ปพนธีร์ แย้มโชติ (เต้ย)
Faculty of ICT, Mahidol University
Teaching Mathematics, Computer Science and Programming:
including thier application such as Data Science, Machine Learning, Artificial Intelligence
Preprint (already presented in conference MLMI 2025, it is in process of public in proceeding)
Click-through rate (CTR) prediction Click-through rate (CTR) predic-tion is essential in online advertising and recommendation systems, affecting user engagement and revenue. Traditional methods like linear models and deep learn-ing often struggle to effectively capture complex, high-order feature interactions. Graph Neural Networks (GNNs) offer improved performance by modeling these intricate relationships but face challenges regarding computational efficiency and interpretability, affecting scalability. We propose a new static graph-based GNN architecture for CTR prediction, incorporating advanced feature selection to en-hance representation and reduce complexity. Our model employs a bilinear inter-action mechanism to efficiently prioritize high-order interactions. Experiments on benchmark datasets show our approach achieves near state-of-the-art results, surpassing traditional deep learning models.
Preprint (already submitted, now it is posted on arxiv)
Feature interaction is crucial in predictive machine learning models, as it captures the relationships between features that influence model performance. In this work, we focus on pairwise interactions and investigate their importance in constructing feature graphs for Graph Neural Networks (GNNs). We leverage existing GNN models and tools to explore the relationship between feature graph structures and their effectiveness in modeling interactions. Through experiments on synthesized datasets, we uncover that edges between interacting features are important for enabling GNNs to model feature interactions effectively. We also observe that including non-interaction edges can act as noise, degrading model performance. Furthermore, we provide theoretical support for sparse feature graph selection using the Minimum Description Length (MDL) principle. We prove that feature graphs retaining only necessary interaction edges yield a more efficient and interpretable representation than complete graphs, aligning with Occam's Razor. Our findings offer both theoretical insights and practical guidelines for designing feature graphs that improve the performance and interpretability of GNN models.
Journal: SAU JOURNAL OF SCIENCE & TECHNOLOGY Vol. 10 No. 2 (2024): July - December 2024
This research paper investigates the presence of conscious bias in LinkedIn job postings within Thailand, focusing on gendered language and its impact on recruitment practices. Conscious bias, manifested through explicit preferences in job descriptions, can perpetuate inequality in the labor market. By analyzing job postings across ten in-demand roles in the tech and data-driven sectors, this study identifies patterns of gendered language and explores their implications for gender diversity in recruitment. Using a combination of Natural Language Processing (NLP) tools and statistical analysis, the research reveals that masculine-coded language is more prevalent in technical roles, potentially discouraging female applicants and contributing to the underrepresentation of women in leadership and senior positions. Additionally, the paper highlights the role of educational requirements as a barrier to inclusivity, particularly in fields that favor candidates with advanced degrees. The findings underscore the need for more inclusive job postings and greater enforcement of anti-discrimination policies to promote equality in the Thai labor market.
Conference Proceedings: Annual Pure and Applied Mathematics Conference 2020
Let R be an expansion of the real field (R, +, ·). For each f : R n → R, we denote the set of zeroes of f by Z(f) = {x ∈ R n : f (x) = 0}. Let f, g : U → R be functions definable in R (where U ⊆ R n), and p ∈ N. We prove that if A is a compact nonempty subset of U and Z(f) ⊆ Z(g), then there exists θ : R → R such that θ is C p , strictly increasing, surjective, and p-flat at 0 (that is, all derivatives of order at most p vanish); and |θ(g(x))| ≤ |f (x)| for all x ∈ A. In addition, we give an application of this result.
(Client Training: no pictures due to confidential agreements)
Train interns of DoHome to understand and apply machine learning techniques by using Azure Machine Learning Studio.
Train corporate staffs of Siam City Cement Public Company Limited (INSEE) to understand and apply machine learning techniques for demand prediction of cement and mortar products via no-code tool in Azure Machine Learning Studio.
Train corporate staffs of DoHome to understand and apply machine learning techniques by using Azure Machine Learning Studio.
Developed a machine learning model for recommendation of orders including item to suggest and amount for particular agents. As well, optimization model for full truck load delivery to maximize the capability of transportation.
Developed a user-friendly Streamlit-based application integrated with Celery for job management.
Develop ETL process migrated from on premise database of Retail chain of a petroleum and energy business in Microsoft Fabric
Investigate the potentiality of applying weak supervision techniques to improve prediction accuracy in uncertainty scenarios for manufacturing data of Western Digital.
Implemented Python optimization algorithms to reduce routine audit report processing time from 5-10 minutes per job to 1-2 seconds.
in audit
in audit
in manufacturing industry
in manufacturing industry / for data science
data science, machine learning and deep learning, mathematics and statistics
teaching of data structures and algorithm
operating system, competitive programming
competitive programming
zmathematics, machine learning, signal processing
artificial intelligence, web application
data management, web application
statistical analysis, machine learning
functional programming teaching
Make me can learn any programming language faster
Scikit learn
Pytorch, TensorFlow and Keras, Apache Mxnet, Pytorch Lightning
Pytorch Geometric
SpaCy, LangChain
Docker
Google Apps Script, Python Django
Pandas, Apache Hadoop, Power BI, Excel ElasticSearch
MySQL, MS Access
Orange (machine learning), Scratch (programming for kids)
Pyomo
Vscode, Linux
