Optimal Scheduling Policy Determination in HSDPA Networks
|
|
- Todd Shields
- 5 years ago
- Views:
Transcription
1 Optimal Scheduling Policy Determination in HSDPA Networks Hussein Al-Zubaidy, Jerome Talim, Ioannis Lambadaris SCE-Carleton University 1125 Colonel By Drive, Ottawa, ON, Canada {hussein, jtalim, 26 June 2007 ussein Al-Zubaidy, Jerome Talim, Ioannis Lambadaris Optimal Scheduling [8ex] SCE-Carleton Policy Determination University1125 in HSDPA Colonel Networks By Drive, Ottawa, 26 June 2007 ON, Canada 1 / Emai 34
2 Outline 1 Objective and Motivation 2 Methodology 3 Problem Definition and Model Description 4 Case Study and Results 5 Conclusion and Future Work ussein Al-Zubaidy, Jerome Talim, Ioannis Lambadaris Optimal Scheduling [8ex] SCE-Carleton Policy Determination University1125 in HSDPA Colonel Networks By Drive, Ottawa, 26 June 2007 ON, Canada 2 / Emai 34
3 Objective Objective and Motivation Objective To devise a methodology to find the optimal scheduling regime in HSDPA networks, that controls the allocation of the time-code resources. This resulting optimal policy should have the following properties: ussein Al-Zubaidy, Jerome Talim, Ioannis Lambadaris Optimal Scheduling [8ex] SCE-Carleton Policy Determination University1125 in HSDPA Colonel Networks By Drive, Ottawa, 26 June 2007 ON, Canada 3 / Emai 34
4 Objective Objective and Motivation Objective To devise a methodology to find the optimal scheduling regime in HSDPA networks, that controls the allocation of the time-code resources. This resulting optimal policy should have the following properties: Fair; Divide the resources fairly between all the active users. ussein Al-Zubaidy, Jerome Talim, Ioannis Lambadaris Optimal Scheduling [8ex] SCE-Carleton Policy Determination University1125 in HSDPA Colonel Networks By Drive, Ottawa, 26 June 2007 ON, Canada 3 / Emai 34
5 Objective and Motivation Objective Objective To devise a methodology to find the optimal scheduling regime in HSDPA networks, that controls the allocation of the time-code resources. This resulting optimal policy should have the following properties: Fair; Divide the resources fairly between all the active users. Optimal Transmission: Maximizes the overall cell throughput. ussein Al-Zubaidy, Jerome Talim, Ioannis Lambadaris Optimal Scheduling [8ex] SCE-Carleton Policy Determination University1125 in HSDPA Colonel Networks By Drive, Ottawa, 26 June 2007 ON, Canada 3 / Emai 34
6 Objective and Motivation Objective Objective To devise a methodology to find the optimal scheduling regime in HSDPA networks, that controls the allocation of the time-code resources. This resulting optimal policy should have the following properties: Fair; Divide the resources fairly between all the active users. Optimal Transmission: Maximizes the overall cell throughput. Optimal Resource Utilization: Provide channel aware (diversity gain) and high speed resource allocation. ussein Al-Zubaidy, Jerome Talim, Ioannis Lambadaris Optimal Scheduling [8ex] SCE-Carleton Policy Determination University1125 in HSDPA Colonel Networks By Drive, Ottawa, 26 June 2007 ON, Canada 3 / Emai 34
7 Objective and Motivation Motivation Motivation 3GPP only suggested some guidelines for HSDPA downlink scheduler and left the design specifics undefined. This resulted in many different scheduling techniques and implementations most of which are proprietary. Most of the available work in scheduler design is based on intuition and creativity of the designers. ussein Al-Zubaidy, Jerome Talim, Ioannis Lambadaris Optimal Scheduling [8ex] SCE-Carleton Policy Determination University1125 in HSDPA Colonel Networks By Drive, Ottawa, 26 June 2007 ON, Canada 4 / Emai 34
8 Methodology Methodology This work presents a different approach for scheduling in HSDPA. A declarative approach is used, Hussein Al-Zubaidy, Jerome Talim, Ioannis Lambadaris Optimal Scheduling [8ex] SCE-Carleton Policy Determination University1125 in HSDPA Colonel Networks By Drive, Ottawa, 26 June 2007 ON, Canada 5 / Emai 34
9 Methodology Methodology This work presents a different approach for scheduling in HSDPA. A declarative approach is used, Develop an analytic model for the HSDPA downlink scheduler. A MDP based discrete stochastic dynamic programming model is used to model the system. This Model is a simplifying abstraction of the real scheduler which estimates system behavior under different conditions and describes the role of various system components in these behaviors. It must be solvable. Hussein Al-Zubaidy, Jerome Talim, Ioannis Lambadaris Optimal Scheduling [8ex] SCE-Carleton Policy Determination University1125 in HSDPA Colonel Networks By Drive, Ottawa, 26 June 2007 ON, Canada 5 / Emai 34
10 Methodology Methodology This work presents a different approach for scheduling in HSDPA. A declarative approach is used, Develop an analytic model for the HSDPA downlink scheduler. A MDP based discrete stochastic dynamic programming model is used to model the system. This Model is a simplifying abstraction of the real scheduler which estimates system behavior under different conditions and describes the role of various system components in these behaviors. It must be solvable. Define an objective function. Hussein Al-Zubaidy, Jerome Talim, Ioannis Lambadaris Optimal Scheduling [8ex] SCE-Carleton Policy Determination University1125 in HSDPA Colonel Networks By Drive, Ottawa, 26 June 2007 ON, Canada 5 / Emai 34
11 Methodology Methodology This work presents a different approach for scheduling in HSDPA. A declarative approach is used, Develop an analytic model for the HSDPA downlink scheduler. A MDP based discrete stochastic dynamic programming model is used to model the system. This Model is a simplifying abstraction of the real scheduler which estimates system behavior under different conditions and describes the role of various system components in these behaviors. It must be solvable. Define an objective function. Value iteration is then used to solve for optimal policy. ussein Al-Zubaidy, Jerome Talim, Ioannis Lambadaris Optimal Scheduling [8ex] SCE-Carleton Policy Determination University1125 in HSDPA Colonel Networks By Drive, Ottawa, 26 June 2007 ON, Canada 5 / Emai 34
12 Methodology Methodology This work presents a different approach for scheduling in HSDPA. A declarative approach is used, Develop an analytic model for the HSDPA downlink scheduler. A MDP based discrete stochastic dynamic programming model is used to model the system. This Model is a simplifying abstraction of the real scheduler which estimates system behavior under different conditions and describes the role of various system components in these behaviors. It must be solvable. Define an objective function. Value iteration is then used to solve for optimal policy. Study the structure of the optimal policy and develop a near-optimal heuristic policy. ussein Al-Zubaidy, Jerome Talim, Ioannis Lambadaris Optimal Scheduling [8ex] SCE-Carleton Policy Determination University1125 in HSDPA Colonel Networks By Drive, Ottawa, 26 June 2007 ON, Canada 5 / Emai 34
13 Problem Definition and Model Description Problem Definition and Conceptualization Problem Definition and Conceptualization The HSDPA downlink channel uses a mix of TDMA and CDMA: Hussein Al-Zubaidy, Jerome Talim, Ioannis Lambadaris Optimal Scheduling [8ex] SCE-Carleton Policy Determination University1125 in HSDPA Colonel Networks By Drive, Ottawa, 26 June 2007 ON, Canada 6 / Emai 34
