Drift Detection Algorithm Research Articles

An assessment of the reliability of sea-ice motion and deformation retrieval using SAR images

AbstractThe local mass balance of sea ice is dependent on the advection of ice into and out of an area and on the deformation processes in that area. Sea-ice motion can be observed from space by synthetic aperture radar (SAR) and quantified by drift-detection algorithms. Due to the scarcity of field observations, it remains a challenging task to validate the resulting motion fields. We analyse the quality of sea-ice motion fields derived from SAR data, using an example dataset from the Weddell Sea region. We apply a quality indicator for sea-ice motion fields which is independent of field data and evaluate it with reference data obtained from visual analysis of the SAR images. Together with the motion field, sea-ice deformation can also be retrieved from SAR data. Similarly to ice motion, it is very difficult to obtain field data to evaluate the quality of the results. Based on a manually derived reference dataset, we introduce a method to validate the retrieved deformation rates. This procedure requires no additional field data. Our analysis shows that deformation rates derived from SAR data are consistent with results obtained from buoy analysis by previous studies.

Advances on Concept Drift Detection in Regression Tasks Using Social Networks Theory

Mining data streams is one of the main studies in machine learning area due to its application in many knowledge areas. One of the major challenges on mining data streams is concept drift, which requires the learner to discard the current concept and adapt to a new one. Ensemble-based drift detection algorithms have been used successfully to the classification task but usually maintain a fixed size ensemble of learners running the risk of needlessly spending processing time and memory. In this paper the authors present improvements to the Scale-free Network Regressor (SFNR), a dynamic ensemble-based method for regression that employs social networks theory. In order to detect concept drifts SFNR uses the Adaptive Window (ADWIN) algorithm. Results show improvements in accuracy, especially in concept drift situations and better performance compared to other state-of-the-art algorithms in both real and synthetic data.

A social approach for learning agents

In this paper, we propose a social approach for learning agents. In dynamic environments, smart agents should detect changes and adapt themselves, applying dynamic learning strategies and drift detection algorithms. Recent studies note that an ensemble of learners can be coordinated by simple protocols based on votes or weighted votes; however, they are not capable of determining the number of learners or the ensemble composition properly. Conversely, we show in this paper that Social Network Theory can provide the multi-agent learning community with sophisticated and well-founded reputation models that outperform well-known ensemble-based drift detection techniques, generating accurate and small ensembles of learning agents. Our approach is evaluated considering dynamic bilateral negotiation scenarios and benchmark databases, presenting statistically significant results that are better than those of other ensemble-based techniques.

NIMS-PL: A Cable-Driven Robot With Self-Calibration Capabilities

We present the Networked InfoMechanical System for Planar Translation, which is a novel two-degree-of-freedom (2-DOF) cable-driven robot with self-calibration and online drift-correction capabilities. This system is intended for actuated sensing applications in aquatic environments. The actuation redundancy resulting from in-plane translation driven by four cables results in an infinite set of tension distributions, thus requiring real-time computation of optimal tension distributions. To this end, we have implemented a highly efficient, iterative linear programming solver, which requires a very small number of iterations to converge to the optimal value. In addition, two novel self-calibration methods have been developed that leverage the robot's actuation redundancy. The first uses an incremental displacement, or jitter method, whereas the second uses variations in cable tensions to determine end-effector location. We also propose a novel least-squares drift-detection algorithm, which enables the robot to detect long-term drift. Combined with self-calibration capabilities, this drift-monitoring algorithm enables long-term autonomous operation. To verify the performance of our algorithms, we have performed extensive experiments in simulation and on a real system.

Novelty detection with application to data streams

This paper presents and evaluates an approach to novelty detection that addresses it as the problem of identifying novel concepts in a continuous learning scenario, as an extension to a single-class classification problem. OLINDDA, an OnLIne Novelty and Drift Detection Algorithm that implements this approach, uses efficient standard clustering algorithms to continuously generate candidate clusters among examples that were not explained by the current known concepts. Clusters complying with a validation criterion that takes cohesiveness and representativeness into account are initially identified as concepts. By merging similar concepts, OLINDDA may enhance the representation of some concepts as it advances toward its final goal of describing novel emerging concepts in an unsupervised way. The proposed approach is experimentally evaluated by the use of several measures taken throughout the learning process. Results show that it is capable of identifying novel concepts that are pure and correspond to real classes, disregarding unrepresentative clusters and outliers.

LEARNING DRIFTING NEGOTIATIONS

In this work, we propose the use of drift detection techniques for learning offer policies in multiissue, bilateral negotiation. Several works aiming to develop adaptive trading agents have been proposed. Such agents are capable of learning their competitors' utility values and functions, thereby obtaining better results in negotiation. However, the learning mechanisms generally used disregard possible changes in a competitor's offer/counter-offer policy. In that case, the agent's performance may decrease drastically. The agent then needs to restart the learning process, as the model previously learned is no longer valid. Drift detection techniques can be used to detect changes in the current offers model and quickly update it. In this work, we demonstrate with simulated data that drift detection algorithms can be used to build adaptive trading agents and offer a number of advantages over the techniques mostly used in this problem. The results obtained with the algorithm IB3 (instance-based) show that the agent's performance can be rapidly recovered even when changes interesting to the competitor are abrupt, moderate, or gradual.