Cortical Computing Primitives and Connectomics

The Intelligence Advanced Research Projects Activity (IARPA) is seeking information about the current state of the art in "executable" (i.e. instantiable) models of cortical computing primitives that are supported by known neuroanatomy, as well as strategies for advancing these models based on new and emerging techniques in connectomics. This information may be used to formulate a new program aimed at developing novel machine learning algorithms based on high fidelity representations of cortical microcircuits. The current request for information (RFI) is issued solely for information gathering and planning purposes and does not constitute a formal solicitation for proposals. The following sections of this RFI contain details of the scope of technical areas of interest, along with instructions for the submission of responses.


IARPA is interested in understanding the neural algorithms that form the basis of inference and recognition in the brain as a potential basis for creating new types of machine learning algorithms that exhibit more human-like performance characteristics than today's leading systems. Many contemporary theories of neural information processing suggest that within a given cortical region or cognitive/sensory domain, the brain employs hierarchical algorithms composed of repeated instances of a limited set of computing "primitives" or modules. These primitives are further theorized to be embodied in cortical microcircuits at various scales. Although there has been significant progress in understanding multiple aspects of cortical microcircuits and the larger networks in which they are embedded, a comprehensive description of their structure, function, and interconnectivity remains elusive. Consequently, myriad mathematical, computational, conceptual, and schematic models have been proposed to describe the nature of the cortical computing primitives and the hierarchical algorithms that employ them.

This RFI specifically addresses mathematical, computational, or otherwise executable models of cortical computing primitives that are supported by known neuroanatomy. Information regarding models that are purely conceptual, schematic, or descriptive in nature (i.e. models that cannot be instantiated), or regarding models that are not grounded in neuroscience, is not sought at this time. Within these constraints, IARPA seeks to better understand the capabilities and limitations of existing models, as well as strategies for improving upon the current state of the art by incorporating new information about the static and dynamic properties of cortical microcircuits and their associated networks.


Responses to this RFI will help IARPA to identify promising areas for seedlings and programs. They will be used also in the planning of an agenda and participant list for a potential workshop to promote discussions on how high fidelity reconstructions of cortical microcircuits could be used to inform models of cortical computing primitives, and on how those models could in turn be used to create novel machine learning algorithms. For example, some cortical computing primitives have recently been described in terms of probabilistic graphical models-IARPA is interested in whether and how these and other types of models of cortical computing primitives could utilize and benefit from the detailed neuroanatomical and neurophysiological information being revealed by structural and functional brain mapping.

Although the scope and purpose of any future workshop or program remain to be determined, one possibility is that it would bring together experts in such fields as connectomics, computational neuroscience, computer vision, graph theory, machine learning, and systems engineering to formulate strategies about how these fields could work together to make significant advances in machine intelligence. To facilitate future collaborations, IARPA encourages experts in all of these topics and in related areas of scientific inquiry to respond to this RFI either individually or as part of a team. If appropriate, a separate workshop announcement will be posted at a later date with additional details.


Through this RFI, IARPA hopes to learn more about the following topics:

  • The biological fidelity and practical utility of existing executable and neurally-grounded models of cortical computing primitives, and algorithms that employ these primitives to perform inference and recognition tasks.
  • Ideas for how these models could be improved by incorporating additional details about cortical microcircuit structure and function.
  • Techniques for generating these details, including (but not limited to) new and emerging tools employed in connectomics research such as high-throughput imaging, automated analysis of imaging data, and graph theoretical statistics.
  • Methods for validating the identified microcircuit characteristics.

More specifically, IARPA seeks responses to this RFI that address any or all of the following questions:

  1. Models of Cortical Computing Primitives
    • What is the state of the art in executable and neurally-grounded models of cortical computing primitives and the larger networks in which they are embedded?
    • Have these models been used to solve real-world problems? If not, why not? If so, what metrics are used to assess their performance? How does this performance compare to other algorithms used to solve similar problems?
    • Are there specific types of data and/or tasks for which these models are particularly well-suited? If so, why?
    • Have there been any simulations or other experiments demonstrating how these models could be implemented by biologically-plausible (e.g., spiking) neurons and neural circuit architectures? How could the correspondence between model and simulation be quantified and validated?
    • Are there other compelling theories of cortical computing that do not include hierarchical algorithms composed of computing primitives? What kinds of models are available to emulate these processes?
  2. Evidence
    • What empirical evidence supports the existence of cortical microcircuits that implement these models of computing primitives? What aspects of the models are not currently supported by empirical evidence?
    • How prevalent are these microcircuits in the brain? How many different types are there likely to be? Where in the brain would one look for microcircuits that support learning and inference tasks? What other tasks do they support?
    • What is the minimum set of new information about cortical microcircuits that would significantly advance the state of the art of these models? How would this information (e.g., connectivity, synaptic weights, glial distribution, neuromodulator state, gene expression, neural activity patterns, etc.) be specified (e.g., precise quantities, probability distributions, etc.)? How many exemplars or samples of cortical microcircuits would be required to specify this information? Is static information sufficient, or are dynamics required in this minimum set?
    • How would this information be used to update the models? What specific aspects and/or parameters of the model would this information affect?
    • How could the fidelity of the updated models be assessed? What are the dimensions in which improvement could be measured? What metrics should be used to assess progress?
    • How is this information likely to affect model performance on inference and recognition tasks?
  3. Data collection
    • What tools or techniques (e.g., electron microscopy, array tomography, super-resolution imaging, immunostaining, genetic engineering, semi-/automated segmentation, graph theory, etc.) could be used to obtain the "new information" described above?
    • What are the major challenges involved in collecting, storing, processing and/or analyzing this information?
    • How long would it take to collect this information using currently available techniques and foreseeable near-term improvements on the current state of the art?
    • How could the accuracy of this information be assessed? What metrics should be used to assess progress?
  4. Implementation
    • What hardware platforms are best suited for implementing these models at a scale sufficient to perform real-world inference and recognition tasks?
    • What are likely to be the key challenges in implementing these models?

Preparation Instructions to Respondents

IARPA solicits respondents to submit ideas for use by the Government in formulating a potential program that uses detailed information on the structure and/or function of cortical microcircuits to create high-fidelity models of cortical computing primitives that can in turn be used by novel machine learning algorithms to perform inference and recognition tasks. IARPA requests that submissions briefly and clearly describe a potential approach or concept on this topic and address the questions posed above in the context of this approach or concept. If appropriate, respondents may also choose to provide a non-proprietary rough order of magnitude (ROM) cost estimate regarding what such approaches might require in terms of funding and other resources for one or more years. This RFI contains all of the information required to submit a response. No additional forms, kits, or other materials are needed.

IARPA appreciates responses from all capable and qualified sources from within and outside of the US. Because IARPA is interested in integrated approaches, responses from teams with complementary areas of expertise are encouraged.

Responses have the following formatting requirements:

  1. A one page cover sheet that identifies the title, organization(s), respondent's technical and administrative points of contact (including names, addresses, phone and fax numbers, and email addresses of all co-authors), and clearly indicating its association with this RFI (RFI-13-05);
  2. A substantive, focused, one page executive summary;
  3. A description of the suggested approach and response to any or all of the questions posed above, limited to 5-10 single-sided, single-spaced 8.5 x 11 inch pages using a minimum of 12 point Times New Roman font (including captions and tables) and 1-inch margins on all sides;
  4. A list of citations and a copy of each reference (to be attached to the PDF or otherwise included in the submission package);
  5. Optionally, a 5-slide PowerPoint presentation that graphically depicts the key ideas.

Disclaimers and Important Notes

This is an RFI issued solely for information and planning purposes and does not constitute a solicitation. Respondents are advised that IARPA is under no obligation to acknowledge receipt of the information received, or provide feedback to respondents with respect to any information submitted under this RFI.

Responses to this notice are not offers and cannot be accepted by the Government to form a binding contract. Respondents are solely responsible for all expenses associated with responding to this RFI. IARPA will not provide reimbursement for costs incurred in responding to this RFI or reimbursement for travel. It is the respondent's responsibility to ensure that the submitted material has been approved for public release by the information owner.

The Government does not intend to award a contract on the basis of this RFI or to otherwise pay for the information solicited, nor is the Government obligated to issue a solicitation based on responses received. Neither proprietary nor classified concepts or information should be included in the submittal. Input on technical aspects of the responses may be solicited by IARPA from non-Government consultants/experts who are bound by appropriate non-disclosure requirements.


For information contact:


Posted Date: July 29, 2013
Responses Due: September 30, 2013