14 Problem Definition and Model Description Problem Definition and Conceptualization Problem Definition and Conceptualization The HSDPA downlink channel uses a mix of TDMA and CDMA: Time is slotted into fixed length 2 ms TTIs. ussein Al-Zubaidy, Jerome Talim, Ioannis Lambadaris Optimal Scheduling [8ex] SCE-Carleton Policy Determination University1125 in HSDPA Colonel Networks By Drive, Ottawa, 26 June 2007 ON, Canada 6 / Emai 34
15 Problem Definition and Model Description Problem Definition and Conceptualization Problem Definition and Conceptualization The HSDPA downlink channel uses a mix of TDMA and CDMA: Time is slotted into fixed length 2 ms TTIs. During each TTI, there are 15 available codes that may be allocated to one or more users. Hussein Al-Zubaidy, Jerome Talim, Ioannis Lambadaris Optimal Scheduling [8ex] SCE-Carleton Policy Determination University1125 in HSDPA Colonel Networks By Drive, Ottawa, 26 June 2007 ON, Canada 6 / Emai 34
16 Problem Definition and Model Description HSDPA Scheduler Model (Downlink) Problem Definition and Conceptualization PDUs User 1 UE 1 SDU Scheduler Transceiver RLC RNC RNC User L Channel State Monitor/Predictor Node-B UE L ussein Al-Zubaidy, Jerome Talim, Ioannis Lambadaris Optimal Scheduling [8ex] SCE-Carleton Policy Determination University1125 in HSDPA Colonel Networks By Drive, Ottawa, 26 June 2007 ON, Canada 7 / Emai 34
17 Problem Definition and Model Description Problem Definition and Conceptualization FSMC Model for HSDPA Downlink Channel P 00 P 01 P M-1 P 10 ussein Al-Zubaidy, Jerome Talim, Ioannis Lambadaris Optimal Scheduling [8ex] SCE-Carleton Policy Determination University1125 in HSDPA Colonel Networks By Drive, Ottawa, 26 June 2007 ON, Canada 8 / Emai 34
18 Problem Definition and Model Description Model Description and Basic Assumptions The Model MDP based Model. HSDPA downlink scheduler is modelled by the 5-tuple (T, S, A, P ss (a), R(s, a)), where, T is the set of decision epochs, S and A are the state and action spaces, P ss (a)=pr(s(t + 1)=s s(t)=s, a(s)=a) is the state transition probability, and R(s, a) is the immediate reward when at state s and taking action a. ussein Al-Zubaidy, Jerome Talim, Ioannis Lambadaris Optimal Scheduling [8ex] SCE-Carleton Policy Determination University1125 in HSDPA Colonel Networks By Drive, Ottawa, 26 June 2007 ON, Canada 9 / Emai 34
19 Problem Definition and Model Description Model Description and Basic Assumptions Basic Assumptions L active users in the cell. Finite buffer with size B per user for each of the L users. Error free transmission. SDUs are segmented by RLC into a fixed number of PDUs (u i ) and delivered to Node-B at the beginning of the next TTI. Independent Bernoulli arrivals with parameter q i. Scheduler can assign c codes chunks at a time, where c {1, 3, 5, 15}. ussein Al-Zubaidy, Jerome Talim, Ioannis Lambadaris Optimal Scheduling [8ex] SCE-Carleton Policy Determination University1125 in HSDPA Colonel Networks26 By Drive, Ottawa, June 2007 ON, Canada 10 / Emai 34
20 Problem Definition and Model Description Model Description and Basic Assumptions Basic Assumptions FSMC State Space The channel state of user i during slot t is denoted by γ i (t). Channel state space is the set M = {0, 1,..., M 1}. user i channel can handle up to γ i (t) PDUs per code. ussein Al-Zubaidy, Jerome Talim, Ioannis Lambadaris Optimal Scheduling [8ex] SCE-Carleton Policy Determination University1125 in HSDPA Colonel Networks26 By Drive, Ottawa, June 2007 ON, Canada 11 / Emai 34
21 Problem Definition and Model Description State and Action Sets State and Action Sets The system state s(t) S is a vector and is given by s(t) = (x 1 (t), x 2 (t),..., x L (t), γ 1 (t), γ 2 (t),..., γ L (t)) (1) ussein Al-Zubaidy, Jerome Talim, Ioannis Lambadaris Optimal Scheduling [8ex] SCE-Carleton Policy Determination University1125 in HSDPA Colonel Networks26 By Drive, Ottawa, June 2007 ON, Canada 12 / Emai 34
22 Problem Definition and Model Description State and Action Sets State and Action Sets The system state s(t) S is a vector and is given by s(t) = (x 1 (t), x 2 (t),..., x L (t), γ 1 (t), γ 2 (t),..., γ L (t)) (1) S = {X M} L is finite, due to the assumption of finite buffers size and channel states. ussein Al-Zubaidy, Jerome Talim, Ioannis Lambadaris Optimal Scheduling [8ex] SCE-Carleton Policy Determination University1125 in HSDPA Colonel Networks26 By Drive, Ottawa, June 2007 ON, Canada 12 / Emai 34
23 Problem Definition and Model Description State and Action Sets State and Action Sets The system state s(t) S is a vector and is given by s(t) = (x 1 (t), x 2 (t),..., x L (t), γ 1 (t), γ 2 (t),..., γ L (t)) (1) S = {X M} L is finite, due to the assumption of finite buffers size and channel states. The action a(s) A is taken when in state s a(s) = (a 1 (s), a 2 (s),..., a L (s)) (2) Hussein Al-Zubaidy, Jerome Talim, Ioannis Lambadaris Optimal Scheduling [8ex] SCE-Carleton Policy Determination University1125 in HSDPA Colonel Networks26 By Drive, Ottawa, June 2007 ON, Canada 12 / Emai 34
24 Problem Definition and Model Description State and Action Sets State and Action Sets The system state s(t) S is a vector and is given by s(t) = (x 1 (t), x 2 (t),..., x L (t), γ 1 (t), γ 2 (t),..., γ L (t)) (1) S = {X M} L is finite, due to the assumption of finite buffers size and channel states. The action a(s) A is taken when in state s a(s) = (a 1 (s), a 2 (s),..., a L (s)) (2) subject to, L i=1 a i (s) 15 c, and a xi (t) i(s) γ i (t)c Hussein Al-Zubaidy, Jerome Talim, Ioannis Lambadaris Optimal Scheduling [8ex] SCE-Carleton Policy Determination University1125 in HSDPA Colonel Networks26 By Drive, Ottawa, June 2007 ON, Canada 12 / Emai 34
25 Problem Definition and Model Description State and Action Sets State and Action Sets The system state s(t) S is a vector and is given by s(t) = (x 1 (t), x 2 (t),..., x L (t), γ 1 (t), γ 2 (t),..., γ L (t)) (1) S = {X M} L is finite, due to the assumption of finite buffers size and channel states. The action a(s) A is taken when in state s a(s) = (a 1 (s), a 2 (s),..., a L (s)) (2) subject to, L i=1 a i (s) 15 c, and a xi (t) i(s) γ i (t)c a i (t)c, number of codes allocated to user i at time epoch t. Hussein Al-Zubaidy, Jerome Talim, Ioannis Lambadaris Optimal Scheduling [8ex] SCE-Carleton Policy Determination University1125 in HSDPA Colonel Networks26 By Drive, Ottawa, June 2007 ON, Canada 12 / Emai 34
26 Problem Definition and Model Description Reward Function Reward Function The reward must achieve the objective function R(s, a) have two components corresponding to the two objectives R(s, a) = L L a i γ i c σ (x i x) 1 {xi =B} (3) i=1 i=1 where we defined the fairness factor (σ) to reflect the significance of fairness in the optimal policy. The positive term of the reward maximizes the cell throughput. The second term guarantees some level of fairness and reduces dropping probability. ussein Al-Zubaidy, Jerome Talim, Ioannis Lambadaris Optimal Scheduling [8ex] SCE-Carleton Policy Determination University1125 in HSDPA Colonel Networks26 By Drive, Ottawa, June 2007 ON, Canada 13 / Emai 34
