Optimal Response Learning and Its Convergence in Multiagent Domains

In multiagent reinforcement learning, with different assumptions of the opponents＇ policies, an agent adopts quite different learning rules, and gets different learning performances. We prove that, in multiagent domains, convergence of the Q values is guaranteed only when an agent behaves optimally and its opponents＇ strategies satisfy certain conditions, and an agent can get best learning performances when it adopts the same learning algorithm as that of its opponents.     

Indirect Watermarking

Watermarking system is a complex system requiring both expertise and specialized techniques with security concerns.Thus to relieve the multimedia manufacturers of the complexities of designing complicated watermarking systems, this paper presents a new watermarking model in which watermarking is provided as a kind of service by the specialized watermarking companies. In order to make this service model work, a simple permutation scheme to hide the true contents is proposed. The paper also proves the security of the scheme against the attacks from the watermarking companies. Finally, the effectiveness of the scheme is demonstrated through experiments.     

Finding Recently Frequent Items over Online Data Streams

In this paper, a new algorithm HCOUNT+ is proposed to find frequent items over data stream based on the HCOUNT algorithm. The new algorithm adopts aided measures to improve the precision of HCOUNT greatly. In addition, HCOUNT+ is introduced to time critical applications and a novel sliding windows-based algorithm SL-HCOUNT+ is proposed to mine the most frequent items occurring recently. This algorithm uses limited memory (nB·(1+α)·eε·ln-M/lnρ(α<1) counters), requires constant processing time per packet (only (1+α)·ln·-M/lnρ(α<1) counters are updated), makes only one pass over the streaming data, and is shown to work well in the experimental results.     

一般和对策中基于协商的多代理强化学习

一般和对策中,只考虑个体理性的多代理协作是一种无全局目标的协作.代理学习基于对手策略假设,不能保证假设的正确性.为此通过定义代理协作的集体目标,提出了一种基于多代理协商的代理强化学习算法.代理选择协商策略,并惩罚偏离该策略的代理来保证协商策略的执行.文中给出了学习收敛的条件及证明,并以实例加以分析.     

Multi-agent reinforcement learning based on policies of global objective

1 .INTRODUCTIONBecause an agent’s rewardis a function of all agents’joint action, when applying RL[1]to multi-agent do-mains ,some fundamental change should be made .Byadopting single agentQlearning[2]to Markovgames,several algorithms have been proposed,suchas Littman’s mini maxQ-learning( mini max-Q)[3],Hu et al’s NashQ-learning(Nash-Q)[4 ,5], Claus etal’s cooperative multi-agentQ-learning[6], Bowlinget al’s multi-agent learningQ-learning using a vari-able learning rate[7 ~9],…     

基于CORBA的PDM系统实现方式

如何将产品数据管理系统（ＰＤＭ）建立在异构的环境上,并保护用户在ＣＩＭＳ中原有的资源是ＰＤＭ能否具有生命力的关键．论述了一个典型的ＰＤＭ系统体系及其基于ＣＯＲＢＡ的实现,重点讨论了ＰＤＭ与ＣＡｘ、ＭＲＰＩ应用系统的集成工具在ＣＯＲＢＡ上的实现以及数据库接口在ＣＯＲＢＡ上的实现,提出一个利用ＣＯＲＢＡ解决上述问题的方案,从而为ＰＤＭ系统的实现提出了一种新的思路．     

Novel Privacy Preserving Method of Countering the Threats from Priori Knowledge

Recently,many data anonymization methods have been proposed to protect privacy in the applications of data mining.But few of them have considered the threats from user's priori knowledge of data patterns.To solve this problem,a flexible method was proposed to randomize the dataset,so that the user could hardly obtain the sensitive data even knowing data relationships in advance.The method also achieves a high level of accuracy in the mining process as demonstrated in the experiments.