In tests, Indyk's team showed that their learning-based approach had upwards of 57percentt fewer errors for estimating the amount of internet traffic in a network, and upwards of 71percentt fewer errors for estimating the number of queries for a given search term.
Although the NNP method resulted in faster skill acquisition, MTL prompting was associated with fewer errors for all participants and better performance during 1- and 2-week maintenance probes for 3 of the 4 participants.
There was also a main effect of gF, F 1,17) = 6.9, p = .018, with fewer errors for the high gF group (3.3%) compared to the low gF group (9.7%).
Similarly, for length 3, there were fewer errors for E-E than for I-I (F 1, 10) = 20.64, p<0.001) and E-I (F 1, 10) = 11.70, p<0.01), which did not differ from each other significantly.
Simple comparisons showed that there were fewer errors for E-E than for E-I (F 1, 22) = 11.06, p<0.01) and I-I (F 1, 22) = 5.66, p<0.05), which did not differ from each other.
Below we show that our approach produces fewer errors for molecular subtype classification.
Consistent with this interpretation is the fact that, unlike keypress responses, pointing responses yielded significantly fewer errors for N = 6 than for N = 2. Thus, it is at least plausible that participants perceived the movements to be more difficult when N = 6 and so slowed down their movements slightly, which made them more accurate.
And he makes a few errors — saying, for instance, that Elvis Costello covered "Hot Burrito #2" on "Almost Blue" when it was actually "Hot Burrito #1".