27 Problem Definition and Model Description State Transition Probability State Transition Probability P ss (a) denotes the probability that choosing an action a at time t when in state s will lead to state s at time t + 1. P ss (a) = Pr(s(t + 1)=s s(t)=s, a(t)=a) = Pr(x 1,..., x L,γ 1,...,γ L x 1,..., x L,γ 1,...,γ L,a 1,...,a L ) Hussein Al-Zubaidy, Jerome Talim, Ioannis Lambadaris Optimal Scheduling [8ex] SCE-Carleton Policy Determination University1125 in HSDPA Colonel Networks26 By Drive, Ottawa, June 2007 ON, Canada 14 / Emai 34
28 Problem Definition and Model Description State Transition Probability State Transition Probability P ss (a) denotes the probability that choosing an action a at time t when in state s will lead to state s at time t + 1. P ss (a) = Pr(s(t + 1)=s s(t)=s, a(t)=a) = Pr(x 1,..., x L,γ 1,...,γ L x 1,..., x L,γ 1,...,γ L,a 1,...,a L ) The evolution of the queue size (x i ) is given by x i = min ( [x i y i ] + + z i, B ) = min ( [x i a i γ i c] + + z i, B ) (4) Hussein Al-Zubaidy, Jerome Talim, Ioannis Lambadaris Optimal Scheduling [8ex] SCE-Carleton Policy Determination University1125 in HSDPA Colonel Networks26 By Drive, Ottawa, June 2007 ON, Canada 14 / Emai 34
29 Problem Definition and Model Description State Transition Probability State Transition Probability P ss (a) denotes the probability that choosing an action a at time t when in state s will lead to state s at time t + 1. P ss (a) = Pr(s(t + 1)=s s(t)=s, a(t)=a) = Pr(x 1,..., x L,γ 1,...,γ L x 1,..., x L,γ 1,...,γ L,a 1,...,a L ) The evolution of the queue size (x i ) is given by x i = min ( [x i y i ] + + z i, B ) = min ( [x i a i γ i c] + + z i, B ) (4) Using the independence of the channel state and queue sizes L ( ) P ss (a) = P xi x i (γ i, a i ) P γi γ i where P γi γ i i =1 is the Markov transition probability of the FSMC. (5) ussein Al-Zubaidy, Jerome Talim, Ioannis Lambadaris Optimal Scheduling [8ex] SCE-Carleton Policy Determination University1125 in HSDPA Colonel Networks26 By Drive, Ottawa, June 2007 ON, Canada 14 / Emai 34
30 Problem Definition and Model Description State Transition Probability cont. State Transition Probability where 1 if x i =x i =B & a i γ i = 0, q i if x i =x i =B & 0 < a i γ i c u i, P xi x i (γ q i if x i =B & x i < B & W 1 B, i, a i )= q i if x i <B & x i = W 1, 1 q i if x i <B & x i = W 2, 0 otherwise. W 1 = [x i a i γ i c] + + u i W 2 = [x i a i γ i c] + (6) ussein Al-Zubaidy, Jerome Talim, Ioannis Lambadaris Optimal Scheduling [8ex] SCE-Carleton Policy Determination University1125 in HSDPA Colonel Networks26 By Drive, Ottawa, June 2007 ON, Canada 15 / Emai 34
31 Value Function Problem Definition and Model Description Value Function Infinite-horizon MDP. Hussein Al-Zubaidy, Jerome Talim, Ioannis Lambadaris Optimal Scheduling [8ex] SCE-Carleton Policy Determination University1125 in HSDPA Colonel Networks26 By Drive, Ottawa, June 2007 ON, Canada 16 / Emai 34
32 Problem Definition and Model Description Value Function Value Function Infinite-horizon MDP. Total expected discounted reward optimality criterion with discount factor λ is used, where 0 < λ < 1. Hussein Al-Zubaidy, Jerome Talim, Ioannis Lambadaris Optimal Scheduling [8ex] SCE-Carleton Policy Determination University1125 in HSDPA Colonel Networks26 By Drive, Ottawa, June 2007 ON, Canada 16 / Emai 34
33 Problem Definition and Model Description Value Function Value Function Infinite-horizon MDP. Total expected discounted reward optimality criterion with discount factor λ is used, where 0 < λ < 1. The objective is to find the policy π among all policies, that maximize the value function V π (s). Hussein Al-Zubaidy, Jerome Talim, Ioannis Lambadaris Optimal Scheduling [8ex] SCE-Carleton Policy Determination University1125 in HSDPA Colonel Networks26 By Drive, Ottawa, June 2007 ON, Canada 16 / Emai 34
34 Problem Definition and Model Description Value Function Value Function Infinite-horizon MDP. Total expected discounted reward optimality criterion with discount factor λ is used, where 0 < λ < 1. The objective is to find the policy π among all policies, that maximize the value function V π (s). The optimal policy is characterized by V (s) = max [R(s, a) + λ P ss (a)v (s )] (7) a A s S where, V (s) = sup π V π (s), attained when applying the optimal policy π. ussein Al-Zubaidy, Jerome Talim, Ioannis Lambadaris Optimal Scheduling [8ex] SCE-Carleton Policy Determination University1125 in HSDPA Colonel Networks26 By Drive, Ottawa, June 2007 ON, Canada 16 / Emai 34
35 Problem Definition and Model Description Value Function Value Function Infinite-horizon MDP. Total expected discounted reward optimality criterion with discount factor λ is used, where 0 < λ < 1. The objective is to find the policy π among all policies, that maximize the value function V π (s). The optimal policy is characterized by V (s) = max [R(s, a) + λ P ss (a)v (s )] (7) a A s S where, V (s) = sup π V π (s), attained when applying the optimal policy π. The model was solved numerically using Value Iteration. ussein Al-Zubaidy, Jerome Talim, Ioannis Lambadaris Optimal Scheduling [8ex] SCE-Carleton Policy Determination University1125 in HSDPA Colonel Networks26 By Drive, Ottawa, June 2007 ON, Canada 16 / Emai 34
36 Case Study and Results Case Study: Two Users with 2-State FSMC The Optimal Policy for Two Symmetrical Users 0 0,0 1,0 2, x 1 0,1 1,1 2,1 3,0 0,2 10 1, ,3 25 x2 P(γ i =1)=0.5 and P(z i =5)=0.5 for all i {1, 2}; c = 5. ussein Al-Zubaidy, Jerome Talim, Ioannis Lambadaris Optimal Scheduling [8ex] SCE-Carleton Policy Determination University1125 in HSDPA Colonel Networks26 By Drive, Ottawa, June 2007 ON, Canada 17 / Emai 34
37 Case Study and Results Case Study: Two Users with 2-State FSMC The Effect of Channel Quality on Policy Structure 0 0,0 1,0 2, x 1 0,1 1,1 2,1 3,0 0,2 10 1,2 15 0, x2 P(γ 1 =1)=0.8, P(γ 2 =1)=0.5 and P(z i =5)=0.5. ussein Al-Zubaidy, Jerome Talim, Ioannis Lambadaris Optimal Scheduling [8ex] SCE-Carleton Policy Determination University1125 in HSDPA Colonel Networks26 By Drive, Ottawa, June 2007 ON, Canada 18 / Emai 34
38 Case Study and Results Case Study: Two Users with 2-State FSMC The Effect of Arrival Probability on Policy Structure x 1 0 0,0 1,0 2,0 1 0,1 1,1 5 0,2 2,1 3,0 10 1, ,3 25 x2 P(γ 1 = 1) = P(γ 2 = 1) = 0.5 and P(z 1 =5)=0.8 P(z 2 =5)=0.5. ussein Al-Zubaidy, Jerome Talim, Ioannis Lambadaris Optimal Scheduling [8ex] SCE-Carleton Policy Determination University1125 in HSDPA Colonel Networks26 By Drive, Ottawa, June 2007 ON, Canada 19 / Emai 34
39 Case Study and Results Case Study: Two Users with 2-State FSMC Heuristic Policy We studied the optimal policy structure by running a wide range of scenarios, we noticed the following trends The policy is a switch-over. Hussein Al-Zubaidy, Jerome Talim, Ioannis Lambadaris Optimal Scheduling [8ex] SCE-Carleton Policy Determination University1125 in HSDPA Colonel Networks26 By Drive, Ottawa, June 2007 ON, Canada 20 / Emai 34
40 Case Study and Results Case Study: Two Users with 2-State FSMC Heuristic Policy We studied the optimal policy structure by running a wide range of scenarios, we noticed the following trends The policy is a switch-over. The weight (w i ) is a function of the difference of the two channel qualities and that of the arrival probabilities: w 1 = f ([ P γ ] +, [ P z ] + ) (8) w 2 = f ([ P γ ] +, [ P z ] + ) (9) where P γ =P(γ 1 =1) P(γ 2 =1) and P z =P(z 1 =u) P(z 2 =u). Hussein Al-Zubaidy, Jerome Talim, Ioannis Lambadaris Optimal Scheduling [8ex] SCE-Carleton Policy Determination University1125 in HSDPA Colonel Networks26 By Drive, Ottawa, June 2007 ON, Canada 20 / Emai 34
