| … | … | @@ -99,6 +99,13 @@ Walk through possible connections, randomly. But rather than uniform |
| 99 | 99 | weighting, we need to spread "attention" according to weights. This |
| 100 | 100 | would require a redesign of the attention allocation subsystem. |
| 101 | 101 | |
| 102 | #### Fun demo! |
| 103 | Use a small collection of words from text chat to deposit attention |
| 104 | into the network. Then trace throught grammatically network crawls |
| 105 | throught the net. Do this on an IRC chat channel. (This would resemble |
| 106 | the behavior of bayesian/markov language generators, but different, |
| 107 | because the network is defined differently.) |
| 108 | |
| 102 | 109 | |
| 103 | 110 | Literature |
| 104 | 111 | ========== |