Jeonghwan Gwak

Novel dynamic diversity controlling EAs for coevolving optimal negotiation strategies

Abstract Finding optimal strategies for negotiation with incomplete information is a challenging issue in agent-based automated negotiation research. Although there are some previous works on finding the strategies through coevolutionary learning using evolutionary algorithms ( EAs ), their coevolving strategies tend to converge to non-global optima (which bring about ineffective negotiation outcomes for participating agents) due to biased coevolution and failures in coevolution. To cope with these drawbacks, this work introduces and compares novel genetic algorithms ( GAs ) and estimation of distribution algorithms ( EDAs ) that have additional capability of dynamic diversity control: (1) the dynamic diversity controlling GA ( D 2 C-GA ), (2) the dynamic diversity controlling EDA ( D 2 C-EDA ), (3) the improved D 2 C-GA ( ID 2 C-GA ) and (4) the improved D 2 C-EDA ( ID 2 C-EDA ). While D 2 C-GA and D 2 C-EDA adopt the novel diversification and refinement ( DR ) procedure, ID 2 C-GA and ID 2 C-EDA adopt the modified and enhanced DR ( mDR ) procedure with two additional local heuristics population repair and local neighborhood search . An extensive series of experiments were carried out to compare and evaluate the performance of the simple GA ( S-GA ), the simple EDA ( S-EDA ), and the novel GAs and EDAs in coevolving effective negotiation strategies of two self-interested negotiation agents for their various deadline combinations. Favorable empirical results showed that (i) ID 2 C-EDA could coevolve (near-)optimal negotiation strategies for all the considered cases due to its good generalization performance and (ii) it also generally outperformed S-GA , S-EDA , D 2 C-GA , D 2 C-EDA and ID 2 C-GA in terms of solution accuracy, coevolutionary search capability and average coevolution restart ratio. Interestingly, it was also found that the coevolution performance of ID 2 C-GA and ID 2 C-EDA is complementary in that ID 2 C-GA and ID 2 C-EDA generally achieved better results in the cases of equal and different deadlines, respectively.

An augmented EDA with dynamic diversity control and local neighborhood search for coevolution of optimal negotiation strategies

In this paper, we present an estimation of distribution algorithm (EDA) augmented with enhanced dynamic diversity controlling and local improvement methods to solve competitive coevolution problems for agent-based automated negotiations. Since optimal negotiation strategies ensure that interacting agents negotiate optimally, finding such strategies—particularly, for the agents having incomplete information about their opponents—is an important and challenging issue to support agent-based automated negotiation systems. To address this issue, we consider the problem of finding optimal negotiation strategies for a bilateral negotiation between self-interested agents with incomplete information through an EDA-based coevolution mechanism. Due to the competitive nature of the agents, EDAs should be able to deal with competitive coevolution based on two asymmetric populations each consisting of self-interested agents. However, finding optimal negotiation solutions via coevolutionary learning using conventional EDAs is difficult because the EDAs suffer from premature convergence and their search capability deteriorates during coevolution. To solve these problems, even though we have previously devised the dynamic diversity controlling EDA (D2C-EDA), which is mainly characterized by a diversification and refinement (DR) procedure, D2C-EDA suffers from the population reinitialization problem that leads to a computational overhead. To reduce the computational overhead and to achieve further improvements in terms of solution accuracy, we have devised an improved D2C-EDA (ID2C-EDA) by adopting an enhanced DR procedure and a local neighborhood search (LNS) method. Favorable empirical results support the effectiveness of the proposed ID2C-EDA compared to conventional and the other proposed EDAs. Furthermore, ID2C-EDA finds solutions very close to the optimum.

