Software Engineer @ Google

Zurich, Switzerland
Joined April 2008
The GAN scene is getting ridiculous.
Enriched Multimodal Representations of Music Performances: Online Access and Visualization: computer.org/multimedia-maga…
2
2
Latest: A3C is cool but threads are slow, I'm just gonna batch all the updates in a big tensor. arxiv.org/pdf/1705.04862.pdf. 13 days ago.
Replying to @lemonzi @overleaf
WYSIWYG, bi***es!
1
Anyone in NYC wants to host me and Marina for a few days in May? 😇 🗽 🗽
Quim Llimona retweeted
The TF wrapper we use internally at DeepMind has been open sourced. Lasagne users might like this one, it shares a lot of design principles. xitter.cf/DeepMindAI/status/8503…
Excited to release #Sonnet - a library for constructing complex Neural Network models in TensorFlow. Get started: github.com/deepmind/sonnet
1
75
2
154
If your model doesn't work after debugging it, what would you do first?
50%Tune hyperparameters
0%Replace some parts of it
50%Debug further (how?)
2 votes • Final results
2
My model is implemented, nothing crashes, summary probes look good, but it doesn't even learn the toy task. It must have ADHD or something.
Seen on Facebook -- Stopping GAN Violence: Generative Unadversarial Networks arxiv.org/pdf/1703.02528.pdf
Quim Llimona retweeted
Google has just published a huge database of everyday sounds for machine learning research.google.com/audioset…
2
264
19
394
Today, I left my #Tensorflow model running all night and it exploded to NaN's within the first 30 min. #FML
Here go another 750 fresh, untested lines of code for my #tensorflow RL model... this can't end well #whowantstodebugit #notme
1
Quim Llimona retweeted
Journal paper and new dataset release related to MTG's Repovizz technology mtg.upf.edu/node/3694
3
3
I'm guessing it's just collaborative filtering... It would be cool if they had image recognition / description on-the-spot
Replying to @github
Those inverted colors are terrible for legibility