### Table 10 Frequency of Occurrence When the Peer Reviewer Identified a Problem and Provided a Solution Topic of Feedback Identified problem and

2003

### Table 5.2: Feedback control action and closed-loop cost for MPC and optimal control.

1994

Cited by 1

### Table 6.1.1 Example case (from the Pre- cooked plans Case-Base) 4. Process user feedback 5. Re-adapt solution

1998

Cited by 9

### Table 6.1.1 Example case (from the Pre- cooked plans Case-Base) 4. Process user feedback 5. Re-adapt solution

### Table 1. Immediate/delay feedback results

"... In PAGE 4: ... The basic statistics for these data- sets are given as following: MrX2 (9032 ham, 40135 spam), SB2 (9274 ham, 2751 spam). The performance of kidult anti-spam solution is given in Table1 -Table 2. Results are included for 2 corpora, with immediate feedback, delayed feedback, and active learning as denoted by the run tag suffix: x2 (MrX2 corpus, immedi- ate feedback), x2d (MrX2 corpus, delayed feedback), x2a (MrX2 corpus, active learning), b2 (SB2 corpus, immediate feedback), b2d (SB2 corpus, delayed feedback), b2a (SB2 corpus, active learning).... ..."

### Table 1: The feedback function apos; Fig. 5 shows the nal solution for the circular- railway example. Less than half of the possible nodes have been inserted in the graph. The feedback func- tion apos; is given in Table 15. Since the LCBE of each node is useless in the closed-loop system, only the marking of each node is kept in the solution. 5In column labeled Ti, i = 1;2;3 or 5, Trk must be read f lt; x = Trk gt;g (k = 1 or 2).

### Table 3: Experiment 1: tPi;core = 9ms The two experiments show that, while purely feed-forward event propagation through properly described local analysis domains will always lead to a solution, the feedback situation is more complicated (as it is in practice). There are cases where the feedback approach used in the example does not lead to a solution (i.e. a convergence), even though there exists a valid schedule in practice. This can be proved by analyzing the average processor utilization. With 91,67% (55 60), the system is in fact schedulable, i.e. there is a point in time where to- tal system workload will be zero and buffers are empty. However, response time analysis is more complicated than only guaranteeing schedulability.

"... In PAGE 5: ... At each step, we calculate new internal jitters. The results of experiment 1 can be seen in Table3 . We see, that the algorithm terminates after three steps, i.... ..."

### Table 1: Experimental results. The average performance (total sum of evaluations) relative to the performance with the optimal feedback. Between brackets the standard deviation over all trials. The * indicate the number of solutions that resulted in an unstable closed loop (these test results are omitted in the computation of the performances).

1998

"... In PAGE 5: ... The performances of the approximated feedbacks are determined by running the system with that feedback and sum- marize the evaluations. In Table1 the experimental results are shown. It is clear that the system identi cation approach has the best result, because it only has to approximate A... In PAGE 6: ... Although the number of parameters of the PPQL approach is much larger than the number of the parametric of Q-learning approach, it still performs reasonably well. However, there are exceptions to this, indicated by the * in Table1 . These represent the situations in which the approximated feedback results in an unstable closed loop.... ..."

Cited by 1

### Table 4.1 shows the solution rates with the two forms of feedback. Note that previous studies have all shown a solution rate of near 0% when given a comparably short period of time to attempt the problem (MacGregor et al. 2001). Of the twelve using the positive version who solved it, three solved it after only one attempt and one on their rst. It is unlikely that they solved it because of the feedback they received, especially given that the three who solved it after only one attempt all received di erent feedback on their rst attempt. Because it is considered likely that these four did not solve because of the feedback, but may have been extraordinarily good at the nine-dots problem, or may have had previous exposure to the problem (which they all claimed they had not), results for positive are also given excluding these four. Even when they are included, an unpaired t-test with unequal variances shows there to be no signi cant di erence between the rates of solution of the two groups.

2004