Piecemeal Learning of an Unknown Environment

We introduce a new learning problem: learning a graph bypiecemeal search, in which the learner must return every so often to its starting point (for refueling, say). We present two linear-time piecemeal-search algorithms for learningcity-block graphs: grid graphs with rectangular obstacles.     

Evolution of neural networks for classification and regression

Although Artificial Neural Networks (ANNs) are importantdata mining techniques, the search for the optimal ANN is a challenging task: the ANN should learn the input–output mapping without overfitting the data and training algorithms may get trapped in local minima. The use of Evolutionary Computation (EC) is a promising alternative for ANN optimization. This work presents two hybrid EC/ANN algorithms: the first evolves neural topologies while the latter performs simultaneous optimization of architectures and weights. Sixteen real-world tasks were used to test these strategies. Competitive results were achieved when compared with a heuristic model selection and other Data Mining algorithms.     

YACS: a new learning classifier system using anticipation

 A new and original trend in the learning classifier system (LCS) framework is focussed on latent learning. These new LCSs call upon classifiers with a (condition), an (action) and an (effect) part. In psychology, latent learning is defined as learning without getting any kind of reward. In the LCS framework, this process is in charge of discovering classifiers which are able to anticipate accurately the consequences of actions under some conditions. Accordingly, the latent learning process builds a model of the dynamics of the environment. This model can be used to improve the policy learning process. This paper describes YACS, a new LCS performing latent learning, and compares it with ACS.     

Labeling Q-learning in hidden state environments

Recently,reinforcement learning (RL) methods have been used for learning problems in environments with embedded hidden states. However, conventional RL methods have been limited to handlingMarkov decision process problems. In order to overcome hidden states, several algorithms were proposed, but these need an extreme amount of memory of past sequences which represent historical state transitions. The aim of this work was to extend our previously proposed algorithm for hidden states in an environment, calledlabeling Q-learning (LQ-learning), which reinforces incompletely observed perception by labeling. In LQ-learning, the agent has a perception structure which consists of pairs of observations and labels. From these pairs, the agent can distinguish more exactly hidden states which look the same but are actually different each other. Labeling is carried out by labeling functions. Numerous labeling functions can be considered, but here we introduce some labeling functions based on the sequence of only the last and the current observations. This extended LQ-learning is applied to grid-world problems which have hidden states. The results of these simulations show the availability of LQ-learning. This work was presented in part at the Sixth International Symposium on Artificial Life and Robotics, Tokyo, January 15–17, 2001     

Mantaining Dynamic Matrices for Fully Dynamic Transitive Closure

In this paper we introduce a general framework for casting fully dynamic transitive closure into the problem of reevaluating polynomials over matrices. With this technique, we improve the best known bounds for fully dynamic transitive closure. In particular, we devise a deterministic algorithm for general directed graphs that achieves O(n ²) amortized time for updates, while preserving unit worst-case cost for queries. In case of deletions only, our algorithm performs updates faster in O(n) amortized time. We observe that fully dynamic transitive closure algorithms with O(1) query time maintain explicitly the transitive closure of the input graph, in order to answer each query with exactly one lookup (on its adjacency matrix). Since an update may change as many as Ω(n ²) entries of this matrix, no better bounds are possible for this class of algorithms. This work has been partially supported by the Sixth Framework Programme of the EU under contract number 507613 (Network of Excellence “EuroNGI: Designing and Engineering of the Next Generation Internet”), and number 001907 (“DELIS: Dynamically Evolving, Large Scale Information Systems”), and by the Italian Ministry of University and Research (Project “ALGO-NEXT: Algorithms for the Next Generation Internet and Web: Methodologies, Design and Experiments”). Portions of this paper have been presented at the 41st Annual Symp. on Foundations of Computer Science, 2000.     

