A5066382032- Machine Learning and Algorithms
- Imbalanced Data Classification Techniques
- Energy Load and Power Forecasting
- Machine Learning and Data Classification
- Forecasting Techniques and Applications
- Mobile Crowdsensing and Crowdsourcing
- Stock Market Forecasting Methods
- Data Stream Mining Techniques
- Data Quality and Management
- Anomaly Detection Techniques and Applications
- Web Data Mining and Analysis
- Complex Network Analysis Techniques
- Data Mining Algorithms and Applications
- Algorithms and Data Compression
- Air Traffic Management and Optimization
- Auction Theory and Applications
- Topic Modeling
- Advanced Statistical Methods and Models
- Gaussian Processes and Bayesian Inference
- Fault Detection and Control Systems
- Oil and Gas Production Techniques
- Reliability and Maintenance Optimization
- Solar Radiation and Photovoltaics
- Point processes and geometric inequalities
- Aviation Industry Analysis and Trends
Middle East Technical University
2015-2024
Istanbul Bilgi University
2022
United States Army Corps of Engineers
2021
Geospatial Research (United Kingdom)
2021
Massachusetts Institute of Technology
2010-2019
University of Washington
2015
Pennsylvania State University
2006-2009
This paper is concerned with the class imbalance problem which has been known to hinder learning performance of classification algorithms. The occurs when there are significantly less number observations target concept. Various real-world tasks, such as medical diagnosis, text categorization and fraud detection suffer from this phenomenon. standard machine algorithms yield better prediction balanced datasets. In paper, we demonstrate that active capable solving by providing learner more...
The class imbalance problem has been known to hinder the learning performance of classification algorithms. Various real-world tasks such as text categorization suffer from this phenomenon. We demonstrate that active is capable solving problem.
In this paper, we propose a nonconvex online Support Vector Machine (SVM) algorithm (LASVM-NC) based on the Ramp Loss, which has strong ability of suppressing influence outliers. Then, again in learning setting, an outlier filtering mechanism (LASVM-I) approximating behavior convex optimization. These two algorithms are built upon another novel SVM (LASVM-G) that is capable generating accurate intermediate models its iterative steps by leveraging duality gap. We present experimental results...
Reactive point processes (RPPs) are a new statistical model designed for predicting discrete events in time based on past history. RPPs were developed to handle an important problem within the domain of electrical grid reliability: short-term prediction failures ("manhole events"), including outages, fires, explosions and smoking manholes, which can cause threats public safety reliability service cities. incorporate self-exciting, self-regulating saturating components. The self-excitement...
Human level recall performance in detecting breast cancer considering microcalcifications from mammograms has a value between 74.5% and 92.3%. In this research, we approach to microcalcification classification problem using convolutional neural networks along with various preprocessing methods such as contrast scaling, dilation, cropping etc. decision fusion ensemble of networks. Various experiments on Digital Database for Screening Mammography dataset showed that poses great importance the...
The problem of "approximating the crowd" is that estimating crowd's majority opinion by querying only a subset it. Algorithms approximate crowd can intelligently stretch limited budget for crowdsourcing task. We present an algorithm, "CrowdSense," works in online fashion to dynamically sample subsets labelers based on exploration/exploitation criterion. algorithm produces weighted combination labelers' votes approximates opinion.
We summarize the first major effort to use analytics for preemptive maintenance and repair of an electrical distribution network. This is a large-scale multiyear between scientists students at Columbia University Massachusetts Institute Technology engineers from Consolidated Edison Company New York (Con Edison), which operates world’s oldest largest underground system. Con Edison’s programs are less than decade old made more effective with developing alongside them. Some data we used our...
We propose a generative model based on latent Dirichlet allocation for mining distinct topics in document collections by integrating the temporal ordering of documents into process. The collection is divided time segments where discovered each segment propagated to influence topic discovery subsequent segments. conduct experiments academic papers from CiteSeer repository. augment text corpus with addition user queries and tags integrate citation graph boost weight topical terms. experiment...
High‐accuracy predictions of future solar power generations are important for monitoring, maintenance, dispatching, and scheduling. The goal this study is to create a forecasting workflow that increases prediction accuracy independent the machine learning method has minimal computational requirements. proposed trend decomposition incorporates irradiance seasonal features as exogenous inputs. In order extract linear part data, moving average filter used. nonlinear (stable) component time...
Previous chapter Next Full AccessProceedings Proceedings of the 2007 SIAM International Conference on Data Mining (SDM)Efficient Multiclass Boosting Classification with Active LearningJian Huang, Seyda Ertekin, Yang Song, Hongyuan Zha, and C. Lee GilesJian Gilespp.297 - 308Chapter DOI:https://doi.org/10.1137/1.9781611972771.27PDFBibTexSections ToolsAdd to favoritesExport CitationTrack CitationsEmail SectionsAboutAbstract We propose a novel multiclass classification algorithm Gentle Adaptive...
In this study, a novel approach is demonstrated for converting calorimeter images from fast simulations to those akin comprehensive full simulations, utilizing conditional Generative Adversarial Networks (GANs). The concept of pix2pix tailored CALPAGAN, where serve as the basis(condition) generating outputs that closely resemble detailed simulations. findings indicate strong correlation between generated and especially in terms key observables like jet transverse momentum distribution, mass,...
Abstract Modern real-world control problems call for continuous domains and robust, sample efficient explainable frameworks. We are presenting a framework recursively composing skills to solve compositional progressively complex tasks. The promotes reuse of skills, as result quick adaptability new decision tree can be observed, providing insight into the agents’ behavior. Furthermore, transferred, modified or trained independently, which simplify reward shaping increase training speeds...
Identification of distinct clusters documents in text collections has traditionally been addressed by making the assumption that data instances can only be represented homogeneous and uniform features. Many real-world data, on other hand, comprise multiple types heterogeneous interrelated components, such as web pages hyperlinks, online scientific publications authors publication venues to name a few. In this paper, we present KSVMeans, clustering algorithm for multi-type datasets integrates...