Panasonic Develops New AI Technology to Improve Model Reliability

Panasonic R&D and Panasonic Holdings have developed a technology to improve the reliability of AI models by detecting unknown objects. The new flow-based generative model, FlowEneDet, enhances recognition accuracy by

Panasonic Develops New AI Technology to Improve Model Reliability

Panasonic R&D Company of America and Panasonic Holdings Corporation have announced the development of a new technology that aims to increase the reliability of AI models. The technology focuses on detecting objects that the AI model has not learned, making them essentially unrecognizable as unknown objects. While image recognition AI models are capable of accurately recognizing pre-trained objects, it is challenging to train them with all objects in the real world. This often leads to AI models facing unknown objects in their usage environment, which can result in unexpected malfunctions.

One of the main challenges with AI models is their inability to judge what they have not seen as unknown. In recent years, this issue has become more prominent as AI continues to be implemented in various social settings. The problem arises when AI models forcibly recognize and pretend to know what they have never encountered before, treating it as something they are familiar with. This can lead to unreliable outcomes and potentially dangerous situations.

To address this issue, Panasonic has developed a new flow-based generative model called FlowEneDet. This model enhances reliability by extending a pre-trained segmentation model with it. By incorporating an energy-based input mechanism, the flow-based detector enables a low-complex architecture that only recognizes objects that have already been trained and are inherently recognizable. Additionally, the model predicts the reliability or uncertainty of the recognition results from the image recognition AI model.

The newly developed technology has gained international recognition and was recently adopted by UAI2023 (The Conference on Uncertainty in Artificial Intelligence). UAI2023 is considered a top conference for showcasing advancements in AI and machine learning technology. The technology will be presented at the plenary session taking place in Pittsburgh, USA from July 31st to August 4th, 2023.

With its potential to enhance the reliability of AI models, Panasonic’s FlowEneDet technology has generated significant interest within the AI community. By accurately detecting and handling unknown objects, the technology aims to prevent unexpected malfunctions and improve the overall performance of AI systems. This development represents a step forward in addressing one of the major challenges associated with AI implementation.

The adoption of Panasonic’s FlowEneDet technology showcases the company’s commitment to advancing AI and machine learning capabilities. By presenting their findings at UAI2023, Panasonic aims to contribute to the broader conversation surrounding uncertainty in artificial intelligence. As AI continues to play an increasingly vital role in various industries, technologies like FlowEneDet will be crucial in ensuring reliable and safe AI systems.