Combining Committee-Based Semi-Supervised Learning and Active Learning

Many data mining applications have a large amount of data but labeling data is usually difficult, expensive, or time consuming, as it requires human experts for annotation. Semi-supervised learning addresses this problem by using unlabeled data together with labeled data in the training process. Co-Training is a popular semi-supervised learning algorithm that has the assumptions that each example is represented by multiple sets of features (views) and these views are sufficient for learning and independent given the class. However, these assumptions are strong and are not satisfied in many real-world domains. In this paper, a single-view variant of Co-Training, called Co-Training by Committee (CoBC) is proposed, in which an ensemble of diverse classifiers is used instead of redundant and independent views. We introduce a new labeling confidence measure for unlabeled examples based on estimating the local accuracy of the committee members on its neighborhood. Then we introduce two new learning algorithms, QBC-then-CoBC and QBC-with-CoBC, which combine the merits of committee-based semi-supervised learning and active learning. The random subspace method is applied on both C4.5 decision trees and 1-nearest neighbor classifiers to construct the diverse ensembles used for semi-supervised learning and active learning. Experiments show that these two combinations can outperform other non committee-based ones.     

Cooperative Multi-Agent Learning: The State of the Art

Cooperative multi-agent systems (MAS) are ones in which several agents attempt, through their interaction, to jointly solve tasks or to maximize utility. Due to the interactions among the agents, multi-agent problem complexity can rise rapidly with the number of agents or their behavioral sophistication. The challenge this presents to the task of programming solutions to MAS problems has spawned increasing interest in machine learning techniques to automate the search and optimization process. We provide a broad survey of the cooperative multi-agent learning literature. Previous surveys of this area have largely focused on issues common to specific subareas (for example, reinforcement learning, RL or robotics). In this survey we attempt to draw from multi-agent learning work in a spectrum of areas, including RL, evolutionary computation, game theory, complex systems, agent modeling, and robotics. We find that this broad view leads to a division of the work into two categories, each with its own special issues: applying a single learner to discover joint solutions to multi-agent problems (team learning), or using multiple simultaneous learners, often one per agent (concurrent learning). Additionally, we discuss direct and indirect communication in connection with learning, plus open issues in task decomposition, scalability, and adaptive dynamics. We conclude with a presentation of multi-agent learning problem domains, and a list of multi-agent learning resources.     

Network-based hybrid genetic algorithm for scheduling in FMS environments

Scheduling in flexible manufacturing systems (FMS) must take account of the shorter lead-time, the multiprocessing environment, the flexibility of alternative workstations with different processing times, and the dynamically changing states. The best scheduling approach, as described here, is to minimize makespan t _M, total flow time t _F, and total tardiness penalty p _T. However, in the case of manufacturing system problems, it is difficult for those with traditional optimization techniques to cope with this. This article presents a new flow network-based hybrid genetic algorithm (hGA) approach for generating static schedules in a FMS environment. The proposed method is combined with the neighborhood search technique in a mutation operation to improve the solution of the FMS problem, and to enhance the performance of the genetic search process. We update the change in swap mutation and the local search-based mutation ration. Numerical experiments show that the proposed flow network-based hGA is both effective and efficient for FMS problems.This work was presented in part at the 8th International Symposium on Artificial Life and Robotics, Oita, Japan, January 24–26, 2003     

Adjunctions in Pyramids,Curve Evolution and Scale-Spaces

We have been witnessing lately a convergence among mathematical morphology and other nonlinear fields, such as curve evolution, PDE-based geometrical image processing, and scale-spaces. An obvious benefit of such a convergence is a cross-fertilization of concepts and techniques among these fields. The concept of adjunction however, so fundamental in mathematical morphology, is not yet shared by other disciplines. The aim of this paper is to show that other areas in image processing can possibly benefit from the use of adjunctions. In particular, a strong relationship between pyramids and adjunctions is presented. We show how this relationship may help in analyzing existing pyramids, and construct new pyramids. Moreover, it will be explained that adjunctions based on a curve evolution scheme can provide idempotent shape filters. This idea is illustrated in this paper by means of a simple affine-invariant polygonal flow. Finally, the use of adjunctions in scale-space theory is also addressed.     