41 Case Study and Results Case Study: Two Users with 2-State FSMC Heuristic Policy We studied the optimal policy structure by running a wide range of scenarios, we noticed the following trends The policy is a switch-over. The weight (w i ) is a function of the difference of the two channel qualities and that of the arrival probabilities: w 1 = f ([ P γ ] +, [ P z ] + ) (8) w 2 = f ([ P γ ] +, [ P z ] + ) (9) where P γ =P(γ 1 =1) P(γ 2 =1) and P z =P(z 1 =u) P(z 2 =u). The intermediate regions has almost a constant width that equals 2c. ussein Al-Zubaidy, Jerome Talim, Ioannis Lambadaris Optimal Scheduling [8ex] SCE-Carleton Policy Determination University1125 in HSDPA Colonel Networks26 By Drive, Ottawa, June 2007 ON, Canada 20 / Emai 34
42 Case Study and Results Case Study: Two Users with 2-State FSMC Heuristic Policy We studied the optimal policy structure by running a wide range of scenarios, we noticed the following trends The policy is a switch-over. The weight (w i ) is a function of the difference of the two channel qualities and that of the arrival probabilities: w 1 = f ([ P γ ] +, [ P z ] + ) (8) w 2 = f ([ P γ ] +, [ P z ] + ) (9) where P γ =P(γ 1 =1) P(γ 2 =1) and P z =P(z 1 =u) P(z 2 =u). The intermediate regions has almost a constant width that equals 2c. a 1 (respectively a 2 ) is increasing in x 1 (respectively x 2 ). ussein Al-Zubaidy, Jerome Talim, Ioannis Lambadaris Optimal Scheduling [8ex] SCE-Carleton Policy Determination University1125 in HSDPA Colonel Networks26 By Drive, Ottawa, June 2007 ON, Canada 20 / Emai 34
43 Case Study and Results Case Study: Two Users with 2-State FSMC Heuristic Policy We studied the optimal policy structure by running a wide range of scenarios, we noticed the following trends The policy is a switch-over. The weight (w i ) is a function of the difference of the two channel qualities and that of the arrival probabilities: w 1 = f ([ P γ ] +, [ P z ] + ) (8) w 2 = f ([ P γ ] +, [ P z ] + ) (9) where P γ =P(γ 1 =1) P(γ 2 =1) and P z =P(z 1 =u) P(z 2 =u). The intermediate regions has almost a constant width that equals 2c. a 1 (respectively a 2 ) is increasing in x 1 (respectively x 2 ). f () is increasing in P γ and decreasing in P z. ussein Al-Zubaidy, Jerome Talim, Ioannis Lambadaris Optimal Scheduling [8ex] SCE-Carleton Policy Determination University1125 in HSDPA Colonel Networks26 By Drive, Ottawa, June 2007 ON, Canada 20 / Emai 34
44 Case Study and Results Weight Function Approximation Weight Function Approximation Following these observations, we approximated w 1 and w 2 as follows ŵ 1 = [ P γ ] + 0.7[ P z ] + (10) ŵ 2 = [ P γ ] + 0.7[ P z ] + (11) ussein Al-Zubaidy, Jerome Talim, Ioannis Lambadaris Optimal Scheduling [8ex] SCE-Carleton Policy Determination University1125 in HSDPA Colonel Networks26 By Drive, Ottawa, June 2007 ON, Canada 21 / Emai 34
45 Case Study and Results Heuristic Policy Structure Heuristic (dotted line) vs. optimal policy; c = x 1 0 0, , ,1 25 x2 P(γ i =1)=0.5 and P(z i =5)=0.5 for all i {1, 2}. ussein Al-Zubaidy, Jerome Talim, Ioannis Lambadaris Optimal Scheduling [8ex] SCE-Carleton Policy Determination University1125 in HSDPA Colonel Networks26 By Drive, Ottawa, June 2007 ON, Canada 22 / Emai 34
46 Case Study and Results Heuristic Policy Structure Heuristic (dotted line) vs. optimal policy; c = x ,0 1, , x2 P(γ 1 =1)=0.8, P(γ 2 =1)=0.5 and P(z i =5)=0.5. ussein Al-Zubaidy, Jerome Talim, Ioannis Lambadaris Optimal Scheduling [8ex] SCE-Carleton Policy Determination University1125 in HSDPA Colonel Networks26 By Drive, Ottawa, June 2007 ON, Canada 23 / Emai 34
47 Case Study and Results Heuristic Policy Structure Heuristic (dotted line) vs. optimal policy; c = x ,0 1, , x2 P(γ 1 = 1) = P(γ 2 = 1) = 0.5, P(z 1 =5)=0.8 and P(z 2 =5)=0.5. ussein Al-Zubaidy, Jerome Talim, Ioannis Lambadaris Optimal Scheduling [8ex] SCE-Carleton Policy Determination University1125 in HSDPA Colonel Networks26 By Drive, Ottawa, June 2007 ON, Canada 24 / Emai 34
48 Case Study and Results Heuristic Policy Structure Heuristic (dotted line) vs. optimal policy; c = x 1 0 0,0 1,0 2, ,1 1,1 3,0 0,2 10 2,1 15 1,2 20 0,3 25 x2 P(γ i =1)=0.5 and P(z i =5)=0.5 for all i {1, 2}. ussein Al-Zubaidy, Jerome Talim, Ioannis Lambadaris Optimal Scheduling [8ex] SCE-Carleton Policy Determination University1125 in HSDPA Colonel Networks26 By Drive, Ottawa, June 2007 ON, Canada 25 / Emai 34
49 Case Study and Results Heuristic Policy Structure Heuristic (dotted line) vs. optimal policy; c = x 1 0 0,0 1,0 2,0 1 0,1 1,1 3,0 5 0,2 2, ,2 20 0,3 25 x2 P(γ 1 =1)=0.8, P(γ 2 =1)=0.5 and P(z i =5)=0.5. ussein Al-Zubaidy, Jerome Talim, Ioannis Lambadaris Optimal Scheduling [8ex] SCE-Carleton Policy Determination University1125 in HSDPA Colonel Networks26 By Drive, Ottawa, June 2007 ON, Canada 26 / Emai 34
50 Case Study and Results Heuristic Policy Structure Heuristic (dotted line) vs. optimal policy; c = x 1 0 0,0 1,0 2,0 1 0,1 5 0,2 1,1 2,1 3,0 10 1, ,3 25 x2 P(γ 1 = 1) = P(γ 2 = 1) = 0.5, P(z 1 =5)=0.8 and P(z 2 =5)=0.5. ussein Al-Zubaidy, Jerome Talim, Ioannis Lambadaris Optimal Scheduling [8ex] SCE-Carleton Policy Determination University1125 in HSDPA Colonel Networks26 By Drive, Ottawa, June 2007 ON, Canada 27 / Emai 34
51 Case Study and Results Performance Evaluation Performance Evaluation: The Effect of Policy Granularity Two actions (c=15) Four actions (c=5) Six actions (c=3) Average Queue Length (PDUs) User1 (c=15) User2 (c=15) User1 (c=5) User2 (c=5) User1 (c=3) User2 (c=3) Average Drop Probability Offered Load (Rho) (a) on Average Queue Length Offered Load (Rho) (b) on Average Drop Probability Where ρ = i P z i u i /r π is the offered load and r π is the measured system capacity under π. P(γ 1 =1)=0.8 and P(γ 2 =1)=0.5. ussein Al-Zubaidy, Jerome Talim, Ioannis Lambadaris Optimal Scheduling [8ex] SCE-Carleton Policy Determination University1125 in HSDPA Colonel Networks26 By Drive, Ottawa, June 2007 ON, Canada 28 / Emai 34
52 Case Study and Results Heuristic Policy Evaluation Performance Evaluation Throughput (PDUs/msec) alpha=0.63, beta=0.12 Rho = 1.2 Rho = 0.8 Rho = 0.5 Round Robin Heuristic MDP Time slots (c) System Throughput for different ρ; P(γ 1 =1)=0.8 and P(γ 2 =1)=0.5. Delay (msec) alpha=0.63, beta=0.12, Pz1=0.8, Pz2=0.5, u=10 Round Robin User1 User2 Heuristic User1 User2 2 Optimal (MDP) User1 (c=5) User2 (c=5) Time slots (d) Queueing Delay Performance; P(γ 2 = 1) = 0.5, q 1 = 0.8, q 2 = 0.5 and u = 10. ussein Al-Zubaidy, Jerome Talim, Ioannis Lambadaris Optimal Scheduling [8ex] SCE-Carleton Policy Determination University1125 in HSDPA Colonel Networks26 By Drive, Ottawa, June 2007 ON, Canada 29 / Emai 34
53 Conclusion Conclusion and Future Work Conclusion The optimal policy can be described as share the codes in proportion to the weighted queue length of the connected users. ussein Al-Zubaidy, Jerome Talim, Ioannis Lambadaris Optimal Scheduling [8ex] SCE-Carleton Policy Determination University1125 in HSDPA Colonel Networks26 By Drive, Ottawa, June 2007 ON, Canada 30 / Emai 34
