Machine Learning meets Biology: Algorithms and Cortical Mechanisms – May 4 2017
A driving goal of the Blue Brain Project is to reconstruct and simulate the anatomy and physiology of rodent neocortical microcircuitry constrained by available experimental data. Leveraging this model, one aim is to deepen the understanding of how learning and computation arise in biological systems at the level of neural circuits.
In light of the impressive and rapidly expanding list of achievements in the field of machine learning, there is significant untapped potential to improve the use of experimental data to constrain simulation neuroscience, and conversely contribute new insights back to the machine learning community through the mutual exchange of ideas.
On Thursday May 4, Blue Brain hosted an interdisciplinary workshop focused on enhancing the dialogue between the machine learning and neuroscience communities, and providing an opportunity for the cross-pollination of ideas.
The event featured a diverse line-up of speakers and included robust round-table style discussions on algorithms, spiking networks, and neural architectures. Henry Markram gave the opening lecture and was followed by speakers from MIT, HUJI, University of Bern, ETH Zurich, EPFL, Demiurge Technologies (CH) and the University of Geneva. .
Session One – Brains to Machines
Henry Markram (EPFL) – ‘Principles of neocortical microcuitry’
Adam Marblestone (MIT) – ‘Towards an integration of deep learning and neuroscience’ PDF
Session Two – Algorithms
Bragi Lovetrue (Demiurge Technologies AG, Switzerland) – ‘Grass-roots collaboration on discovering the algorithmic principles of cross-level and cross-area thalamocortical interactions’
Joachim Buhmann (ETH Zurich, Switzerland) – ‘Data science for medicine: we need resilient algorithms’ PDF
Session Three – Spiking Networks
Mihai Petrovici (University of Bern, Switzerland) – ‘Fast interference with spiking networks’ PDF
Aditya Gilra and Wulfram Gerstner (EPFL) – ‘Astable local learning scheme for recurrent spiking neural networks’
Session Four – Neural Architectures
Stephane Marchand-Maillet (University of Geneva, Switzerland) – ‘Manifold learning for complex data visualization: benefits to and from neural architectures’ PDF
Idan Segev (HUJI) – ‘The Neuron as multisite plastic unit’