OpenCog HK 2014 Generic AI Task Breakdown

Task 1: Make ECAN work scalably

The current ECAN implementation has a number of shortcomings, and could use a major refactoring, e.g.

• Elimination of internal use of matrices, in favor of direct use of the Atomspace
• Elimination of dependence on the cognitive cycle, as MindAgents will later this year be made to run in their own threads
• Various small fixes to the equations

Tentatively assigned to: New Addis hire
Current time estimate: 3 months
Tentatively scheduled for: July-Sep

Task 2: Port PLN to unified rule engine

Once we have a unified rule engine, the PLN rules can be ported there, and PLN can be implemented as a customization of the new rule engine

Tentatively assigned to: Alex, with support from Misgana/Amen/Cosmo
Current time estimate: 3 months
Tentatively scheduled for: Aug-Oct

Task 3: Integrate temporal reasoning code into PLN

Keyvan has written code for temporal inference based on Interval Algebra. It needs to be integrated into PLN, to work with the code Dario/Jade wrote for temporal inference using Interval Algebra.

Tentatively assigned to: New Addis Hire
Current time estimate: 2 months (because this is for a new hire who won't understand what's going on)
Tentatively scheduled for: Sep-Oct

Task 4: Implement spatial reasoning using RCC-3D, and integrate with PLN

Something analogous to what has been done for Interval Algebra (Task 3) must be done for spatial reasoning as well.

Tentatively assigned to: New Addis Hire
Current time estimate: 4 months 
Tentatively scheduled for: Nov 2014 - Feb 2015

Task 5: Get ECAN-guided inference control to accelerate PLN reasoning

This is best pursued after ECAN is fixed and PLN is ported to the new rule engine, it would seem.

Tentatively assigned to: New Addis Hire
Current time estimate: 4 months 
Tentatively scheduled for: Nov 2014 - Feb 2015

Task 7: Get basic syllogistic inference to work based on English language inputs

Get RelEx2Logic, PLN and NLGen to work together for basic syllogistic inference.

See List of Test Syllogisms here

Task 7: Improve DeSTIN functionality for object and image classification

During Summer 2014, Teddy and Yuhuang Hu and others are experimenting with ways to make the DeSTIN vision processing tool work better. Some of this work is in the main DeSTIN C codebase, and some is in various MATLAB versions. The improvements found to actually help should be ported to the C codebase and effectively organized, and the system should then be tested not only on standard image and video corpora but also on the output of robot vision.

Tentatively assigned to: New Addis Hire (replacing Teddy who is leaving for grad school at the end of August)
Current time estimate: ongoing
Tentatively scheduled for: the rest of 2014