### Table 2: Features of POMDP and DBC induced policies

"... In PAGE 4: ... The following two subsections discuss properties of POMDP and DBC induced policies in the light of features such as user preference interpretation and similarity to controller mechanisms. The summary of this discussion is available in Table2 , with numbers in parentheses denoting the subsection in which specific... ..."

### Table 1: Sample Conversation with Flo ment calender) are allowed, and can be disambiguated through proper action selection. This was done auto- matically when solving the POMDP models and ob- taining a policy.

### Table 1: Ratios of the weighted execution time of the rollout scheduler to the DEC scheduler. A ratio of less than one means that the rollouts outperformed the DEC scheduler.

1998

"... In PAGE 3: ... Because the running time increases quadratically with the number of rollouts, we focused our rollout experiments on one program in the SPEC95 suite: applu. Table1 gives the performance of each rollout scheduler as compared to the DEC scheduler on all 33,007 basic blocks of size 200 or less from applu. To assess the performance of each rollout policy , we used the ratio of the weighted execution time of the rollout scheduler to the weighted execution time of the DEC scheduler.... ..."

Cited by 10

### Table 2: Value Iteration for an E-PCA POMDP

2005

"... In PAGE 22: ... For each transition ~b i ! b ! ba ! b0 ! ~b0 ! ~b j we can assign a probability p(z; jji; a) = p(zjba) w(~b j;~b0) = w(~b j;~b0) jSj X l=1 p(zjsl)ba(sl) (36) The total transition probability ~ T (~b i ; a;~b j) is the sum, over all observations z, of p(z; jji; a). Step 3 in Table2 performs this computation, but shares work between the computation of ~ T (~b i ; a;~b j) for di erent posterior beliefs ~bj which are reachable from the same prior belief ~bi under action a. Computing the Value Function With the reward and transition functions computed in the previous sections, we can use value iteration to compute the value function for our belief space MDP.... ..."

Cited by 23

### Table 2: Value Iteration for an E-PCA POMDP

2005

"... In PAGE 22: ... For each transition ~b i ! b ! ba ! b0 ! ~b0 ! ~b j we can assign a probability p(z; jji; a) = p(zjba) w(~b j;~b0) = w(~b j;~b0) jSj X l=1 p(zjsl)ba(sl) (36) The total transition probability ~ T (~b i ; a;~b j) is the sum, over all observations z, of p(z; jji; a). Step 3 in Table2 performs this computation, but shares work between the computation of ~ T (~b i ; a;~b j) for di erent posterior beliefs ~bj which are reachable from the same prior belief ~bi under action a. Computing the Value Function With the reward and transition functions computed in the previous sections, we can use value iteration to compute the value function for our belief space MDP.... ..."

Cited by 23

### Table 2: Value Iteration for an E-PCA POMDP

2005

Cited by 23

### Table 1: Ratios of the weighted execution time of the rollout scheduler to the DEC sched- uler. A ratio of less than one means that the rollouts outperformed the DEC scheduler.

1998

"... In PAGE 3: ... Because the running time increases quadratically with the number of rollouts, we focused our rollout experiments on one program in the SPEC95 suite: applu. Table1 gives the performance of each rollout scheduler as compared to the DEC scheduler on all 33,007 basic blocks of size 200 or less from applu. To assess the performance of each rollout policy a3 , we used the ratio of the weighted execution time of the rollout scheduler to the weighted execution time of the DEC scheduler.... ..."

Cited by 10

### Table 1 Comparison of different power management policy. POMDP MDP+partial weight

in Stochastic Modeling and Optimization for Robust Power Management in a Partially Observable System

"... In PAGE 6: ... It out performs the MDP policy in a partially observable system. Figure 4 Comparison of energy latency tradeoff Table1 gives the comparison of the average latency (D), loss rate (L) and power (P). The first column gives the latency weight in the reward function.... ..."

### Table 1 Policy iteration

"... In PAGE 4: ... There are two related dynamic programming algorithms for indefinite-horizon MDPs: policy iteration and value iteration. Policy iteration is summarized in Table1 . It interleaves the dynamic-programming update, used for policy improvement, with policy evaluation.... ..."