On June 1, the Defense Advanced Research Projects Agency (DARPA) released the Enabling Confidence (EC) Artificial Intelligence Exploration (AIE) Opportunity. Submissions are due by 4:00 p.m. Eastern on June 30.
DARPA invites submissions of innovative basic or applied research concepts in the technical domain of scalable methods to generate accurate statistical models for the outputs of machine learning (ML) systems.
This AIE Opportunity is issued under the Program Announcement for AIE, DARPA-PA-21-04. All awards will be made in the form of an Other Transaction (OT) for prototype project. The total award value for the combined Phase 1 Feasibility Study (Base) and Phase 2 Proof of Concept (Option) is limited to $1,000,000 per proposal. This total award value includes Government funding and performer cost share, if required or if proposed.
Accurate processing of statistical information related to environmental variations and sensor noise is paramount to the performance of statistics-based estimators (e.g., the use of covariance matrices in Kalman filters), and is the key enabler for optimally combining information originating from multiple heterogeneous sensors and subsystems.
Unlike many signal processing systems, ML systems do not currently fit neatly into statistics-based estimation chains, where input noise, frequently represented by covariance matrices, can be accurately reflected in the output covariance. Transforming input noise from sensors or environmental sources into accurate ML output statistics – in a way that scales favorably with the size of the ML system – is challenging. Several techniques have been developed to quantify the uncertainty in ML outputs; however, even the most advanced techniques only estimate output variances – not covariances.
EC will develop scalable methods to generate accurate statistical models for the outputs of ML systems, to enable enhanced performance when using this information to combine multiple subsystems. The EC AIE Opportunity will encourage performers to consider a range of ML techniques – e.g., Deep Learning, Bayesian techniques, etc. – in addressing the following research questions:
- Can input sensor and environmental covariance information be faithfully reflected in ML system output statistical models in a way that is computationally tractable? a. Example approaches could include: propagating input covariance information through a neural network to determine the output model; modifying a ML algorithm to impose constraints so that the output model is accurately represented by a covariance matrix; etc.
- Can confident ML subsystems be composed hierarchically to increase inference accuracy, and can they be combined with statistics-based estimation systems (e.g., Kalman filters) to reduce errors?
Review the DARPA EC AIE opportunity.
Source: SAM
IC News brings you business opportunities like this one each week. If you find value in our work, please consider supporting IC News with a subscription.
By 
