Trustworthy Intelligent System Design Laboratory

About Laboratory

Our research agenda spans from developing intelligent systems to designing and implementing measures to ensure societal acceptance of systems.

Develop intelligent systems

As the complexity of information processing tasks increases, the criteria for evaluation become more varied. For basic arithmetic operations, the evaluation is simply based on whether the answer is correct or incorrect. However, for more complex tasks, the evaluation is based on probabilities, such as the percentage of correct answers from past data for validation. It’s important to note that this percentage cannot be confirmed until the future. The ability to produce accurate outputs for unknown inputs is referred to as robustness.

In the field of AI, researchers are working on enhancing the explainability and interpretability of results. Generative AI tasks, which have gained significant attention in recent years, are evaluated based on validity instead of correct answer rates due to the absence of a single correct answer. When developing new intelligent systems, evaluation criteria should be established concurrently.

Design a framework in which the system is trusted

In order for a system to be trusted, it needs to undergo various performance and property verifications. It encompasses the concepts of durability, availability, fault-tolerant, error-tolerant, correctness, reliability, dependability, explainability, responsibility, traceability, dependability, explainability, responsibility, traceability, consistency, integrity, confidentiality, social acceptability, feasibility, possibilities, verifiability, validity, and so on. By properly evaluating these factors and disclosing the results, the created system can earn trust from society.

It is important to ensure that evaluation results are communicated in a way that is understandable to many people. Valid arguments should not be overshadowed by unproven falsehoods. When designing an advanced intelligent system, it is crucial to establish a system that can be verified by anyone at any time.

A chiral aperiodic monotile as a motif

The background image is inspired by the shape “Spectres,” discovered by David Smith et al. Spectres are shapes that can fill a plane with only one type of shape, and only in aperiodic fills. It expresses our intention to address a wide variety of issues without resorting to simple repetition.

Members

We open the lab session on Monday evening with the Media Systems Lab in the 2024 academic year.

Master's degree enrollment in April 2023

Yan Ningxin

Shotaro Tanigawa

Tsubasa Kinoshita

Master's degree enrollment in September 2023

Yu Hyunmin

Master's degree enrollment in April 2024

Atsuko Utsumi

Shin Soonnam

Li Jinxuan

Master's degree enrollment in September 2024

Kim Eunmi

Siregar Josua Aldivio

Shoji Harata