Sander Dieleman

I will be at #ICLR2019 this week, find me if you want to talk about generative models and/or ML for audio/music . Also make sure to check out the poster and talk for MAESTRO on Tuesday at 10AM! https://magenta.tensorflow.org/maestro-wave2midi2wave … https://arxiv.org/abs/1810.12247 pic.twitter.com/roUfsNPA1v

We learn discrete high-level representations (codes) of images with auxiliary decoders. Likelihood measured in code space cannot get lost in the details and focuses on large-scale coherence! Autoregressive decoders can plausibly fill in the missing detail in pixel space. (2/2)pic.twitter.com/3Ii4uKaGXL

Likelihood is a great loss fn, it's all about the space you measure it in! Our latest work on hierarchical AR image models (w/ @JeffreyDeFauw, Karen Simonyan): https://arxiv.org/abs/1903.04933  We generated 128x128 & 256x256 samples for all ImageNet classes: https://bit.ly/2FJkvhJ  (1/2)pic.twitter.com/4SsaOlqzV6

Hello @CamdenCouncil @kingscrossN1C, the Pancras Road taxi rank regularly extends onto the Goods Way cycle lane, forcing cyclists to slalom (recorded Wednesday ~11AM). The new layout of the Pancras Road crossing has only made this more dangerous. What can be done about this?pic.twitter.com/OOL3CZomsF

I will be at #NeurIPS2018 to present our work on music generation in the raw audio domain, using a stack of WaveNet autoencoders. Poster #87 on Tuesday Dec 4th, 5PM-7PM! Paper: http://papers.nips.cc/paper/8023-the-challenge-of-realistic-music-generation-modelling-raw-audio-at-scale … Samples: https://goo.gl/A9nTZa pic.twitter.com/ANmIljicI2

Invertible neural networks are really cool! Check out this excellent blog post about a new paper where they are used to analyse inverse problems: https://hci.iwr.uni-heidelberg.de/vislearn/inverse-problems-invertible-neural-networks/ … paper: https://arxiv.org/abs/1808.04730  (1/4)pic.twitter.com/0rwkBKtwTY

Stacking WaveNet autoencoders on top of each other leads to raw audio models that can capture long-range structure in music. Check out our new paper: https://arxiv.org/abs/1806.10474  Listen to some minute-long piano music samples: https://goo.gl/A9nTZa pic.twitter.com/snWRKHbnwr

"We conclude that the common association between sequence modeling and recurrent nets should be reconsidered, and convolutional nets should be regarded as a natural starting point for sequence modeling tasks." Great to see more work in this direction! https://arxiv.org/abs/1803.01271 pic.twitter.com/Xi7SJ7FPat

Having lunch with @ugent alumni and students at #nips2017! Now at @IBMResearch, @DeepMindAI, @GoogleBrain and @OpenAIpic.twitter.com/sPOwULYk8D

#nips4creativity workshop beginning shortly! Join us downstairs at the Hyatt hotel. https://nips2017creativity.github.io  #NIPS2017pic.twitter.com/IWdWAo7gS6

I'm speaking at the MIR meetup in Berlin tomorrow https://www.meetup.com/Berlin-Music-Information-Retrieval-Meetup/events/243855597/ … and at NDSS in Stockholm on Thursday http://www.nordicdatasciencesummit.com/ pic.twitter.com/E9WLVVSL9I

PSA: #nips2017 will sell out today or tomorrow! Their predictive model does not seem to account for the feedback loop :)pic.twitter.com/SQKHBhJ0Uk

PatternNet & PatternLRP: nice work from a former colleague on interpreting neural network classification decisions. https://arxiv.org/abs/1705.05598 pic.twitter.com/K0pvw69sS3

I've been working on WaveNet autoencoders with @GoogleBrain Magenta. blog post: https://magenta.tensorflow.org/nsynth  paper: https://arxiv.org/abs/1704.01279 pic.twitter.com/ErvVihcD5f

New @kaggle competition: lung cancer detection. First prize $500k, prizes for the entire top 10. Nice! https://www.kaggle.com/c/data-science-bowl-2017 …pic.twitter.com/4hQ4jLYfQQ

More on analyzing & fixing GANs: https://openreview.net/forum?id=Hk4_qw5xe … https://arxiv.org/abs/1610.01945  -- let me know what I've missed! [2/3]pic.twitter.com/YAbOfuDGaA

Lots of interesting work on "fixing" GANs right now: https://arxiv.org/abs/1612.02780  https://arxiv.org/abs/1611.04273  https://arxiv.org/abs/1611.02163  [1/3]pic.twitter.com/nlHwCVHtE8

Harmonic networks (@deworrall92 et al.) are fully rotation equivariant convnets. Very cool! https://arxiv.org/abs/1612.04642  https://www.youtube.com/watch?v=qoWAFBYOtoU …pic.twitter.com/JvgL1NFUcU

A human rendition of one of the #WaveNet piano samples, and some detailed analysis from Magenta: https://magenta.tensorflow.org/2016/09/23/learning-music-from-learned-music/ …pic.twitter.com/hGGOYbkWgn

Presenting our poster on cyclic symmetry in CNNs this afternoon at #ICML2016! (With @JeffreyDeFauw and @koraykv)pic.twitter.com/3rTq5OdoGp