Diagnosis of Alzheimer’s Disease Based on Structural MRI Images Using a Regularized Extreme Learning Machine and PCA Features

Alzheimers disease (AD) is a progressive, neurodegenerative brain disorder that attacks neurotransmitters, brain cells, and nerves, affecting brain functions, memory, and behaviors and then finally causing dementia on elderly people. Despite its significance, there is currently no cure for it. However, there are medicines available on prescription that can help delay the progress of the condition. Thus, early diagnosis of AD is essential for patient care and relevant researches. Major challenges in proper diagnosis of AD using existing classification schemes are the availability of a smaller number of training samples and the larger number of possible feature representations. In this paper, we present and compare AD diagnosis approaches using structural magnetic resonance (sMR) images to discriminate AD, mild cognitive impairment (MCI), and healthy control (HC) subjects using a support vector machine (SVM), an import vector machine (IVM), and a regularized extreme learning machine (RELM). The greedy score-based feature selection technique is employed to select important feature vectors. In addition, a kernel-based discriminative approach is adopted to deal with complex data distributions. We compare the performance of these classifiers for volumetric sMR image data from Alzheimers disease neuroimaging initiative (ADNI) datasets. Experiments on the ADNI datasets showed that RELM with the feature selection approach can significantly improve classification accuracy of AD from MCI and HC subjects.

Coevolving Negotiation Strategies for P-S-Optimizing Agents

In this paper, we consider the negotiation between two competitive agents that consider both time and cost criteria. Therefore, the negotiation agents are designed to not only optimize price utility but also be successful in optimizing (negotiation) speed utility. To this end, the objective of this work is to find effective strategies for the negotiation. The strategies are coevolved through an evolutionary learning process using two different evolutionary algorithms (EAs)—a genetic algorithm (GA) and an estimation of distribution algorithm (EDA). We present an empirical comparison of GA and EDA in coevolving negotiation strategies with different preference criteria in optimizing the price and (negotiation) speed. The experimental results show that both EAs are successful in finding good solutions with respect to both the price-optimizing (P-Optimizing) and the speed-optimizing (S-Optimizing) negotiation. However, both EAs are not effective in the negotiation for the concurrent optimization of the price and speed (P-S-Optimizing negotiation). This is because in some cases, the original fitness function cannot characterize the difference among P-Optimizing, S-Optimizing, and P-S-Optimizing solutions. Hence, this paper proposes a new fitness function that can better differentiate among the P-Optimizing, S-Optimizing, and P-S-Optimizing solutions. The experiments showed that the EAs using the proposed fitness function can coevolve effective strategies for the exact P-S-Optimizing negotiation.

Multi-Object Tracking Through Learning Relational Appearance Features and Motion Patterns

Absract Multi-object tracking (MOT) is to simultaneously track multiple targets, e.g., pedestrians in this work, through locating them and maintaining their identities to make their individual trajectories. Despite of recent advances in object detection, MOT based on the tracking-by-detection principle is a still yet challenging and difficult task in complex and crowded conditions. For example, due to occlusion, missed object detection, and frequent entering and leaving of object in a scene, tracking failures such as identity switches and trajectory fragmentation can often occur. To tackle the issues, a new data association approach, namely, the relational appearance features and motion patterns learning (RAFMPL)-based data association, is proposed for facilitating MOT. In RAFMPL-MOT, the proposed relational features-based appearance model is different from conventional approaches in that it generates tracklets based on relational information by selecting one reference object and utilizing the feature differences between the reference object and the other objects. In addition, the motion patterns learning-based motion model enables linear and nonlinear confident motions patterns to be considered in data association. The proposed approach can effectively cover the key difficulties of MOT. In particular, using RAFMPL-MOT, it is possible to assign the same ID for the object that has been disappeared (even for moderately long period) and then is reappeared in the scene more robustly. Further, it also improves its robustness for occlusion problems frequently occurring in real situations. The experimental results show that the RAFMPL-MOT could generally achieve outperformance compared to the existing competitive MOT approaches.

Bolstering efficient SSGAs based on an ensemble of probabilistic variable-wise crossover strategies

A crossover operator in genetic algorithms (GAs) plays an essential role as the main search operator to breed offspring by exchanging information between individuals. Although different types of crossover operators have been developed for real-coded GAs (RCGAs), there has been very little research on combining different crossover operators to build more effective and efficient RCGAs. In this work, we propose new steady-state generation alternation-based RCGAs (SSGAs) ameliorated with (i) an ensemble of different probabilistic variable-wise crossover strategies, which is realized by the corresponding parallel populations, to utilize synergetic and complementary effect with their efficient operations, and (ii) efficient operation at each evolution step to obtain further performance enhancement. To investigate the performance of this ensemble with respect to search abilities and computation time, we compare the proposed algorithms against various SSGAs when running 27 benchmark functions. Empirical studies showed that the proposed algorithms exhibit better performance than the contestant SSGAs on these functions. Moreover, a comparison with the state-of-the-art evolutionary algorithms on eight difficult benchmark functions clearly demonstrated outperformance of the proposed algorithms.

