Machine learning method and machine learning apparatus

A machine learning apparatus comprising:

• a storage configured to store a data set including pieces of position data, first ground truth occupancy data, and ground truth flow data,
• the pieces of position data indicating respective positions of one or more objects at respective first points in time,
• the first ground truth occupancy data comprising ground truth map data indicating occupancy probabilities of the one or more objects at a second point in time later than the first points in time, the first ground truth occupancy data including one or more ground truth occupancy regions,
• the ground truth flow data comprising ground truth map data indicating respective velocity vectors of the one or more objects at the second point in time;
• and
• a processor configured to perform a machine learning process, based on the data set, wherein
• the processor is configured to:
• input the pieces of position data to a machine learning model to generate predicted occupancy data and predicted flow data,
• the predicted occupancy data comprising map data indicating predicted occupancy probabilities of the one or more objects at the second point in time,
• the predicted flow data comprising map data indicating predicted velocity vectors of the one or more objects at the second point in time,
• generate, before calculating a loss parameter, second ground truth occupancy data by expanding the one or more ground truth occupancy regions to generate one or more expanded ground truth occupancy regions, the second ground truth occupancy data including the one or more expanded ground truth occupancy regions,
• calculate the loss parameter based on the predicted occupancy data, the second ground truth occupancy data, the predicted flow data, and the ground truth flow data, the calculating including calculating an occupancy loss by comparing the predicted occupancy data with the second ground truth occupancy data and calculating a flow loss by comparing the predicted flow data with the ground truth flow data, and
• update model parameters of the machine learning model, based on the loss parameter.