54 Conclusion and Future Work Conclusion Conclusion The optimal policy can be described as share the codes in proportion to the weighted queue length of the connected users. A policy with finer granularity will perform better in light to moderate loading conditions, while a coarse policy is more desirable in heavy loading conditions. ussein Al-Zubaidy, Jerome Talim, Ioannis Lambadaris Optimal Scheduling [8ex] SCE-Carleton Policy Determination University1125 in HSDPA Colonel Networks26 By Drive, Ottawa, June 2007 ON, Canada 30 / Emai 34
55 Conclusion and Future Work Conclusion Conclusion The optimal policy can be described as share the codes in proportion to the weighted queue length of the connected users. A policy with finer granularity will perform better in light to moderate loading conditions, while a coarse policy is more desirable in heavy loading conditions. However, the performance gain when using c < 5 is marginal and does not justify the added complexity. Hussein Al-Zubaidy, Jerome Talim, Ioannis Lambadaris Optimal Scheduling [8ex] SCE-Carleton Policy Determination University1125 in HSDPA Colonel Networks26 By Drive, Ottawa, June 2007 ON, Canada 30 / Emai 34
56 Conclusion cont. Conclusion and Future Work Conclusion The suggested heuristic policy has a reduced constant time complexity (O(1)) as compared to the exponential time complexity needed in the determination of the optimal policy. ussein Al-Zubaidy, Jerome Talim, Ioannis Lambadaris Optimal Scheduling [8ex] SCE-Carleton Policy Determination University1125 in HSDPA Colonel Networks26 By Drive, Ottawa, June 2007 ON, Canada 31 / Emai 34
57 Conclusion and Future Work Conclusion Conclusion cont. The suggested heuristic policy has a reduced constant time complexity (O(1)) as compared to the exponential time complexity needed in the determination of the optimal policy. The performance of the resulted heuristic policy matches very closely to the optimal policy. ussein Al-Zubaidy, Jerome Talim, Ioannis Lambadaris Optimal Scheduling [8ex] SCE-Carleton Policy Determination University1125 in HSDPA Colonel Networks26 By Drive, Ottawa, June 2007 ON, Canada 31 / Emai 34
58 Conclusion and Future Work Conclusion Conclusion cont. The suggested heuristic policy has a reduced constant time complexity (O(1)) as compared to the exponential time complexity needed in the determination of the optimal policy. The performance of the resulted heuristic policy matches very closely to the optimal policy. The results also proved that RR is undesirable in HSDPA system due to the poor performance and lack of fairness. ussein Al-Zubaidy, Jerome Talim, Ioannis Lambadaris Optimal Scheduling [8ex] SCE-Carleton Policy Determination University1125 in HSDPA Colonel Networks26 By Drive, Ottawa, June 2007 ON, Canada 31 / Emai 34
59 Conclusion and Future Work Conclusion Conclusion cont. The suggested heuristic policy has a reduced constant time complexity (O(1)) as compared to the exponential time complexity needed in the determination of the optimal policy. The performance of the resulted heuristic policy matches very closely to the optimal policy. The results also proved that RR is undesirable in HSDPA system due to the poor performance and lack of fairness. The suggested heuristic policy can be extended to the case with more than two active users. It also can be easily adapted to accommodate more than one class of service. ussein Al-Zubaidy, Jerome Talim, Ioannis Lambadaris Optimal Scheduling [8ex] SCE-Carleton Policy Determination University1125 in HSDPA Colonel Networks26 By Drive, Ottawa, June 2007 ON, Canada 31 / Emai 34
60 Future Work Conclusion and Future Work Future Work Prove analytically some of the optimal policy and value function characteristics, such as monotonicity, multi-modularity, and the switch-over behavior that we noticed before. ussein Al-Zubaidy, Jerome Talim, Ioannis Lambadaris Optimal Scheduling [8ex] SCE-Carleton Policy Determination University1125 in HSDPA Colonel Networks26 By Drive, Ottawa, June 2007 ON, Canada 32 / Emai 34
61 Conclusion and Future Work Future Work Future Work Prove analytically some of the optimal policy and value function characteristics, such as monotonicity, multi-modularity, and the switch-over behavior that we noticed before. Relax the assumption of error free transmission and extend the model to take into account retransmissions. ussein Al-Zubaidy, Jerome Talim, Ioannis Lambadaris Optimal Scheduling [8ex] SCE-Carleton Policy Determination University1125 in HSDPA Colonel Networks26 By Drive, Ottawa, June 2007 ON, Canada 32 / Emai 34
62 Conclusion and Future Work Future Work Future Work Prove analytically some of the optimal policy and value function characteristics, such as monotonicity, multi-modularity, and the switch-over behavior that we noticed before. Relax the assumption of error free transmission and extend the model to take into account retransmissions. Study the effect of using different arrival process statistics using simulation obviously. ussein Al-Zubaidy, Jerome Talim, Ioannis Lambadaris Optimal Scheduling [8ex] SCE-Carleton Policy Determination University1125 in HSDPA Colonel Networks26 By Drive, Ottawa, June 2007 ON, Canada 32 / Emai 34
63 Conclusion and Future Work Future Work Thank You Discussion Hussein Zubaidy hussein/ iscussion [8ex] Hussein Zubaidy [4ex] hussein/ Thank You () 33 / 34
64 Conclusion and Future Work Acronyms Acronyms HSDPA High Speed Downlink Packet Access. 3GPP Third Generation Partnership Project MDP Markov Decision Process TDMA Time Division Multiple Access CDMA Code Division Multiple Access TTI Transmission Time Interval (2 ms) FSMC Finite State Markov Channel SDU Service Data Unit RLC Radio Link Control Protocol located at Radio Network Controller (RNC) PDU Protocol data unit LQF Longest Queue First iscussion [8ex] Hussein Zubaidy [4ex] hussein/ Thank You () 34 / 34
Intelligent Systems (AI-2)
Intelligent Systems (AI-2) Computer Science cpsc422, Lecture 9 Sep, 28, 2016 Slide 1 CPSC 422, Lecture 9 An MDP Approach to Multi-Category Patient Scheduling in a Diagnostic Facility Adapted from: Matthew
More informationOptimal Policies for Distributed Data Aggregation in Wireless Sensor Networks
Optimal Policies for Distributed Data Aggregation in Wireless Sensor Networks Hussein Abouzeid Department of Electrical Computer and Systems Engineering Rensselaer Polytechnic Institute abouzeid@ecse.rpi.edu
More informationLecture 7: Bayesian approach to MAB - Gittins index
Advanced Topics in Machine Learning and Algorithmic Game Theory Lecture 7: Bayesian approach to MAB - Gittins index Lecturer: Yishay Mansour Scribe: Mariano Schain 7.1 Introduction In the Bayesian approach
More informationMaking Complex Decisions
Ch. 17 p.1/29 Making Complex Decisions Chapter 17 Ch. 17 p.2/29 Outline Sequential decision problems Value iteration algorithm Policy iteration algorithm Ch. 17 p.3/29 A simple environment 3 +1 p=0.8 2
More informationModelling Anti-Terrorist Surveillance Systems from a Queueing Perspective
Systems from a Queueing Perspective September 7, 2012 Problem A surveillance resource must observe several areas, searching for potential adversaries. Problem A surveillance resource must observe several
More informationMaking Decisions. CS 3793 Artificial Intelligence Making Decisions 1
Making Decisions CS 3793 Artificial Intelligence Making Decisions 1 Planning under uncertainty should address: The world is nondeterministic. Actions are not certain to succeed. Many events are outside
More informationFinal exam solutions
