[visionlist] NeurIPS20 workshop: BeyondBackpropagation - Novel Ideas for Training Neural Architectures

Viorica Patraucean vpatrauc at gmail.com
Thu Sep 17 15:25:40 -04 2020

Call for papers (apologies for crossposting)
NeurIPS workshop: BeyondBackpropagation - Novel Ideas for Training Neural
Web: https://beyondbackprop.github.io/

Important dates
Submission deadline: 9th of October, 2020
Final decisions: 30th of October, 2020

Motivation and description
Many have questioned the biological plausibility of backpropagation as a
learning mechanism since its discovery. In addition, backpropagation limits
the amount of computation parallelism, yielding high latency. In this
workshop, we promote discussions around more efficient training schemes by
bringing together researchers from machine learning, neuroscience, and
hardware engineering.

We invite high-quality 4-page papers on the following topics:
- Biologically plausible training of deep learning models
- Large scale learning; distributed and parallel training
- Efficient training or inference
- Alternatives to backprop such as target-prop or alternating minimization
- Statistical analysis, convergence, and generalization bounds
- Neural networks with reconfigurable architectures
- Energy efficient deep learning
- Scalable training at the system level (hardware, compilation)
- Applications to new domains
Accepted papers will be presented during the poster session. Top
submissions will be selected for short oral presentations. The reviews will
be double-blind.
Link to submissions system:

Confirmed speakers
Danielle Bassett (U of Pennsylvania)
Yoshua Bengio (MILA)
David Duvenaud (U of Toronto)
Karl Fristen (UCL)
Cristina Savin (NYU)
Olivier Teytaud (Facebook)
Bastiaan Veeling (U of Amsterdam)

Anna Choromanska (NYU)
Marco Gori (U of Siena)
Yanping Huang (Google)
Sindy Lowe (U of Amsterdam)
Mateusz Malinowski (DeepMind)
Viorica Patraucean (DeepMind)
Grzegorz Swirszcz (DeepMind)

More info: https://beyondbackprop.github.io/
Contact: viorica at google.com
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://visionscience.com/pipermail/visionlist_visionscience.com/attachments/20200917/d76c7bef/attachment.html>

More information about the visionlist mailing list