation.

Research Areas:

Impact Areas:

This event is not part of a series.

Created by Jacob Andreas Email at Thursday, October 19, 2023 at 11:05 AM.