Depending upon your perspective Artificial Intelligence is stuck in the doldrums. Its long touted promise has yet to materialize and instead produced boom-bust cycles of grand expectations followed by disappointing results. Maybe the problem isn’t the algorithms but training data sets. This is the latest message from IARPA which has issued a request for information (RFI) for “Novel Training Datasets and Environments to Advance Artificial Intelligence.”
Here’s a brief excerpt from the RFI’s Background and Scope:
“Until recently, the conventional wisdom has been that new algorithms were the limiting factor in making steady progress towards artificial intelligence. However, recent advances in machine learning, a sub-field of artificial intelligence, have established that historical algorithms (e.g. backpropagation) in conjunction with high-performance computers can be used to achieve nearly human-level performance on diverse tasks such as image and speech recognition, language translation, and video game play.
“In each of these instances, and in many others, rapid progress was facilitated by the availability of massive amounts of training data well-suited to the problem under study. This realization raises the prospect that many additional artificial intelligence problems may be solvable in the near-term, without significant innovations in the underlying algorithms, if the right training resources become widely available”
IARPA (Intelligence Advanced Research Projects Activity) is seeking input from the artificial intelligence research community on training resources most likely to drive progress in new problem domains within the field of artificial intelligence (including machine learning). In particular, the RFI seeks input in five areas:
- Which problem domain(s) has the greatest potential to benefit from the availability of new training resources and why?
- What new training resources are needed to achieve significant progress in this domain? How should these resources be structured? How do the proposed resources compare with currently available resources?
- What kind of effort is needed to create and/or curate these training resources? What technical, logistical, and/or legal challenges would be associated with such an effort? How much would such an effort cost, and how long would it take? How much effort and money would be required to store, maintain, distribute, and/or utilize the proposed training resources?
- Who would be the major stakeholders in the proposed training resources? How would these stakeholders use the proposed resources?
- Annual challenges (e.g. ImageNet Large Scale Visual Recognition Challenge) employing a standard set of data for training and/or evaluation have helped to catalyze progress in many machine learning problem domains. Should a challenge be created in the proposed problem domain, and if so, how should it be designed, implemented, and judged?
Responses are due by April 1. Here is a link for more information on the RFI, https://www.fbo.gov/index?s=opportunity&mode=form&id=bdbf68f721889b06eafc934d562fe20f&tab=core&_cview=0