EE365 Stochastic Control / MS&E251 Stochastic Decision Models Profs. S. Lall, S. Boyd June 5 6 or June 6 7, 2013 Final exam solutions This is a 24 hour take-home final. Please turn it in to one of the
More informationThe Irrevocable Multi-Armed Bandit Problem
The Irrevocable Multi-Armed Bandit Problem Ritesh Madan Qualcomm-Flarion Technologies May 27, 2009 Joint work with Vivek Farias (MIT) 2 Multi-Armed Bandit Problem n arms, where each arm i is a Markov Decision
More informationDeep RL and Controls Homework 1 Spring 2017
10-703 Deep RL and Controls Homework 1 Spring 2017 February 1, 2017 Due February 17, 2017 Instructions You have 15 days from the release of the assignment until it is due. Refer to gradescope for the exact
More informationMarkov Decision Processes
Markov Decision Processes Robert Platt Northeastern University Some images and slides are used from: 1. CS188 UC Berkeley 2. AIMA 3. Chris Amato Stochastic domains So far, we have studied search Can use
More informationReinforcement Learning (1): Discrete MDP, Value Iteration, Policy Iteration
Reinforcement Learning (1): Discrete MDP, Value Iteration, Policy Iteration Piyush Rai CS5350/6350: Machine Learning November 29, 2011 Reinforcement Learning Supervised Learning: Uses explicit supervision
More informationDecision Theory: Value Iteration
Decision Theory: Value Iteration CPSC 322 Decision Theory 4 Textbook 9.5 Decision Theory: Value Iteration CPSC 322 Decision Theory 4, Slide 1 Lecture Overview 1 Recap 2 Policies 3 Value Iteration Decision
More informationThe Value of Information in Central-Place Foraging. Research Report
The Value of Information in Central-Place Foraging. Research Report E. J. Collins A. I. Houston J. M. McNamara 22 February 2006 Abstract We consider a central place forager with two qualitatively different
More informationDynamic pricing and scheduling in a multi-class single-server queueing system
DOI 10.1007/s11134-011-9214-5 Dynamic pricing and scheduling in a multi-class single-server queueing system Eren Başar Çil Fikri Karaesmen E. Lerzan Örmeci Received: 3 April 2009 / Revised: 21 January
More informationReinforcement Learning (1): Discrete MDP, Value Iteration, Policy Iteration
Reinforcement Learning (1): Discrete MDP, Value Iteration, Policy Iteration Piyush Rai CS5350/6350: Machine Learning November 29, 2011 Reinforcement Learning Supervised Learning: Uses explicit supervision
More informationEconomics 2010c: Lecture 4 Precautionary Savings and Liquidity Constraints
Economics 2010c: Lecture 4 Precautionary Savings and Liquidity Constraints David Laibson 9/11/2014 Outline: 1. Precautionary savings motives 2. Liquidity constraints 3. Application: Numerical solution
More informationEssays on Some Combinatorial Optimization Problems with Interval Data
Essays on Some Combinatorial Optimization Problems with Interval Data a thesis submitted to the department of industrial engineering and the institute of engineering and sciences of bilkent university
More informationMartingale Pricing Theory in Discrete-Time and Discrete-Space Models
IEOR E4707: Foundations of Financial Engineering c 206 by Martin Haugh Martingale Pricing Theory in Discrete-Time and Discrete-Space Models These notes develop the theory of martingale pricing in a discrete-time,
More informationLecture 2: Making Good Sequences of Decisions Given a Model of World. CS234: RL Emma Brunskill Winter 2018
Lecture 2: Making Good Sequences of Decisions Given a Model of World CS234: RL Emma Brunskill Winter 218 Human in the loop exoskeleton work from Steve Collins lab Class Structure Last Time: Introduction
More information4 Reinforcement Learning Basic Algorithms
Learning in Complex Systems Spring 2011 Lecture Notes Nahum Shimkin 4 Reinforcement Learning Basic Algorithms 4.1 Introduction RL methods essentially deal with the solution of (optimal) control problems
More informationAM 121: Intro to Optimization Models and Methods
AM 121: Intro to Optimization Models and Methods Lecture 18: Markov Decision Processes Yiling Chen and David Parkes Lesson Plan Markov decision processes Policies and Value functions Solving: average reward,
More informationThe Agent-Environment Interface Goals, Rewards, Returns The Markov Property The Markov Decision Process Value Functions Optimal Value Functions
The Agent-Environment Interface Goals, Rewards, Returns The Markov Property The Markov Decision Process Value Functions Optimal Value Functions Optimality and Approximation Finite MDP: {S, A, R, p, γ}
More informationIEOR E4703: Monte-Carlo Simulation
IEOR E4703: Monte-Carlo Simulation Simulating Stochastic Differential Equations Martin Haugh Department of Industrial Engineering and Operations Research Columbia University Email: martin.b.haugh@gmail.com
More informationMarkov Decision Processes: Making Decision in the Presence of Uncertainty. (some of) R&N R&N
Markov Decision Processes: Making Decision in the Presence of Uncertainty (some of) R&N 16.1-16.6 R&N 17.1-17.4 Different Aspects of Machine Learning Supervised learning Classification - concept learning
More informationarxiv: v1 [math.pr] 6 Apr 2015
Analysis of the Optimal Resource Allocation for a Tandem Queueing System arxiv:1504.01248v1 [math.pr] 6 Apr 2015 Liu Zaiming, Chen Gang, Wu Jinbiao School of Mathematics and Statistics, Central South University,
More informationMulti-TSO system reliability: cross-border balancing
Multi-TSO system reliability: cross-border balancing Fridrik Mar Baldursson, Reykjavik University Ewa Lazarczyk, Reykjavik University Marten Ovaere, KU Leuven Stef Proost, KU Leuven, April 5, 2016 Background
More informationPOMDPs: Partially Observable Markov Decision Processes Advanced AI
POMDPs: Partially Observable Markov Decision Processes Advanced AI Wolfram Burgard Types of Planning Problems Classical Planning State observable Action Model Deterministic, accurate MDPs observable stochastic
More informationGame Theory for Wireless Engineers Chapter 3, 4
Game Theory for Wireless Engineers Chapter 3, 4 Zhongliang Liang ECE@Mcmaster Univ October 8, 2009 Outline Chapter 3 - Strategic Form Games - 3.1 Definition of A Strategic Form Game - 3.2 Dominated Strategies
More informationFinancial Giffen Goods: Examples and Counterexamples
Financial Giffen Goods: Examples and Counterexamples RolfPoulsen and Kourosh Marjani Rasmussen Abstract In the basic Markowitz and Merton models, a stock s weight in efficient portfolios goes up if its
More informationEE266 Homework 5 Solutions
EE, Spring 15-1 Professor S. Lall EE Homework 5 Solutions 1. A refined inventory model. In this problem we consider an inventory model that is more refined than the one you ve seen in the lectures. The
More informationNon-Deterministic Search
Non-Deterministic Search MDP s 1 Non-Deterministic Search How do you plan (search) when your actions might fail? In general case, how do you plan, when the actions have multiple possible outcomes? 2 Example:
More informationReasoning with Uncertainty
Reasoning with Uncertainty Markov Decision Models Manfred Huber 2015 1 Markov Decision Process Models Markov models represent the behavior of a random process, including its internal state and the externally
More informationLog-Robust Portfolio Management
Log-Robust Portfolio Management Dr. Aurélie Thiele Lehigh University Joint work with Elcin Cetinkaya and Ban Kawas Research partially supported by the National Science Foundation Grant CMMI-0757983 Dr.
