Po-Wei Harn
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Previous Organizations / Institutes
The organizations / institutes from the scholar's other articles in the past ten years
Research Subjects
Po-Wei Harn has a total of 9 co-authors. Here are the top co-authors by number of shared publications:
- Libo Sun
Computer Science and Software Engineering, Auburn University, Auburn, Alabama, USA
3 shared publications - Bo Hui
Computer science, The University of Tulsa, Tulsa, Oklahoma, USA
2 shared publications - Wei-Shinn Ku
Auburn University
2 shared publications - Min-Te Sun
National Central University
1 shared publication - Peixiong He
Department of Computer Science and Software Engineering, Auburn University, Auburn, AL 36849, USA
1 shared publication - Sai Deepthi Yeddula
Computer Science and Engineering, Oakland University, Rochester, Michigan, USA
1 shared publication
Latest Publications
- Open Access
- 13 July 2026
- Published by Association for Computing Machinery (ACM)
- p. 401-404
Evolving Maze Environments for Deep Q-Learning via Evolutionary Algorithms
- 13 July 2026
- Published by Association for Computing Machinery (ACM)
- p. 401-404
Open AccessEvolving Maze Environments for Deep Q-Learning via Evolutionary Algorithms
- Open Access
- 22 April 2026
- Published by Elsevier in SSRN Electronic Journal
Power-Law Temporal Scaling for Cost-Per-Purchase Advertising Prediction
Cost-per-purchase (CPP) digital advertising requires forecasting both purchase volume and advertising cost for campaigns whose durations span days to months, a heterogeneity that distorts cumulative targets and undermines standard regression. We propose a temporal-scaling pipeline that normalizes cumulative outcomes to a reference horizon via a power-law factor, $(T_{ref}/T_{actual})^{\alpha}$, integrated into a stacked ensemble of gradient-boosted and tree-based learners. The power-law form is derived rather than imposed: under a minimal scale-invariance assumption on the accumulation rate, the cumulative outcome scales as $T^{\alpha}$, the canonical scale-free candidate. Experiments on a proprietary dataset of 30{,}380 e-commerce campaigns show the pipeline improves RMSE most in fixed-horizon planning and in high-value segments. Cross-corpus validation on four public datasets (UCI, Olist, Instacart, Criteo) confirms applicability to duration-scalable targets and correct fallback to the baseline for binary or per-event targets, and a log-correlation diagnostic identifies when the transformation is appropriate.- 22 April 2026
- Published by Elsevier in SSRN Electronic Journal
Open AccessPower-Law Temporal Scaling for Cost-Per-Purchase Advertising Prediction
Cost-per-purchase (CPP) digital advertising requires forecasting both purchase volume and advertising cost for campaigns whose durations span days to months, a heterogeneity that distorts cumulative targets and undermines standard regression. We propose a temporal-scaling pipeline that normalizes cumulative outcomes to a reference horizon via a power-law factor, $(T_{ref}/T_{actual})^{\alpha}$, integrated into a stacked ensemble of gradient-boosted and tree-based learners. The power-law form is derived rather than imposed: under a minimal scale-invariance assumption on the accumulation rate, the cumulative outcome scales as $T^{\alpha}$, the canonical scale-free candidate. Experiments on a proprietary dataset of 30{,}380 e-commerce campaigns show the pipeline improves RMSE most in fixed-horizon planning and in high-value segments. Cross-corpus validation on four public datasets (UCI, Olist, Instacart, Criteo) confirms applicability to duration-scalable targets and correct fallback to the baseline for binary or per-event targets, and a log-correlation diagnostic identifies when the transformation is appropriate. - Open Access
- 13 July 2025
- Published by Association for Computing Machinery (ACM)
- p. 377-385
Evolutionary Quadtree Pooling for Convolutional Neural Networks
- 13 July 2025
- Published by Association for Computing Machinery (ACM)
- p. 377-385
Open AccessEvolutionary Quadtree Pooling for Convolutional Neural Networks
- Open Access
- 15 July 2023
- Published by Association for Computing Machinery (ACM)
- p. 583-586
A Novel Quadtree-Based Genetic Programming Search for Searchable Encryption Optimization
The encoding method of a searchable encryption can significantly impact the performance of a location-based alert system. While there were attempts to design searchable encryption manually, Gray Encoding is considered the most preferable method. However, if the alert zones are scattered unevenly, Gray Encoding fails to achieve token aggregation. In this research, a novel Quadtree-based Genetic Programming (Quadtree-GP) is proposed to iteratively identify superior searchable encryption candidates for the location-based alert system. Quadtree-GP can be effectively applied on customized requirements and different grid maps. Extensive experimental results show that Quadtree-GP is able to find searchable encryption candidates that outperform GP search, random search, and the baseline Gray Encoding in terms of user response time, token remaining percentage, and execution time.
- 15 July 2023
- Published by Association for Computing Machinery (ACM)
- p. 583-586
Open AccessA Novel Quadtree-Based Genetic Programming Search for Searchable Encryption Optimization
The encoding method of a searchable encryption can significantly impact the performance of a location-based alert system. While there were attempts to design searchable encryption manually, Gray Encoding is considered the most preferable method. However, if the alert zones are scattered unevenly, Gray Encoding fails to achieve token aggregation. In this research, a novel Quadtree-based Genetic Programming (Quadtree-GP) is proposed to iteratively identify superior searchable encryption candidates for the location-based alert system. Quadtree-GP can be effectively applied on customized requirements and different grid maps. Extensive experimental results show that Quadtree-GP is able to find searchable encryption candidates that outperform GP search, random search, and the baseline Gray Encoding in terms of user response time, token remaining percentage, and execution time.