Artificial Intelligence and Spatial-temporal Semantics Lab

• Head of Research Group: Dr. Kristóf KARACS
• Members of the Group: Anna GELENCSÉR-HORVÁTH, András Attila SULYOK, Subham SHOME, Hyeon YU
• The laboratory focuses on environmental understanding tasks related to sensory processing. As sensors are getting smaller, they are being incorporated in a wide variety of devices. In the classical approach to processing, data is collected from sensors as external units. With the increased amount of data, transmitting all the data not only to the cloud, but even to another processor requires an unrealistic amount of bandwidth and power.
• Processing locally, as close to the sensor as possible, requires appropriate knowledge and architectural organization. The key principle we have developed is semantic embedding. This means that the hierarchical recognition process must be designed in such a way that information coming form different levels of abstraction that have a contribution to the recognition process can be used at the appropriate level.

Training and inference pipelines of an animal tracking composite AI system