Sander Dieleman
As of October 2026, Sander Dieleman is a research scientist at Google DeepMind whose work spans generative audio, music, images, video, and language. He studied deep learning for musical audio at Ghent University and explored convolutional networks for Spotify recommendations from audio. He helped develop Lasagne, coauthored WaveNet, and contributed to the MAESTRO piano dataset. He later coauthored an autoregressive model for long sequences and papers on diffusion for language, then wrote about generative models that work in latent spaces.
Timeline
-
2014–2015
Music recommendation and Lasagne
In 2014, he interns at Spotify and develops convolutional neural networks for music recommendations from audio. In 2015, he contributes to Lasagne, a neural network library built on Theano.
-
2016
A doctorate in music learning
At Ghent University, he studies hierarchical representations of musical audio for classification and recommendation, and defends his PhD thesis in January 2016.
-
2016
WaveNet and raw audio
In 2016, he coauthors WaveNet, a deep generative model that generates raw audio, speech, and piano music.
-
2018
The MAESTRO piano dataset
In October 2018, he and his collaborators introduce the MAESTRO dataset, containing over 172 hours of aligned piano recordings, and present the Wave2Midi2Wave process for transcription, composition, and synthesis.
-
2022
Long sequences and language diffusion
In July 2022, he coauthors Perceiver AR, an autoregressive model that handles long sequences and generates piano music. In November 2022, he coauthors two papers on continuous diffusion for language, SED and CDCD.
-
April 2025
Generative modelling in latent space
In April 2025, he publishes a post about the compact latent representations used by generative models.
-
August 2026
Continuous diffusion for language
In August 2026, he publishes an overview of recent research on continuous diffusion models for language.
-
As of October 2026
Google DeepMind research scientist
He is a Google DeepMind research scientist whose work includes generative models for music, images, and video.
Sources and credits
Photo: We have no published photo of Sander Dieleman yet. Send us a free-licensed photo.
- Recommending music on Spotify with deep learning – Sander Dieleman
- New Lasagne feature: arbitrary expressions as layer parameters – Sander Dieleman
- About me – Sander Dieleman
- Research – Sander Dieleman
- WaveNet: A generative model for raw audio — Google DeepMind
- The MAESTRO Dataset and Wave2Midi2Wave
- Perceiver AR: general-purpose, long-context autoregressive generation — Google DeepMind
- Self-conditioned Embedding Diffusion for Text Generation
- Continuous diffusion for categorical data
- Generative modelling in latent space – Sander Dieleman
- Continuous diffusion language models – Sander Dieleman
- Sander Dieleman: Bio, Work & Ideas | AI Engineer