Tokyo, June 19, 2026 – Ricoh today announced that its paper on technology for developing highly reliable AI models with limited data has been accepted as a poster presentation at the International Joint Conference on Neural Networks (IJCNN) 2026, one of the major international conferences in the field of neural networks, a fundamental technology for artificial intelligence (AI). This technology allows AI systems to recognize when they cannot reliably predict an answer, helping to improve AI reliability in real-world applications where training data is limited.
The conference will be held in the Netherlands from June 21 to 26, 2026, and is co-sponsored by the Institute of Electrical and Electronics Engineers (IEEE) and the International Neural Network Society (INNS).
AI adoption continues to expand across a wide range of industries, but obtaining sufficient training data remains a challenge. Additionally, AI systems may behave as if they know the correct answer even when presented with unfamiliar or never-before-seen data, raising concerns about the reliability of AI decision-making. In practical applications, there is a growing need for AI that can not only make accurate decisions with limited data, but also be able to recognize when the answer is unknown.
To address these challenges, this paper proposes a method that combines a Bayesian machine learning model that can evaluate the reliability of predictions with Contrastive Language-Image Pre-training (CLIP), a multimodal underlying model that captures the relationship between images and text. This method quantitatively estimates prediction uncertainty by evaluating the similarity between images and text using different criteria. This allows the AI system to recognize that it cannot reliably predict an answer when given input it has never seen before. Moreover, by adopting a training-free optimization approach, our method reduces the need for additional training during deployment and enables rapid implementation in a wide range of practical applications.
This paper was selected based on its proven utility in improving performance using existing multimodal underlying models with minimal additional training. It was also recognized for its ability to process unprecedented data and maintain stable performance under a variety of conditions. This technology is expected to improve the reliability of AI and expand its scope of application in fields where avoiding misjudgments is important, such as visual inspection in the manufacturing industry and inspection of equipment and infrastructure.
Ricoh will continue to leverage its neural network expertise to accelerate research and development and further strengthen the technology needed to rapidly develop and deploy reliable AI, even in environments with limited training data. Through these initiatives, Ricoh aims to provide reliable AI services tailored to customers’ industries and business needs, create new value, and support job satisfaction.
