Machine Learning on Geospatial Data with Malte Loller-Anderson & Mathilde Ørstavik
2024/08/29
再生時間： 54 分
ポッドキャスト

カートのアイテムが多すぎます

ご購入は五十タイトルがカートに入っている場合のみです。

カートに追加できませんでした。

しばらく経ってから再度お試しください。

ウィッシュリストに追加できませんでした。

しばらく経ってから再度お試しください。

ほしい物リストの削除に失敗しました。

しばらく経ってから再度お試しください。

ポッドキャストのフォローに失敗しました

ポッドキャストのフォロー解除に失敗しました

Machine Learning on Geospatial Data with Malte Loller-Anderson & Mathilde Ørstavik

無料で聴く

ポッドキャストの詳細を見る

サマリー
What can machine learning do for geospatial data? Carl and Richard talk to Malte Loller-Anderson and Mathilde Ørstavik about their work at Norkart, using aerial imagery to build detailed maps around Norway. Mathilde dives into the critical role of machine learning - identifying buildings in images. Usually done by hand with each new image, Norkart has a machine learning model that automates the process trained on previous vector maps of buildings. But there are many things that look like buildings in Norway, including patches of snow, mountains, and even shapes under water. Malte also discusses how Norkart has decided to train in-house with nVidia L40 processors rather than in the cloud - the hardware is used 24 hours a day since some models can take weeks to train! There are many interesting ideas about geospatial data and machine learning from people who have been doing it for years.

続きを読む一部表示

あらすじ・解説

What can machine learning do for geospatial data? Carl and Richard talk to Malte Loller-Anderson and Mathilde Ørstavik about their work at Norkart, using aerial imagery to build detailed maps around Norway. Mathilde dives into the critical role of machine learning - identifying buildings in images. Usually done by hand with each new image, Norkart has a machine learning model that automates the process trained on previous vector maps of buildings. But there are many things that look like buildings in Norway, including patches of snow, mountains, and even shapes under water. Malte also discusses how Norkart has decided to train in-house with nVidia L40 processors rather than in the cloud - the hardware is used 24 hours a day since some models can take weeks to train! There are many interesting ideas about geospatial data and machine learning from people who have been doing it for years.

続きを読む一部表示