Two of Panasonic's Advanced AI Technologies (Home Action Genome and AutoDO)
Panasonic Develops Two Advanced AI Technologies, Home Action Genome and AutoDO - The Panasonic company has developed two advanced AI technologies, which were accepted at CVPR2021 (IEEE Conference on Computer Vision and Pattern Recognition), the world's leading international conference on AI technology.
Home Action Genome
On May 28, 2021, Panasonic announced that we have developed a new "Home Action Genome" dataset that collects people's daily activities in their homes using several types of sensors, including cameras, microphones, and thermal sensors.
This AI technology was developed to gain understanding of contrastive compositional actions.
Panasonic Japan has built and released the world's largest multimodal data set for living rooms, while most of the data sets for living rooms are small.
By applying this data set, AI researchers can use it as training data for machine learning and AI research to support people in space.
In addition to the above, Panasonic is developing cooperative learning technologies for the introduction of hierarchical activities in multimodal and multiple perspectives.
By applying this technology, it is hoped that it will be able to learn consistent features between different viewpoints, sensors, behavior hierarchies, and detailed behavior labels, and thereby improve the recognition performance of complex activities in living spaces.
This technology is the result of research conducted in collaboration between the Digital AI Technology Center, the Technology Division, and the Stanford Vision and Learning Lab at Stanford University.
Powerful AutoAugment for Biased Data with Label Noise through Scalable Probabilistic Implicit Differentiation
Panasonic also announced that it has developed a new machine learning technology that automatically performs optimal data augmentation according to the distribution of training data.
This technology can be applied to real-world situations, where very little data is available.
There are many cases in their main business area, where it is difficult to apply AI technology due to the limited data available.
By applying this technology, the process of setting data augmentation parameters can be eliminated, and parameters can be adjusted automatically.
Thus, it is hoped that the range of application of AI technology can be even wider.
In the future, by further accelerating research and development of these technologies, we will seek to realize AI technologies that can be used in real-world environments such as familiar devices and systems.
This technology is the result of research conducted by the Digital AI Technology Center, Technology Division, AI Laboratory of Panasonic R&D Company of America.
Figure 2: AutoDO solves the data augmentation problem (Shared-policy DA dilemma).
Details of this technology will be presented at CVPR2021 (to be held from 19 June 2017).