分散ラグランジュ緩和プロトコルにおけるバンドル法 (original) (raw)
分散ラグランジュ緩和プロトコルにおけるバンドル法
2016, Transactions of the Japanese Society for Artificial Intelligence
The Generalized Mutual Assignment Problem (GMAP) is a maximization problem in distributed environments, where multiple agents select goods under resource constraints. Distributed Lagrangian Relaxation Protocols (DisLRP) are peer-to-peer communication protocols for solving GMAP instances. In DisLRPs, agents seek a good quality upper bound for the optimal value by solving the Lagrangian dual problem, which is a convex minimization problem. Existing DisLRPs exploit a subgradient method to explore a better upper bound by updating the Lagrange multipliers (prices) of goods. While the computational complexity of the subgradient method is very low, it cannot detect the fact that an upper bound converges to the minimum. Moreover, solution oscillation sometimes occurs, which is one of the critical issues for the subgradient method. In this paper, we present a new DisLRP with a Bundle Method and refer to it as Bundle DisLRP (BDisLRP). The bundle method, which is also called the stabilized cutting planes method, has recently attracted much attention as a way to solve Lagrangian dual problems in centralized environments. We show that this method can also work in distributed environments. We experimentally compared BDisLRP with Adaptive DisLRP (ADisLRP), which is a previous protocol that exploits the subgradient method, to demonstrate that BDisLRP performed convergence faster with better quality upper bounds than ADisLRP.
Sign up for access to the world's latest research.
checkGet notified about relevant papers
checkSave papers to use in your research
checkJoin the discussion with peers
checkTrack your impact
Related papers
1999
Kyoto University (京都大学)0048新制・課程博士博士(工学)甲第7838号工博第1818号新制||工||1140(附属図書館)UT51-99-G432京都大学大学院工学研究科原子核工学専攻(主査)教授 芹澤 昭示, 教授 小森 悟, 教授 三島 嘉一郎学位規則第4条第1項該
Transactions of the Japanese Society for Artificial Intelligence, 2013
A "hub" is an object closely surrounded by, or very similar to, many other objects in the dataset. Recent studies by Radovanović et al. demonstrated that in high dimensional spaces, objects close to the data centroid tend to become hubs. In this paper, we show that the family of kernels based on the graph Laplacian makes all objects in the dataset equally similar to the centroid, and thus they are expected to make less hubs when used as a similarity measure. We investigate this hypothesis using both synthetic and real-world data. It turns out that these kernels suppress hubs in some cases but not always, and the results seem to be affected by the size of the data-a factor not discussed previously. However, for the datasets in which hubs are indeed reduced by the Laplacian-based kernels, these kernels work well in classification and information retrieval tasks. This result suggests that the amount of hubs, which can be readily computed in an unsupervised fashion, can be a yardstick of whether Laplacian-based kernels work effectively for a given data.
Journal of applied mechanics, 2006
When we conduct forced vibration using a small shaker to identify structural damage, the responses have the possibility to contain measurement noise due to ambient vibrations. To surmount this, the independent component analysis (ICA) was applied to extract the excitation responses from contaminated measurements. ICA is an algorithm recently used for blind source separation. It separates signals only using the statistical independence of original signals. We assume that a structure is excited by independent input forces: harmonic excitation and ambient vibrations, and tried to extract the harmonic excitation responses. We proposed its application as preprocess of damage identification to improve the identification accuracy. Though numerical simulation, we confirmed that the technique worked efficiently to obtain identification results of high accuracy.
Transactions of the Japanese Society for Artificial Intelligence, 2013
Distributed Constraint Optimization problems (DCOPs) have been studied as a fundamental model of multiagent cooperation. In traditional DCOPs, all agents cooperate to optimize the sum of their cost functions. However, in practical systems some agents may desire to select the value of their variables without cooperation. In special cases, such agents may take the values with the worst impact on the quality of the result reachable by the optimization process. Similar classes of problems have been studied as Quantified (Distributed) Constraint Problems, where the variables of the CSP have existential/universal quantifiers. All constraints should be satisfied independently of the value taken by universal variables. In this paper, a Quantified Distributed Constraint Optimization problem (QDCOP) that extends the framework of DCOPs is presented. We apply existential/universal quantifiers to distinct uncooperative variables. A universally quantified variable is left unassigned by the optimization as the result has to hold when it takes any value from its domain, while an existentially quantified variable takes exactly one of its values for each context. We consider that the QDCOP applies the concept of game tree search to DCOP. If the original problem is a minimization problem, agents that own universally quantified variables may intend to maximize the cost value in the worst case. Other agents normally intend to optimize the minimizing problems. Therefore, only the bounds, especially the upper bounds, of the optimal value are guaranteed. The purpose of the new class of problems is to compute such bounds, as well as to compute sub-optimal solutions. For the QDCOP, we propose solution methods that are based on min-max/alpha-beta and ADOPT algorithms.
2017
日本では「不思議のダンジョン」シリーズで名高いローグライクゲームは，マップや敵配置がランダムなこと，そのため臨機応変にアイテム等を使い分ける必要があることなどを特徴とする一人用ダンジョン探索ゲームの総称である．囲碁や将棋などよりは複雑で，StarcraftやCivilizationなどよりは単純なこのゲーム群は，短期的戦術と長期的スケジューリングの両方を要するなど，高度なゲームAIを研究するうえでの良い課題をいくつも持つ．本論文では，学術用プラットフォームを共有できるようにするために，ローグライクゲームの最低限の要素を持たせたゲームのルールを提案した．そのうえで，ルールベースのコンピュータプレイヤ，短期的な戦術行動を改善するためのモンテカルロ法プレイヤ，長期的な視点での行動が取れるようにするための教師あり学習を提案した．: Roguelike Game, known as "Mystery Dungeon" in Japan, is a genre of one-player role-playing video games. In roguelike games, the map is randomly created and the monsters are randomly located every play, then the player needs to select his strategy properly according to the given situation. Roguelike games are more complex than the game of Go or Chess, and less complex than Starcraft or Civilization. They are less complex, still they have several typical problems for sophisticated game AIs, for example both of short-time strategy and long-time scheduling are necessary and the balancing is important. In this paper, ...
Loading Preview
Sorry, preview is currently unavailable. You can download the paper by clicking the button above.