Compositional CSP Traces Refinement Checking

Compositional verification using assume-guarantee reasoning has recently seen an uprise due to the introduction of automatic techniques for learning assumptions. In this paper, we transfer this technique to a setting with CSP as modelling and property specification language, and present an approach to compositional traces refinement checking. The approach has been implemented using the CSP model checker FDR as teacher during learning. The implementation shows that the compositional approach can both drastically outperform as well as underperform FDR's performance, depending on the example at hand.     

Feasible Direction Decomposition Algorithms for Training Support Vector Machines

The article presents a general view of a class of decomposition algorithms for training Support Vector Machines (SVM) which are motivated by the method of feasible directions. The first such algorithm for the pattern recognition SVM has been proposed in Joachims, T. (1999, Schölkopf et al. (Eds.) Advances in kernel methods-Support vector learning (pp. 185–208). MIT Press). Its extension to the regression SVM—the maximal inconsistency algorithm—has been recently presented by the author (Laskov, 2000, Solla, Leen, & Müller (Eds.) Advances in neural information processing systems 12 (pp. 484–490). MIT Press). A detailed account of both algorithms is carried out, complemented by theoretical investigation of the relationship between the two algorithms. It is proved that the two algorithms are equivalent for the pattern recognition SVM, and the feasible direction interpretation of the maximal inconsistency algorithm is given for the regression SVM. The experimental results demonstrate an order of magnitude decrease of training time in comparison with training without decomposition, and, most importantly, provide experimental evidence of the linear convergence rate of the feasible direction decomposition algorithms.     

Dynamic learning from multiple examples for semantic object segmentation and search

We present a novel “dynamic learning” approach for an intelligent image database system to automatically improve object segmentation and labeling without user intervention, as new examples become available, for object-based indexing. The proposed approach is an extension of our earlier work on “learning by example,” which addressed labeling of similar objects in a set of database images based on a single example. The proposed dynamic learning procedure utilizes multiple example object templates to improve the accuracy of existing object segmentations and labels. Multiple example templates may be images of the same object from different viewing angles, or images of related objects. This paper also introduces a new shape similarity metric called normalized area of symmetric differences (NASD), which has desired properties for use in the proposed “dynamic learning” scheme, and is more robust against boundary noise that results from automatic image segmentation. Performance of the dynamic learning procedures has been demonstrated by experimental results.     

Property Testing in Bounded Degree Graphs

Abstract. We further develop the study of testing graph properties as initiated by Goldreich, Goldwasser and Ron. Loosely speaking, given an oracle access to a graph, we wish to distinguish the case when the graph has a pre-determined property from the case when it is ``far' from having this property. Whereas they view graphs as represented by their adjacency matrix and measure the distance between graphs as a fraction of all possible vertex pairs, we view graphs as represented by bounded-length incidence lists and measure the distance between graphs as a fraction of the maximum possible number of edges. Thus, while the previous model is most appropriate for the study of dense graphs, our model is most appropriate for the study of bounded-degree graphs. In particular, we present randomized algorithms for testing whether an unknown bounded-degree graph is connected, k -connected (for k>1 ), cycle-free and Eulerian. Our algorithms work in time polynomial in 1/ɛ , always accept the graph when it has the tested property, and reject with high probability if the graph is ɛ -far from having the property. For example, the 2-connectivity algorithm rejects (with high probability) any N -vertex d -degree graph for which more than ɛ dN edges need to be added in order to make the graph 2-edge-connected. In addition we prove lower bounds of Ω(\sqrt N ) on the query complexity of testing algorithms for the bipartite and expander properties.     

