Decentralized Stochastic Planning with Anonymity in Interactions

Authors: Pradeep Varakantham, Yossiri Adulyasak, Patrick Jaillet

AAAI 2014 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental In our experimental results, we also compare against Softmax based Flow Update (SMFU) algorithm, which was developed for competitive games and can be adapted to solve D-SPAIT problems. In this section, we demonstrate the following: (1) Run-time performance of SOLVEDSPAIT-LINEAR(), SOLVEDSPAIT-PWC approaches. (2) Run-time performance, scalability and solution quality for the local optimal approach in SOLVEDSPAIT-PWLC. (3) Performance comparison against the SMFU(Varakantham et al. 2012), which also exploits anonymous interactions in competitive settings.
Researcher Affiliation Academia School of Information Systems, Singapore Management University Singapore-MIT Alliance for Research and Technology (SMART), Massachusetts Institute of Technology Department of Electrical Engineering and Computer Science, Massachussets Institute of Technology
Pseudocode Yes Algorithm 1 SOLVEDSPAIT(), Algorithm 2 SOLVEDSPAIT-HOMOGENOUS(), Algorithm 3 SOLVEDSPAIT-LINEAR ( m, c ), Algorithm 4 SOLVEDSPAIT-PWC(), Algorithm 5 Solve DSPAIT-PWLC()
Open Source Code No The paper does not provide any explicit statements about releasing source code or links to a code repository.
Open Datasets No We generated random DSPAIT problems, where both the reward and transition functions were generated randomly. For the transition function, we varied the reachability (number of states with positive probability) of the states and generated random transition probabilities. For the reward function, we generated random numbers between a range while satisfying the assumptions of the specific categories of functions (ex: negative slopes, monotonically non-increasing, etc.). We then used our optimization algorithms to solve these random DSPAIT problems. No access information is provided for these randomly generated problems.
Dataset Splits No The paper mentions generating 'random DSPAIT problems' and 'randomly generated 25 instances for each problem type', but it does not provide specific details on train/validation/test splits.
Hardware Specification Yes All the linear and quadratic optimization problems were solved using the commercial optimization software CPLEX 12.2 on a 1.8 GHz Intel Core i5 machine with 8GB 1600 MHz DDR3 RAM.
Software Dependencies Yes All the linear and quadratic optimization problems were solved using the commercial optimization software CPLEX 12.2 on a 1.8 GHz Intel Core i5 machine with 8GB 1600 MHz DDR3 RAM.
Experiment Setup Yes We generated random DSPAIT problems, where both the reward and transition functions were generated randomly. For the transition function, we varied the reachability (number of states with positive probability) of the states and generated random transition probabilities. For the reward function, we generated random numbers between a range while satisfying the assumptions of the specific categories of functions (ex: negative slopes, monotonically non-increasing, etc.). A problem type is denoted by a cross product of number of states (zones), number of actions (zones) and number of decision epochs (time horizon). We randomly generated 25 instances for each problem type... We show the performance of the SOLVEDSPAIT-PWLC function by considering 10 piecewise linear components for every state action pair, time step corresponding to the reward function.