Conditional Random Field (CRF)-Boosting: Constructing a Robust Online Hybrid Boosting Multiple Object Tracker Facilitated by CRF Learning

Due to the reasonably acceptable performance of state-of-the-art object detectors, tracking-by-detection is a standard strategy for visual multi-object tracking (MOT). In particular, online MOT is more demanding due to its diverse applications in time-critical situations. A main issue of realizing online MOT is how to associate noisy object detection results on a new frame with previously being tracked objects. In this work, we propose a multi-object tracker method called CRF-boosting which utilizes a hybrid data association method based on online hybrid boosting facilitated by a conditional random field (CRF) for establishing online MOT. For data association, learned CRF is used to generate reliable low-level tracklets and then these are used as the input of the hybrid boosting. To do so, while existing data association methods based on boosting algorithms have the necessity of training data having ground truth information to improve robustness, CRF-boosting ensures sufficient robustness without such information due to the synergetic cascaded learning procedure. Further, a hierarchical feature association framework is adopted to further improve MOT accuracy. From experimental results on public datasets, we could conclude that the benefit of proposed hybrid approach compared to the other competitive MOT systems is noticeable.

Feature flow-based abnormal event detection using a scene-adaptive cuboid determination method

As closed circuit television which had been used only for surveillance or identification has developed rapidly the research on intelligent surveillance systems is getting increased interest. Above all, abnormal event detection is becoming an essential part of surveillance systems by detecting or identifying actions or situations which are not commonly occurred in general. In this work, we propose an abnormal event detection method using trajectory modeling with an automatic scene-adaptive cuboid determination scheme. First, we constructed a human appearance model to determine the human size without using any detection method. Then, HOG feature extracted from human images which is the predetermined input is used to construct a human appearance model. We applied a background subtraction to input datasets and then compared HOG feature extracted from the bounding box of the foreground with the human appearance model. The human size is determined by the size of the foreground bounding box with the highest similarity. With the ratio obtained through the experiments, the cuboid size is calculated according to the human size and histogram of oriented tracklets model is constructed by the cuboid size. We used the UCSD dataset to validate the proposed approach. From the experimental results, we verified the significance of the proposed AED method adopting the automatic scene-adaptive cuboid size determination scheme.

Object recognition using depth information of a consumer depth camera

Many computer vision applications adopting consumer depth cameras have recently received much attention due to the availability at low prices and the potential benefits to provide more useful information, which can result in a higher accuracy (e.g., for object recognition). In this work, to address the problem of drinking activity recognition in vision-based Ambient Assisted Living by using depth information, we propose a novel pre-processing scheme on extracting useful depth information of a target object, such as a mug, from a histogram of a depth image. By analyzing the histogram, the specific area or shape of an object can be localized. Then, the extracted depth information that is mostly related to the object is used for the object recognition. In the experiments, we demonstrated that the proposed approach can achieve a higher accuracy than the existing methods in recognizing an object.

An incremental learning approach for restricted boltzmann machines

Determination of model complexity is a challenging issue to solve computer vision problems using restricted boltzmann machines (RBMs). Many algorithms for feature learning depend on cross-validation or empirical methods to optimize the number of features. In this work, we propose an learning algorithm to find the optimal model complexity for the RBMs by incrementing the hidden layer. The proposed algorithm is composed of two processes: 1) determining incrementation necessity of neurons and 2) computing the number of additional features for the increment. Specifically, the proposed algorithm uses a normalized reconstruction error in order to determine incrementation necessity and prevent unnecessary increment for the number of features during training. Our experimental results demonstrated that the proposed algorithm converges to the optimal number of features in a single layer RBMs. In the classification results, our model could outperform the non-incremental RBM.