The Setting of Shelf Heights and the Distribution of Box Sizes in Two-dimensional Shelf Packing

In this paper we explore two factors which have been relatively neglected in the study of shelf packing algorithms. Boxes arrive one by one at a bin that is shelved over its width and are placed into the shelves side by side, left-justified. We consider the setting of the shelf heights, which needs to be decided upon before packing starts. We investigate the relationship between the number of boxes to be packed and the number of shelf heights that leads to minimal space wastage in the resulting packings. We also consider the distribution of the box sizes, for which we distinguish three types: (a) the continuous uniform distribution U(0, 1); (b) the discrete uniform distribution where M sizes {1/M, 2/M,……,1} have equal probability of being chosen; and (c) an alternative discrete uniform distribution where M sizes {s₁, s₂,……,s_M} are drawn from U(0, 1), which then each have equal probability of being chosen. In simulation studies we illustrate important differences between the three types of distribution.     

A linear-time algorithm to find a pair of arc-disjoint spanning in-arborescence and out-arborescence in a directed acyclic graph

Suppose that we are given a directed graph D=(V,A) with specified vertices r₁,r₂V. In this paper, we consider the problem of discerning the existence of a pair of arc-disjoint spanning in-arborescence rooted at r₁ and out-arborescence rooted at r₂, and finding such arborescences if they exist. It is known (Bang-Jensen (1991) [1]) that this problem is -complete even if r₁=r₂. As a special case, it is only known (Bang-Jensen (1991) [1]) that this problem in a tournament can be solved in polynomial time. In this paper, we give a linear-time algorithm for this problem in a directed acyclic graph. We also consider an extension of our problem to the case where we have multiple roots for in-arborescences and out-arborescences, respectively.     

Introspective Sorting and Selection Algorithms

Quicksort is the preferred in-place sorting algorithm in many contexts, since its average computing time on uniformly distributed inputs is Θ(N log N), and it is in fact faster than most other sorting algorithms on most inputs. Its drawback is that its worst-case time bound is Θ(N²). Previous attempts to protect against the worst case by improving the way quicksort chooses pivot elements for partitioning have increased the average computing time too much – one might as well use heapsort, which has a Θ(N log N) worst-case time bound, but is on the average 2–5 times slower than quicksort. A similar dilemma exists with selection algorithms (for finding the i-th largest element) based on partitioning. This paper describes a simple solution to this dilemma: limit the depth of partitioning, and for subproblems that exceed the limit switch to another algorithm with a better worst-case bound. Using heapsort as the ‘stopper’ yields a sorting algorithm that is just as fast as quicksort in the average case, but also has an Θ(N log N) worst case time bound. For selection, a hybrid of Hoare's FIND algorithm, which is linear on average but quadratic in the worst case, and the Blum–Floyd–Pratt–Rivest–Tarjan algorithm is as fast as Hoare's algorithm in practice, yet has a linear worst-case time bound. Also discussed are issues of implementing the new algorithms as generic algorithms, and accurately measuring their performance in the framework of the C+:+ Standard Template Library. ©1997 by John Wiley & Sons, Ltd.     

Approximation Algorithms for Reconstructing the Duplication History of Tandem Repeats

Computing the duplication history of a tandem repeated region is an important problem in computational biology (Fitch in Genetics 86:623–644, 1977; Jaitly et al. in J. Comput. Syst. Sci. 65:494–507, 2002; Tang et al. in J. Comput. Biol. 9:429–446, 2002). In this paper, we design a polynomial-time approximation scheme (PTAS) for the case where the size of the duplication block is 1. Our PTAS is faster than the previously best PTAS in Jaitly et al. (J. Comput. Syst. Sci. 65:494–507, 2002). For example, to achieve a ratio of 1.5, our PTAS takes O(n ⁵) time while the PTAS in Jaitly et al. (J. Comput. Syst. Sci. 65:494–507, 2002) takes O(n ¹¹) time. We also design a ratio-6 polynomial-time approximation algorithm for the case where the size of each duplication block is at most 2. This is the first polynomial-time approximation algorithm with a guaranteed ratio for this case. Part of work was done during a Z.-Z. Chen visit at City University of Hong Kong.     

