DARPA funds $11 million tool that will make coding a lot easier


That's a lot of money for a good idea (using large bodies of existing code to complete what the programmer write's) that everyone seems to be working on. Does Rice have some special advantage here that justifies the investment?

Comment viewing options

Select your preferred way to display the comments and click "Save settings" to activate your changes.

Standard DARPA

This is part of a large DARPA project called MUSE, more info here: http://www.darpa.mil/Our_Work/I2O/Programs/Mining_and_Understanding_Software_Enclaves_%28MUSE%29.aspx

The size of the grant is big, but not out of line with other DARPA grants to multi-institution projects, which are regularly multiple millions.

It doesn't say any other

It doesn't say any other institutions are involved; $11 million is huge for one PI. In the Rice press release, you find out at the end that U Wisc, UTA, and Grammartech is also involved.

I wonder why they decided to go with Bayesian statistics rather than recurrent neural networks? Hinton et al trained up some RNNs using Wikipedia and was able to "auto-complete" based on that...pretty freaky. Their choice seems to be pretty weak from an ML perspective.