More informationTDT4171 Artificial Intelligence Methods
TDT47 Artificial Intelligence Methods Lecture 7 Making Complex Decisions Norwegian University of Science and Technology Helge Langseth IT-VEST 0 helgel@idi.ntnu.no TDT47 Artificial Intelligence Methods
More informationOptimal routing and placement of orders in limit order markets
Optimal routing and placement of orders in limit order markets Rama CONT Arseniy KUKANOV Imperial College London Columbia University New York CFEM-GARP Joint Event and Seminar 05/01/13, New York Choices,
More informationA Markovian Futures Market for Computing Power
Fernando Martinez Peter Harrison Uli Harder A distributed economic solution: MaGoG A world peer-to-peer market No central auctioneer Messages are forwarded by neighbours, and a copy remains in their pubs
More informationHandout 4: Deterministic Systems and the Shortest Path Problem
SEEM 3470: Dynamic Optimization and Applications 2013 14 Second Term Handout 4: Deterministic Systems and the Shortest Path Problem Instructor: Shiqian Ma January 27, 2014 Suggested Reading: Bertsekas
More informationAN ABSTRACT OF THE THESIS OF
AN ABSTRACT OF THE THESIS OF Pingan Zhu for the degree of Master of Science incomputer Science and Mathematics presented on September 23, 2013. Title: Revenue-Based Spectrum Management Via Markov Decision
More informationOptimal Securitization via Impulse Control
Optimal Securitization via Impulse Control Rüdiger Frey (joint work with Roland C. Seydel) Mathematisches Institut Universität Leipzig and MPI MIS Leipzig Bachelier Finance Society, June 21 (1) Optimal
More informationCall Admission Control for Preemptive and Partially Blocking Service Integration Schemes in ATM Networks
Call Admission Control for Preemptive and Partially Blocking Service Integration Schemes in ATM Networks Ernst Nordström Department of Computer Systems, Information Technology, Uppsala University, Box
More informationReduced Complexity Approaches to Asymmetric Information Games
Reduced Complexity Approaches to Asymmetric Information Games Jeff Shamma and Lichun Li Georgia Institution of Technology ARO MURI Annual Review November 19, 2014 Research Thrust: Obtaining Actionable
More information17 MAKING COMPLEX DECISIONS
267 17 MAKING COMPLEX DECISIONS The agent s utility now depends on a sequence of decisions In the following 4 3grid environment the agent makes a decision to move (U, R, D, L) at each time step When the
More informationCPS 270: Artificial Intelligence Markov decision processes, POMDPs
CPS 270: Artificial Intelligence http://www.cs.duke.edu/courses/fall08/cps270/ Markov decision processes, POMDPs Instructor: Vincent Conitzer Warmup: a Markov process with rewards We derive some reward
More informationAn optimal policy for joint dynamic price and lead-time quotation
Lingnan University From the SelectedWorks of Prof. LIU Liming November, 2011 An optimal policy for joint dynamic price and lead-time quotation Jiejian FENG Liming LIU, Lingnan University, Hong Kong Xianming
More informationDependence Structure and Extreme Comovements in International Equity and Bond Markets
Dependence Structure and Extreme Comovements in International Equity and Bond Markets René Garcia Edhec Business School, Université de Montréal, CIRANO and CIREQ Georges Tsafack Suffolk University Measuring
More informationCS 188: Artificial Intelligence Spring Announcements
CS 188: Artificial Intelligence Spring 2011 Lecture 9: MDPs 2/16/2011 Pieter Abbeel UC Berkeley Many slides over the course adapted from either Dan Klein, Stuart Russell or Andrew Moore 1 Announcements
More informationDynamically Scheduling and Maintaining a Flexible Server
Dynamically Scheduling and Maintaining a Flexible Server Jefferson Huang Operations Research Department Naval Postgraduate School INFORMS Annual Meeting November 7, 2018 Co-Authors: Douglas Down (McMaster),
More informationCSEP 573: Artificial Intelligence
CSEP 573: Artificial Intelligence Markov Decision Processes (MDP)! Ali Farhadi Many slides over the course adapted from Luke Zettlemoyer, Dan Klein, Pieter Abbeel, Stuart Russell or Andrew Moore 1 Outline
More informationCS 188: Artificial Intelligence. Outline
C 188: Artificial Intelligence Markov Decision Processes (MDPs) Pieter Abbeel UC Berkeley ome slides adapted from Dan Klein 1 Outline Markov Decision Processes (MDPs) Formalism Value iteration In essence
More informationCS 188: Artificial Intelligence
CS 188: Artificial Intelligence Markov Decision Processes Dan Klein, Pieter Abbeel University of California, Berkeley Non Deterministic Search Example: Grid World A maze like problem The agent lives in
More informationLecture 12: MDP1. Victor R. Lesser. CMPSCI 683 Fall 2010
Lecture 12: MDP1 Victor R. Lesser CMPSCI 683 Fall 2010 Biased Random GSAT - WalkSat Notice no random restart 2 Today s lecture Search where there is Uncertainty in Operator Outcome --Sequential Decision
More informationReinforcement Learning
Reinforcement Learning MDP March May, 2013 MDP MDP: S, A, P, R, γ, µ State can be partially observable: Partially Observable MDPs () Actions can be temporally extended: Semi MDPs (SMDPs) and Hierarchical
More informationHaiyang Feng College of Management and Economics, Tianjin University, Tianjin , CHINA
RESEARCH ARTICLE QUALITY, PRICING, AND RELEASE TIME: OPTIMAL MARKET ENTRY STRATEGY FOR SOFTWARE-AS-A-SERVICE VENDORS Haiyang Feng College of Management and Economics, Tianjin University, Tianjin 300072,
More informationPakes (1986): Patents as Options: Some Estimates of the Value of Holding European Patent Stocks
Pakes (1986): Patents as Options: Some Estimates of the Value of Holding European Patent Stocks Spring 2009 Main question: How much are patents worth? Answering this question is important, because it helps
More informationBudget Management In GSP (2018)
Budget Management In GSP (2018) Yahoo! March 18, 2018 Miguel March 18, 2018 1 / 26 Today s Presentation: Budget Management Strategies in Repeated auctions, Balseiro, Kim, and Mahdian, WWW2017 Learning
More informationReport for technical cooperation between Georgia Institute of Technology and ONS - Operador Nacional do Sistema Elétrico Risk Averse Approach
Report for technical cooperation between Georgia Institute of Technology and ONS - Operador Nacional do Sistema Elétrico Risk Averse Approach Alexander Shapiro and Wajdi Tekaya School of Industrial and
More informationAssembly systems with non-exponential machines: Throughput and bottlenecks
Nonlinear Analysis 69 (2008) 911 917 www.elsevier.com/locate/na Assembly systems with non-exponential machines: Throughput and bottlenecks ShiNung Ching, Semyon M. Meerkov, Liang Zhang Department of Electrical
More informationAlgorithmic Trading under the Effects of Volume Order Imbalance
Algorithmic Trading under the Effects of Volume Order Imbalance 7 th General Advanced Mathematical Methods in Finance and Swissquote Conference 2015 Lausanne, Switzerland Ryan Donnelly ryan.donnelly@epfl.ch
More informationCongestion Control for Best Effort
1 Congestion Control for Best Effort Prof. Jean-Yves Le Boudec Prof. Andrzej Duda Prof. Patrick Thiran ICA, EPFL CH-1015 Ecublens Andrzej.Duda@imag.fr http://icawww.epfl.ch Contents 2 Congestion control
More informationCS 188: Artificial Intelligence
CS 188: Artificial Intelligence Markov Decision Processes Dan Klein, Pieter Abbeel University of California, Berkeley Non-Deterministic Search 1 Example: Grid World A maze-like problem The agent lives
More informationCommunication Networks
Stochastic Simulation of Communication Networks Part 3 Prof. Dr. C. Görg www.comnets.uni-bremen.de VSIM 3-1 Table of Contents 1 General Introduction 2 Random Number Generation 3 Statistical i Evaluation
More informationReal Options and Game Theory in Incomplete Markets
Real Options and Game Theory in Incomplete Markets M. Grasselli Mathematics and Statistics McMaster University IMPA - June 28, 2006 Strategic Decision Making Suppose we want to assign monetary values to
More informationReinforcement Learning. Monte Carlo and Temporal Difference Learning
Reinforcement Learning Monte Carlo and Temporal Difference Learning Manfred Huber 2014 1 Monte Carlo Methods Dynamic Programming Requires complete knowledge of the MDP Spends equal time on each part of
More informationMulti-period mean variance asset allocation: Is it bad to win the lottery?
Multi-period mean variance asset allocation: Is it bad to win the lottery? Peter Forsyth 1 D.M. Dang 1 1 Cheriton School of Computer Science University of Waterloo Guangzhou, July 28, 2014 1 / 29 The Basic
More informationThe investment game in incomplete markets.
The investment game in incomplete markets. M. R. Grasselli Mathematics and Statistics McMaster University RIO 27 Buzios, October 24, 27 Successes and imitations of Real Options Real options accurately
More informationReinforcement Learning. Slides based on those used in Berkeley's AI class taught by Dan Klein
Reinforcement Learning Slides based on those used in Berkeley's AI class taught by Dan Klein Reinforcement Learning Basic idea: Receive feedback in the form of rewards Agent s utility is defined by the
More informationCombining Real Options and game theory in incomplete markets.
