I am pleased to announce a Proposers' Day conference for IARPA's Machine Intelligence from Cortical Networks (MICrONS) program. The conference will be held on Thursday, July 17, 2014 in the College Park, Maryland metropolitan area. A brief summary of the program and its goals is copied below; the full announcement is available at https://www.fbo.gov/notices/9b5c9fedae00bd2f2b3f4a1770451596
. Hope to see you there!
Program Description and Goals
For many information processing tasks, the brain employs algorithms composed of multiple instances of a limited set of computing "primitives" arrayed in a multi-stage processing architecture. Neurons in these primitives operate in parallel and communicate with their neighbors above, below, and laterally within the network to make sense of the complex environments in which we live. Today's state of the art algorithms for machine learning take a similar form, but deviate significantly in the details of implementation. Presumably, a significant part of the performance gap separating artificial and biological computing today is due to these deviations. The MICrONS program is predicated on the notion that it will be possible to revolutionize machine intelligence if we can construct algorithms that utilize the same data representations, transformations, and learning rules as those employed and implemented by the cortical computing primitives.
Although a significant body of neuroscience data has been collected over the past 100+ years, the majority of what is known about the brain is about its microscale (one or a few neurons) or macroscale (hundreds of thousands or millions of neurons) operation. Much less is known about the detailed structure and function of the mesoscale cortical microcircuits (hundreds to tens of thousands of neurons) that embody the cortical computing primitives, because until recently there have been few tools available to interrogate the brain at the requisite resolution (nanometers) and scale (millimeters). MICrONS seeks to use emerging technologies in high-resolution and high-throughput brain mapping-such as serial electron microscopy and volumetric calcium imaging-to address this gap in our understanding of cortical computation and to exploit the findings to enhance machine intelligence.
The overall and specific goal of the MICrONS program is to create a new generation of machine learning algorithms derived from high-fidelity representations of cortical microcircuits to achieve human-like performance on complex information processing tasks. To achieve this goal, multidisciplinary teams will:
• Propose an algorithmic framework for information processing that is consistent with existing neuroscience data, but that cannot be fully realized without additional specific knowledge about the data representations, computations, and network architectures employed by the brain;
• Collect and analyze high-resolution data on the structure and function of cortical microcircuits believed to embody the cortical computing primitives underlying key components of the proposed framework;
• Generate computational neural models of cortical microcircuits informed and constrained by this data and by the existing neuroscience literature to elucidate the nature of the cortical computing primitives; and
• Implement novel machine learning algorithms that use mathematical abstractions of the identified cortical computing primitives as their basis of operation.
It is anticipated that algorithms created under MICrONS will be validated through their performance on complex auditory or visual scene parsing tasks, and will also demonstrate capacity for generalization to abstract, non-sensory data.