Comments on: Time for the next generation of knowledge automation http://haleyai.com/wordpress/2009/11/01/time-for-the-next-generation-of-knowledge-acquisition-management-and-automation/ systems that know and understand and think and learn Fri, 20 Jan 2012 02:37:12 +0000 http://wordpress.org/?v=2.9.1 hourly 1 By: Peter Lin http://haleyai.com/wordpress/2009/11/01/time-for-the-next-generation-of-knowledge-acquisition-management-and-automation/comment-page-1/#comment-67 Peter Lin Thu, 03 Dec 2009 15:10:17 +0000 http://haleyai.com/wordpress/2009/11/01/time-for-the-next-generation-of-knowledge-acquisition-management-and-automation/#comment-67 I just noticed you responded to my comment. The kind of change I'm thinking isn't really stale data. I was thinking more about data that has business meaning like a FIX transaction on the stock exchange. If a system is calculating aggregates on securities transactions and an order gets cancelled, the system may need to perform corrective action on any of the calculated aggregates. What I did to solve this problem was to extend CLIPS syntax. I added the concept of modification logic to the RHS. This way, the user can define corrective action when a fact is retracted. I wrote a paper on it and explained the implementation details. I believe I emailed it to your old email address a while back. You can download it here http://jamocha.svn.sourceforge.net/viewvc/jamocha/morendo/doc/modification_logic.pdf?revision=2586 I just noticed you responded to my comment. The kind of change I’m thinking isn’t really stale data. I was thinking more about data that has business meaning like a FIX transaction on the stock exchange. If a system is calculating aggregates on securities transactions and an order gets cancelled, the system may need to perform corrective action on any of the calculated aggregates. What I did to solve this problem was to extend CLIPS syntax. I added the concept of modification logic to the RHS. This way, the user can define corrective action when a fact is retracted. I wrote a paper on it and explained the implementation details. I believe I emailed it to your old email address a while back. You can download it here
http://jamocha.svn.sourceforge.net/viewvc/jamocha/morendo/doc/modification_logic.pdf?revision=2586

]]> By: pvhaley http://haleyai.com/wordpress/2009/11/01/time-for-the-next-generation-of-knowledge-acquisition-management-and-automation/comment-page-1/#comment-54 pvhaley Wed, 11 Nov 2009 22:31:41 +0000 http://haleyai.com/wordpress/2009/11/01/time-for-the-next-generation-of-knowledge-acquisition-management-and-automation/#comment-54 No, Peter, ART did not have anything for time or events. The CEP engines I am familiar with remove events when they are no longer in a window that they are tracking, but there are open-ended issues. I'm not sure what it would mean to "cancel" an event, though. Something is an event after it happens, unless it is a process, in which case I guess you could cancel it if it was present progressive or future tense. With regard to data, I would think it might be better to describe when it holds than to call it a version. Version doesn't have the connotation of state that I think you want. BTW, there are a number of approaches to data that varies over time or hypothetical states. ART had a context mechanism that could address it and UTexas' "Knowledge Machine" has situations an Cyc has contexts. These are all better for parallel hypotheticals than sequential states, however (for reasons that I will spare you here). You might also want to look at Kowalski's event calculus. He reminded me about it recently, but I still think some architectural support for states is better than the flat alternatives. No, Peter, ART did not have anything for time or events. The CEP engines I am familiar with remove events when they are no longer in a window that they are tracking, but there are open-ended issues. I’m not sure what it would mean to “cancel” an event, though. Something is an event after it happens, unless it is a process, in which case I guess you could cancel it if it was present progressive or future tense. With regard to data, I would think it might be better to describe when it holds than to call it a version. Version doesn’t have the connotation of state that I think you want. BTW, there are a number of approaches to data that varies over time or hypothetical states. ART had a context mechanism that could address it and UTexas’ “Knowledge Machine” has situations an Cyc has contexts. These are all better for parallel hypotheticals than sequential states, however (for reasons that I will spare you here). You might also want to look at Kowalski’s event calculus. He reminded me about it recently, but I still think some architectural support for states is better than the flat alternatives.

]]> By: Peter Lin http://haleyai.com/wordpress/2009/11/01/time-for-the-next-generation-of-knowledge-acquisition-management-and-automation/comment-page-1/#comment-50 Peter Lin Fri, 06 Nov 2009 16:11:31 +0000 http://haleyai.com/wordpress/2009/11/01/time-for-the-next-generation-of-knowledge-acquisition-management-and-automation/#comment-50 I like the new layout. Did ART provide support for temporal logic or event semantics? One thing I feel the CEP/ESP engines do not address is temporal versioning. What happens when an event needs to be removed or cancelled? Most of the CEP engines do not tackle this issue. In my discussion with CEP proponents, they consider a change a second event, which "was triggered" by the same piece of data. The problem is that data has changed, so each derived event has to know the version. I enjoyed the entry. peter lin I like the new layout. Did ART provide support for temporal logic or event semantics? One thing I feel the CEP/ESP engines do not address is temporal versioning. What happens when an event needs to be removed or cancelled? Most of the CEP engines do not tackle this issue. In my discussion with CEP proponents, they consider a change a second event, which “was triggered” by the same piece of data. The problem is that data has changed, so each derived event has to know the version.

I enjoyed the entry.

peter lin

]]>