China
Africa
Explore chapters and articles related to this topic
Distributed Artificial Intelligence and Agents
Published in Weiming Shen, Douglas H. Norrie, Jean-Paul A. Barthès, Multi-Agent Systems for Concurrent Intelligent Design and Manufacturing, 2019
Weiming Shen, Douglas H. Norrie, Jean-Paul A. Barthes
In some cases (as with robots), a pure reactive architecture is not enough to ensure a consistent permanent behavior in agents or the necessary machinery to build symbolic internal representations to use for symbolic planning is too complex to be of any practical use. For such situations, researchers have proposed new mechanisms for introducing goals and goal directed behaviour into reactive agents. Maes (1990) introduced the concept of a network propagating energy activation, for selecting among competing goals while taking into account external stimuli. Noteworthy is the anytime algorithm used for the level of the threshold controlling activation of actions. Other researchers have introduced a multi-layered architecture with a reactive lower level and more deliberative upper levels, as in the InteRRaP approach, of Müller et al. (1995). Such architectures are quite promising but may be difficult to implement.
An adaptive patch approximation algorithm for bicriteria convex mixed-integer problems
Published in Optimization, 2022
In this paper, we develop an algorithm which is designed contrarily to the above characteristics. We pose the following goals to achieve with our method: The approximation returned by the algorithm in each step should be an inner approximation, i.e. all points in the approximation should correspond to feasible solutions that can be directly computed.The algorithm should provide an approximation of the Pareto frontier with a clearly quantified approximation guarantee at every step. In particular, the algorithm can be stopped at every moment and we still obtain a result with a clear approximation guarantee, i.e. the algorithm should be an anytime algorithm [25]. The approximation quality obtained until each given time limit should be competitive with every other algorithm that requires a similar amount of time.The approximation quality obtained by the algorithm, respectively, the needed number of iterations to reach some given accuracy, should be bounded with respect to some intrinsic property of the problem we solve. In particular, the performance of the algorithm should be not too far away from a theoretical optimum that is given by the number of patches minimally needed to describe an approximation to the Pareto frontier.
Anytime clustering of data streams while handling noise and concept drift
Published in Journal of Experimental & Theoretical Artificial Intelligence, 2022
Jagat Sesh Challa, Poonam Goyal, Ajinkya Kokandakar, Dhananjay Mantri, Pranet Verma, Sundar Balasubramaniam, Navneet Goyal
To address this issue, anytime stream clustering algorithms were proposed, which perform anytime online maintenance of micro-clusters for streams that have varying inter-arrival rate of objects. They can handle objects arriving at any stream speed, deliver a result at any given point in time, and use more time if available, to refine the result to the highest possible degree. Figure 1 depicts the characteristic of an anytime algorithm where the accuracy of the result improves with increase in time allowance. Only a few anytime stream clustering algorithms exist in literature, which include ClusTree (Kranen et al., 2011), LiarTree (Hassani et al., 2011) and SubClusTree (Hassani et al., 2014). However, there are a few drawbacks associated with them. ClusTree’s (Kranen et al., 2011) algorithm for insertion of incoming objects, uses distance computations performed from each object to the mean of the existing micro-clusters, which leads to a reduction in overall purity of the micro-clusters indexed especially for high dimensional data. Moreover, ClusTree does not handle noise and concept drift. LiarTree (Hassani et al., 2011) is a variation of ClusTree that handles noise & concept drift. However, its method builds new sub-trees for each noise-to-concept transition, which distorts spatial locality and thus hampers purity of the micro-clusters produced. SubClusTree is an extension of ClusTree for anytime subspace clustering. Also, all of the above are proposed for single-port data streams only.
Solving Partially Observable Environments with Universal Search Using Dataflow Graph-Based Programming Model
Published in IETE Journal of Research, 2021
Swarna Kamal Paul, Parama Bhaumik
Silver et al. [20] used Monte-Carlo tree search to solve large POMDPs. It constructs a state belief tree and combines Monte-Carlo updates with Monte-Carlo tree search to handle the curse of dimensionality. However, in a goal-oriented environment with sparse and delayed reward distribution, planning in belief state with a bounded horizon may become difficult, as the agent rarely receives a reward to update the value of belief state. Ye et al. presented an online planning algorithm for solving POMDPs using Determinized Sparse Partially Observable Tree (DESPOT) [21] which is a sparse approximation of belief trees. It is created by randomly sampling state, observation, and action trajectories. Though it has been shown, relatively smaller size of belief tree can achieve near-optimal performance yet it can be quite large for large observation and action space. An anytime algorithm is proposed to incrementally build DESPOT using some heuristic based on available planning. However, the choice of heuristic is important and non-trivial for delivering optimal performance. Sunberg et al. [22] explored for the solution of POMDPs in continuous state, observation, and action space. They proposed two variants of POMCP and particle filter trees.