Distributed scheduling algorithms for channel access in TDMA wireless mesh networks

In this paper, we have considered the distributed scheduling problem for channel access in TDMA wireless mesh networks. The problem is to assign time-slot(s) for nodes to access the channels, and it is guaranteed that nodes can communicate with all their one-hop neighbors in the assigned time-slot(s). And, the objective is to minimize the cycle length, i.e., the total number of different time-slots in one scheduling cycle. In single-channel ad hoc networks, the best known result for this problem is proved to be K ² in arbitrary graphs (Chlamtac and Pinter in IEEE Trans. Comput. C-36(6):729–737, 1987) and 25K in unit disk graphs () with K as the maximum node degree. There are multiple channels in wireless mesh networks, and different nodes can use different control channels to reduce congestion on the control channels. In this paper, we have considered two scheduling models for wireless mesh networks. The first model is that each node has two radios, and the scheduling is simultaneously done on the two radios. We have proved that the upper bound of the cycle length in arbitrary graphs can be 2K. The second model is that the time-slots are scheduled for the nodes regardless of the number of radios on them. In this case, we have proved that the upper bound can be (4K−2). We also have proposed greedy algorithms with different criterion. The basic idea of these algorithms is to organize the conflicting nodes by special criterion, such as node identification, node degree, the number of conflicting neighbors, etc. And, a node cannot be assigned to a time-slot(s) until all neighbor nodes, which have higher criterion and might conflict with the current node, are assigned time-slot(s) already. All these algorithms are fully distributed and easy to realize. Simulations are also done to verify the performance of these algorithms.     

Efficient parallel algorithms for computing all pair shortest paths in directed graphs

We present parallel algorithms for computing all pair shortest paths in directed graphs. Our algorithm has time complexityO(f(n)/p+I(n)logn) on the PRAM usingp processors, whereI(n) is logn on the EREW PRAM, log logn on the CCRW PRAM,f(n) iso(n ³). On the randomized CRCW PRAM we are able to achieve time complexityO(n ³/p+logn) usingp processors. A preliminary version of this paper was presented at the 4th Annual ACM Symposium on Parallel Algorithms and Architectures, June 1992. Support by NSF Grant CCR 90-20690 and PSC CUNY Awards #661340 and #662478.     

Efficient Parallel Graph Algorithms for Coarse-Grained Multicomputers and BSP

In this paper we present deterministic parallel algorithms for the coarse-grained multicomputer (CGM) and bulk synchronous parallel (BSP) models for solving the following well-known graph problems: (1) list ranking, (2) Euler tour construction in a tree, (3) computing the connected components and spanning forest, (4) lowest common ancestor preprocessing, (5) tree contraction and expression tree evaluation, (6) computing an ear decomposition or open ear decomposition, and (7) 2-edge connectivity and biconnectivity (testing and component computation). The algorithms require O(log p) communication rounds with linear sequential work per round (p = no. processors, N = total input size). Each processor creates, during the entire algorithm, messages of total size O(log (p) (N/p)) . The algorithms assume that the local memory per processor (i.e., N/p ) is larger than p ^ε , for some fixed ε > 0 . Our results imply BSP algorithms with O(log p) supersteps, O(g log (p) (N/p)) communication time, and O(log (p) (N/p)) local computation time. It is important to observe that the number of communication rounds/ supersteps obtained in this paper is independent of the problem size, and grows only logarithmically with respect to p . With growing problem size, only the sizes of the messages grow but the total number of messages remains unchanged. Due to the considerable protocol overhead associated with each message transmission, this is an important property. The result for Problem (1) is a considerable improvement over those previously reported. The algorithms for Problems (2)—(7) are the first practically relevant parallel algorithms for these standard graph problems. Received July 5, 2000; revised April 16, 2001.