Reading time ~4 minutes
Applied machine learning is the practice of leveraging machine learning techniques to solve complex, domain-specific problems across various fields of science and industry. These techniques can be broadly categorized into supervised learning, where models are trained on labeled data to make predictions, and unsupervised learning, which identifies patterns and structures in unlabeled data. By applying these methods effectively, we can uncover new insights, automate processes, and enhance decision-making in diverse disciplines.
The papers below are a selection; the full list is on my Publications page.
My Research and Approach
My work focuses on bridging machine learning with other scientific domains, fostering interdisciplinary collaborations that push the boundaries of both fields. Rather than viewing machine learning as an isolated discipline, I see it as a powerful tool that, when carefully applied, can unlock new discoveries in areas ranging from archaeology and digital archives to sensor data analysis and beyond.
Machine Learning for Archaeological Analysis
One of my research areas involves applying deep learning techniques to archaeological studies, particularly for the non-invasive analysis of pottery fragments. As part of the Deep Research for Akkon project, I use convolutional neural networks (CNNs) to classify X-ray images of archaeological potsherds from the Late Jomon period (ca. 4000 to 3200 aBP). In Japan, recent research has shown that ceramic vessels from this era contain impressions of plant seeds, providing valuable insights into ancient agricultural practices and the diffusion of rice cultivation. Traditionally, archaeologists analyze these impressions by visually inspecting X-ray images, but this approach is often inconclusive, leading to the use of invasive techniques that can damage fragile artifacts.
My work addresses this challenge by developing deep learning models that can automatically classify X-ray images of pottery fragments, significantly improving the accuracy and efficiency of the analysis. Our approach has achieved a 90% classification rate across a dataset of 1036 images spanning seven distinct classes. By integrating AI into archaeological workflows, this research helps preserve artifacts while enhancing the ability of researchers to extract meaningful insights from historical materials.
Selected publications
- “Classification of unexposed potsherd cavities by using deep learning” — Journal of Archaeological Science: Reports, 2023
- “Improving the Classification of Unexposed Potsherd Cavities by Means of Preprocessing” — Information, 2024
Automatic Image Tagging for Disaster Digital Archives
Another key area of my research focuses on automatic image tagging for disaster digital archives. Disaster archives serve as crucial resources for documenting and analyzing past natural disasters, but managing these vast collections effectively requires comprehensive and accurate tagging of images. Traditional machine-learning-based tagging models often fail to extract the detailed, disaster-specific information needed for meaningful retrieval and analysis.
To address this challenge, I apply generative AI techniques to improve disaster image tagging. By generating detailed descriptions of images and extracting structured tags from them, my approach allows for more nuanced and context-aware tagging compared to conventional methods. Additionally, by incorporating prior knowledge that the images are disaster-related directly into the prompts, we can achieve more specialized tagging that enhances searchability and usability for researchers, policymakers, and responders. This work has been applied to real-world test cases, including images from the 2011 Tohoku Earthquake and the 2016 Kumamoto Earthquake, demonstrating its effectiveness in generating disaster-relevant metadata.
Selected publications
- “Disaster Image Tagging Using Generative AI for Digital Archives” — JCDL 2024 (Vannevar Bush Best Paper Award, Third Place)
- “Semi-automated Disaster Image Tagging While Protecting Privacy: A Case Study” — DEXA 2024
Natural Language Processing, Text Mining, and Recommendation
Much of my work sits on the language side of applied ML: making models more robust on scarce or imbalanced text, and turning unstructured text into structure that people can act on. This includes text augmentation techniques that improve aspect-based sentiment classification under class imbalance, attribution-guided augmentation, and knowledge-graph-based recommendation deployed with a local news organization.
Selected publications
- “TAWC: Text Augmentation with Word Contributions for Imbalance Aspect-Based Sentiment Classification” — Applied Sciences, 2024
- “Text Augmentation Based on Integrated Gradients Attribute Score for Aspect-based Sentiment Analysis” — BigComp 2023
- “Knowledge Graphs for News Recommendation in a Local News Organization” — 2023
Imbalanced Learning and Time-Series Imputation
Real-world datasets are rarely clean or balanced. A recurring thread of my research develops principled ways to handle skewed class distributions and missing values—including meta-learning to recommend the right imputation method for a given time series, noise-free sampling for imbalanced classification, and sampling frameworks for overlapped classes.
Selected publications
- “Meta-learning for vessel time series data imputation method recommendation” — Expert Systems with Applications, 2024
- “Noise-free sampling with majority framework for an imbalanced classification problem” — Knowledge and Information Systems, 2024
- “Two-Stage Sampling: A Framework for Imbalanced Classification With Overlapped Classes” — IEEE BigData 2022
Broader Interests and Collaboration
Beyond these domains, my broader interests lie in applying AI to real-world challenges, particularly in fields that traditionally rely on manual analysis. Whether in computer vision, natural language processing, or large-scale data integration, I am interested in finding ways to reduce repetitive workloads while improving the depth and precision of insights.
I am always looking for motivated students and collaborators who are interested in applied machine learning, especially in interdisciplinary settings. If you share these interests and would like to work on projects at the intersection of AI and other scientific domains, feel free to reach out!
Questions?
Have some question about it? You can send me an email