Mutual information oriented deep skill chaining for multi‐agent reinforcement learning

Multi‐agent reinforcement learning relies on reward signals to guide the policy networks of individual agents.However,in high‐dimensional continuous spaces,the non‐stationary environment can provide outdated experiences that hinder convergence,resulting in ineffective training performance for multi‐agent systems.To tackle this issue,a novel reinforcement learning scheme,Mutual Information Oriented Deep Skill Chaining(MioDSC),is proposed that generates an optimised cooperative policy by incorporating intrinsic rewards based on mutual information to improve exploration efficiency.These rewards encourage agents to diversify their learning process by engaging in actions that increase the mutual information between their actions and the environment state.In addition,MioDSC can generate cooperative policies using the options framework,allowing agents to learn and reuse complex action sequences and accelerating the convergence speed of multi‐agent learning.MioDSC was evaluated in the multi‐agent particle environment and the StarCraft multi‐agent challenge at varying difficulty levels.The experimental results demonstrate that MioDSC outperforms state‐of‐the‐art methods and is robust across various multi‐agent system tasks with high stability.

Fast Distributed Demand Response Algorithm in Smart Grid

This paper proposes a fast distributed demand response U+0028 DR U+0029 algorithm for future smart grid based on primaldual interior method and Gaussian belief propagation U+0028 GaBP U+0029 solver. At the beginning of each time slot, each end-user U+002F energysupplier exchanges limited rounds of messages that are not private with its neighbors, and computes the amount of energy consumption U+002F generation locally. The proposed demand response algorithm converges rapidly to a consumption U+002F generation decision that yields the optimal social welfare when the demands of endusers are low. When the demands are high, each end-user U+002F energysupplier estimates its energy consumption U+002F generation quickly such that a sub-optimal social welfare is achieved and the power system is ensured to operate within its capacity constraints. The impact of distributed computation errors on the proposed algorithm is analyzed theoretically. The simulation results show a good performance of the proposed algorithm. © 2017 Chinese Association of Automation.

Decentralized Multigrid for In-situ Big Data Computing

Modern seismic sensors are capable of recording high precision vibration data continuously for several months. Seismic raw data consists of information regarding earthquake’s origin time, location, wave velocity, etc.Currently, these high volume data are gathered manually from each station for analysis. This process restricts us from obtaining high-resolution images in real-time. A new in-network distributed method is required that can obtain a high-resolution seismic tomography in real time. In this paper, we present a distributed multigrid solution to reconstruct seismic image over large dense networks. The algorithm performs in-network computation on large seismic samples and avoids expensive data collection and centralized computation. Our evaluation using synthetic data shows that the proposed method accelerates the convergence and reduces the number of messages exchanged. The distributed scheme balances the computation load and is also tolerant to severe packet loss.