Combining Real Options and game theory in incomplete markets. M. R. Grasselli Mathematics and Statistics McMaster University Further Developments in Quantitative Finance Edinburgh, July 11, 2007 Successes
More informationOptimal Investment for Worst-Case Crash Scenarios
Optimal Investment for Worst-Case Crash Scenarios A Martingale Approach Frank Thomas Seifried Department of Mathematics, University of Kaiserslautern June 23, 2010 (Bachelier 2010) Worst-Case Portfolio
More informationMonte-Carlo Planning Look Ahead Trees. Alan Fern
Monte-Carlo Planning Look Ahead Trees Alan Fern 1 Monte-Carlo Planning Outline Single State Case (multi-armed bandits) A basic tool for other algorithms Monte-Carlo Policy Improvement Policy rollout Policy
More informationSOLVING ROBUST SUPPLY CHAIN PROBLEMS
SOLVING ROBUST SUPPLY CHAIN PROBLEMS Daniel Bienstock Nuri Sercan Özbay Columbia University, New York November 13, 2005 Project with Lucent Technologies Optimize the inventory buffer levels in a complicated
More informationComplex Decisions. Sequential Decision Making
Sequential Decision Making Outline Sequential decision problems Value iteration Policy iteration POMDPs (basic concepts) Slides partially based on the Book "Reinforcement Learning: an introduction" by
More informationOptimal Control of Batch Service Queues with Finite Service Capacity and General Holding Costs
Queueing Colloquium, CWI, Amsterdam, February 24, 1999 Optimal Control of Batch Service Queues with Finite Service Capacity and General Holding Costs Samuli Aalto EURANDOM Eindhoven 24-2-99 cwi.ppt 1 Background
More informationScheduling of multi-class multi-server queueing systems with abandonments
Scheduling of multi-class multi-server queueing systems with abandonments U. Ayesta b,c,e,f, P. Jaco d, V. Nova a a CERGE-EI, a joint worplace of Charles University in Prague and the Economics Institute
More informationarxiv: v2 [q-fin.tr] 29 Oct 2017
Instantaneous order impact and high-frequency strategy optimization in limit order books arxiv:1707.01167v2 [q-fin.tr] 29 Oct 2017 Federico Gonzalez and Mark Schervish, Department of Statistics, Carnegie
More informationSequential Decision Making
Sequential Decision Making Dynamic programming Christos Dimitrakakis Intelligent Autonomous Systems, IvI, University of Amsterdam, The Netherlands March 18, 2008 Introduction Some examples Dynamic programming
More informationSolving dynamic portfolio choice problems by recursing on optimized portfolio weights or on the value function?
DOI 0.007/s064-006-9073-z ORIGINAL PAPER Solving dynamic portfolio choice problems by recursing on optimized portfolio weights or on the value function? Jules H. van Binsbergen Michael W. Brandt Received:
More informationOptimal Control of an Inventory System With Joint Production and Pricing Decisions
IEEE TRANSACTIONS ON AUTOMATIC CONTROL, VOL. 6, NO. 2, DECEMBER 206 4235 Optimal Control of an Inventory System With Joint Production and Pricing Decisions Ping Cao and Jingui Xie Abstract In this study,
More informationOnline Appendix: Extensions
B Online Appendix: Extensions In this online appendix we demonstrate that many important variations of the exact cost-basis LUL framework remain tractable. In particular, dual problem instances corresponding
More information16 MAKING SIMPLE DECISIONS
247 16 MAKING SIMPLE DECISIONS Let us associate each state S with a numeric utility U(S), which expresses the desirability of the state A nondeterministic action A will have possible outcome states Result
More information,,, be any other strategy for selling items. It yields no more revenue than, based on the
ONLINE SUPPLEMENT Appendix 1: Proofs for all Propositions and Corollaries Proof of Proposition 1 Proposition 1: For all 1,2,,, if, is a non-increasing function with respect to (henceforth referred to as
More informationUQ, STAT2201, 2017, Lectures 3 and 4 Unit 3 Probability Distributions.
UQ, STAT2201, 2017, Lectures 3 and 4 Unit 3 Probability Distributions. Random Variables 2 A random variable X is a numerical (integer, real, complex, vector etc.) summary of the outcome of the random experiment.
More informationThe investment game in incomplete markets
The investment game in incomplete markets M. R. Grasselli Mathematics and Statistics McMaster University Pisa, May 23, 2008 Strategic decision making We are interested in assigning monetary values to strategic
More informationModelling the Sharpe ratio for investment strategies
Modelling the Sharpe ratio for investment strategies Group 6 Sako Arts 0776148 Rik Coenders 0777004 Stefan Luijten 0783116 Ivo van Heck 0775551 Rik Hagelaars 0789883 Stephan van Driel 0858182 Ellen Cardinaels
More informationChapter 5 Univariate time-series analysis. () Chapter 5 Univariate time-series analysis 1 / 29
Chapter 5 Univariate time-series analysis () Chapter 5 Univariate time-series analysis 1 / 29 Time-Series Time-series is a sequence fx 1, x 2,..., x T g or fx t g, t = 1,..., T, where t is an index denoting
More informationDynamic Resource Allocation for Spot Markets in Cloud Computi
Dynamic Resource Allocation for Spot Markets in Cloud Computing Environments Qi Zhang 1, Quanyan Zhu 2, Raouf Boutaba 1,3 1 David. R. Cheriton School of Computer Science University of Waterloo 2 Department
More informationIntroduction to Real Options
IEOR E4706: Foundations of Financial Engineering c 2016 by Martin Haugh Introduction to Real Options We introduce real options and discuss some of the issues and solution methods that arise when tackling
More informationFairness concept in terms of the utilization Hakyong Kim, Yongtak Lee, and Kiseon Kim Indexing terms: fairness, fairness index In this Letter, we disc
Fairness concept in terms of the utilization Hakyong Kim, Yongtak Lee, and Kiseon Kim Indexing terms: fairness, fairness index In this Letter, we discuss the concept of the fairness and dene a fairness
More informationMonte-Carlo Planning: Introduction and Bandit Basics. Alan Fern
Monte-Carlo Planning: Introduction and Bandit Basics Alan Fern 1 Large Worlds We have considered basic model-based planning algorithms Model-based planning: assumes MDP model is available Methods we learned
More informationFor students electing Macro (8701/Prof. Roe) & Micro (8703/Prof. Glewwe) option
WRITTEN PRELIMINARY Ph.D EXAMINATION Department of Applied Economics Jan./Feb. - 2011 Trade, Development and Growth For students electing Macro (8701/Prof. Roe) & Micro (8703/Prof. Glewwe) option Instructions
More informationOptimal rebalancing of portfolios with transaction costs assuming constant risk aversion
Optimal rebalancing of portfolios with transaction costs assuming constant risk aversion Lars Holden PhD, Managing director t: +47 22852672 Norwegian Computing Center, P. O. Box 114 Blindern, NO 0314 Oslo,
More informationTemporal Abstraction in RL
Temporal Abstraction in RL How can an agent represent stochastic, closed-loop, temporally-extended courses of action? How can it act, learn, and plan using such representations? HAMs (Parr & Russell 1998;
More informationFramework and Methods for Infrastructure Management. Samer Madanat UC Berkeley NAS Infrastructure Management Conference, September 2005
Framework and Methods for Infrastructure Management Samer Madanat UC Berkeley NAS Infrastructure Management Conference, September 2005 Outline 1. Background: Infrastructure Management 2. Flowchart for
More informationReinforcement Learning and Simulation-Based Search
Reinforcement Learning and Simulation-Based Search David Silver Outline 1 Reinforcement Learning 2 3 Planning Under Uncertainty Reinforcement Learning Markov Decision Process Definition A Markov Decision
More informationPerformance Analysis of Cognitive Radio Spectrum Access with Prioritized Traffic
Performance Analysis of Cognitive Radio Spectrum Access with Prioritized Traffic Vamsi Krishna Tumuluru, Ping Wang, and Dusit Niyato Center for Multimedia and Networ Technology (CeMNeT) School of Computer
More informationDynamic Pricing with Varying Cost
Dynamic Pricing with Varying Cost L. Jeff Hong College of Business City University of Hong Kong Joint work with Ying Zhong and Guangwu Liu Outline 1 Introduction 2 Problem Formulation 3 Pricing Policy
More informationUtility Indifference Pricing and Dynamic Programming Algorithm
Chapter 8 Utility Indifference ricing and Dynamic rogramming Algorithm In the Black-Scholes framework, we can perfectly replicate an option s payoff. However, it may not be true beyond the Black-Scholes
More informationMDP Algorithms. Thomas Keller. June 20, University of Basel
MDP Algorithms Thomas Keller University of Basel June 20, 208 Outline of this lecture Markov decision processes Planning via determinization Monte-Carlo methods Monte-Carlo Tree Search Heuristic Search
More informationMDPs and Value Iteration 2/20/17
MDPs and Value Iteration 2/20/17 Recall: State Space Search Problems A set of discrete states A distinguished start state A set of actions available to the agent in each state An action function that,
More informationRisk-Averse Anticipation for Dynamic Vehicle Routing
Risk-Averse Anticipation for Dynamic Vehicle Routing Marlin W. Ulmer 1 and Stefan Voß 2 1 Technische Universität Braunschweig, Mühlenpfordtstr. 23, 38106 Braunschweig, Germany, m.ulmer@tu-braunschweig.de
More informationCSE 473: Artificial Intelligence
CSE 473: Artificial Intelligence Markov Decision Processes (MDPs) Luke Zettlemoyer Many slides over the course adapted from Dan Klein, Stuart Russell or Andrew Moore 1 Announcements PS2 online now Due
More information