Machine learning helped improve expected real-time arrival time by up to 50%.
It took nearly 13 years to provide traffic data to Google Maps, helping people clear their way, along with information about whether the traffic along the road is heavy or light, approximate travel time, and Estimated Time of Arrival (ETAs). ).
In an effort to bolster these traffic prediction capabilities, Google’s Artificial Intelligence Research Laboratory and Alphabet DeepMind have expanded to include Sydney, Tokyo, Berlin, Jakarta, Sao Paulo, and Washington, D.C. For example, real-time ETA has been improved by up to 50% in sites using a machine learning technology known as neural networks. For the graph.
Google Maps product manager Johan Lau said Google Maps uses aggregated location data and historical traffic patterns to understand traffic conditions to determine current traffic estimates, but previously ignored what traffic might happen during a traffic congestion trip.
“Our forecasts for the estimated time of arrival already contain a high-resolution bar – in fact, we see that our forecasts have been consistently accurate for over 97% of the visits … This is a technology that makes Google Maps better that enables us to predict whether or not you will be affected by it. Starts after “.
Using neural networks in the graph, DeepMind researchers said this allowed Google Maps to incorporate “a relational learning bias to model the communication structure of road networks in the real world.”
“Our tests have shown the benefit of expanding to roads adjacent to adjacent roads that are not part of the main road,” DeepMind explained in a blog post.
For example, consider how congestion on a side street spreads to affect traffic on a large street. By bypassing multiple intersections, the model naturally gains the ability to predict delays and cornering delays due to merges and overtakes in interrupted traffic. In mapping neural networks to normalize harmonic spaces lends our power to our modeling technique. “
At the same time, Google Maps has also indicated how to update its machine learning model to account for changes in global traffic patterns that have changed since the emergence of the new coronavirus epidemic.
According to Lau, the model automatically prioritizes past traffic patterns for the past two to four weeks and reduces pattern preferences from any earlier.