Reverse-engineering neocortical intelligence

BranchesTask_4bThe ultimate test for understanding brain computations is to use this understanding to create a machine that can perform like the brain.

We have assembled an international team of neurophysiologists, biophysicists, theorists, statisticians, anatomists, and machine learning practitioners, in order to collect and synthesize neuroscience knowledge with unprecedented completeness and diversity into fundamentally new computer algorithms for practical use.

This highly ambitious research effort is sure to produce an incredible data set and foundational insights into the function and structure of the cortex.

The project is funded by IARPA (Intelligence Advance Research Projects Agency) through the MICrONS program (Machine Intelligence from COrtical NetworkS), and is associated with the BRAIN Initiative Initative.

We welcome applications by talented and motivated individuals to participate in this large project. Please forward this call to your friends and colleagues!

Major Project components:

  • Record optically from 105 neurons (hundreds at a time) of a behaving mouse across all layers of a 1mm3 volume of visual cortex (Tolias, Xu, Paninski)
  • Analyze distributed representations of probabilistic information (Pitkow, Bethge)
  • Measure cell-type-specific plasticity rules (Tolias, Siapas)
  • Electron microscopy to reconstruct the nanoscale wiring of this circuit (Reid)
  • Construct a detailed wiring diagram from the electron micrographs (Seung)
  • Build theories of probabilistic inference based on cortical population codes and nonlinear transformations (Pitkow, Bethge, Patel)
  • Develop learning networks from neural principles (Patel, Baraniuk, Pitkow)
  • Apply networks to computer vision problems (Patel, Baraniuk, Urtasun, Zemel)