Research Themes

Machine Learning＋Computational Algebra


Improving AI Accuracy and
Building Trustworthy Systems

Machine LearningComputational AlgebraDeep LearningTransformersLogical Reasoning Technologies

Mathematically Uncovering the Differences Between Human and AI Perception

I began researching AI and machine learning＊ because I was strongly interested in questions such as, “Do humans and AI see the same world?” and “If they differ, what exactly is different?” AI can recognize images and generate text in ways that seem human-like, yet its internal mechanisms are fundamentally different from ours.
For example, adding only a tiny amount of noise to an image that looks like nothing but a panda to humans can cause an AI to classify it as a gibbon. By analyzing such phenomena, I aim to mathematically reveal AI’s fragility and limitations. That curiosity became the starting point of my research.

Tips

Machine Learning
A technique in which computers learn patterns and regularities from large amounts of data. It is an important technology that improves AI accuracy and enables data-driven classification, prediction, and decision-making.


How Are Humans and AI Different?
Some images look like a panda to humans, but an AI may misclassify them as a “gibbon.” AI can be like an “alien”: it interprets the world in ways that differ from humans. Understanding these differences—and exploring how to use them—is also an important research area.

Seeking a Theory That Enables AI to Think for Itself and Produce “Solutions”

My background is in computational algebra and mathematical informatics—fields that study algorithms for efficiently solving equations and the logical structures behind computation. In recent years, AI has been supported by enormous amounts of data and computation, but mathematical grounding is indispensable for such computation to be carried out correctly and stably.
I wanted to describe the processes of machine learning and logical reasoning not merely as empirical rules (patterns), but in a form that can be explained mathematically. This motivation led me to connect machine learning with computational algebra.
Currently, I work on visualizing the basis for AI’s decisions and developing methods for “ambiguous yet rigorous matching” that can detect “similarity” across different images. In addition, I pursue two directions: “learning of algebra,” in which AI itself learns algebra, and “algebra for learning,” in which AI is used to discover new algebraic theory.
All of these connect to the question of how far AI can carry out “logical reasoning” like humans. Ultimately, I aim to build a theory that enables AI not merely to follow familiar “question” patterns and output “solutions,” but to understand and think for itself.

Building a “Trust Relationship” Between Humans and AI

What I want to convey to students through my research is that behind eye-catching applications lie quiet, fundamental theories. To make effective use of AI, it is important to understand how it works and to be able to explain why it behaves as it does.
In my classes, I often use a cooking analogy. Anyone can make a dish by following a recipe, but to change ingredients or seasonings—or to create an entirely new dish—you need both logic and experience. The same is true for AI: developing a reliable “palate” through fundamental theory is the first step toward new creation.
In the future, AI will become more deeply integrated into society. That is why it will become increasingly important to understand the differences between humans and AI and to build a relationship of trust.
My goal is not only to bring AI closer to humans, but to understand the differences and conduct research that makes use of them. I hope students will keep their curiosity alive and find their own “questions” to pursue.

Profile

Assoc. Prof. Hiroshi KERA

Assoc. Prof.  Hiroshi KERA

Associate Professor, Faculty of Informatics / Graduate School of Informatics, Chiba University. He completed the doctoral program in Information & Communication Engineering at The University of Tokyo in 2020. He then served as a Project Researcher at the Graduate School of Information Science and Technology, The University of Tokyo, an Assistant Professor at the Graduate School of Informatics, Chiba University, and a Visiting Researcher at Zuse Institute Berlin, before assuming his current position in 2025.

Addressing Social Issues Through Data Science

View Another Research