DeepMind has improved travel times by 50% using a dedicated AI model. Also, the algorithm will help optimize the driver’s path and make it faster.
DeepMind, using its artificial intelligence (AI) model, has improved navigation in Google maps – according to engineers, they made the calculation of travel times more accurate by 50%. The experiments were carried out in several regions at once, including Berlin, Jakarta, Sao Paulo, Sydney, Tokyo, and Washington. Through the use of machine learning techniques, they were able to reduce the errors in traffic prediction by incorporating relational learning that models road networks.
Google Maps analyzes traffic in real-time on roads around the world, but it does not use many inputs – traffic load, average speed at a particular site, and others. Machine learning enables Google Maps to combine traffic conditions with historical road models around the world. To achieve this goal, DeepMind has developed neural networks of graphs that conduct space-time reasoning.
All of this information feeds into neural networks developed by DeepMind, which pick patterns in the data and use them to predict future traffic. Google says its new models have improved forecasting accuracy, but it will become even more accurate in the future. In this case, the data will be automatically correlated; they do not need the help of researchers.
The models work by dividing maps into what Google calls “super segments” – the total traffic of adjacent roads. Moreover, each of them is connected to an individual neural network, which makes a prediction of traffic intensity for a specific sector. It is not known how large these “super-segments” are, but Google notes that they are “dynamically sized”, each of which uses data processing. The uniqueness of the approach is that a special neural network is used for this, which is well suited for analyzing cartographic data.