June 7th we're heading to the RDM in Rotterdam for some interesting talks on what people are doing in practice with Machine Learning

Schedule:
18:00 - Doors open & food
18:30 - Automatic Detection of Zebra Crossings in Aerial Imagery
19:00 - Break
19:15 - Deep Reinforcement Learning with Unity 3D
20:00 - Networking
21:00 - End of event

Directions:

Getting to RDM is a little tricky by public transport, so if going by car is an option for you, I'd definitely do it. The advantage of going with public transport is the free Watertaxi ride at the end of the event. (More info below)

By car : This one's easy. There is plenty of free parking space nearby, from the highway follow the signs Heijplaat / RDM

By public transport : From Rotterdam Centraal, it's easiest to rent an OV fiets and ride to Jobshaven, where there is a Waterbus leaving. Taking the tram to the Waterbus is also possible, but you'd likely miss the connection. The Waterbus only leaves every 30 min. You can take your bike on the Waterbus, but if you plan on taking the Watertaxi back with us, please leave your bike at Jobshaven as you are not allowed to take it with you on the Watertaxi!

Return trip : The last Waterbus back to Jobshaven departs at 20:20. So we have arranged (free!) Watertaxi's leaving at around the end of the event. Watertaxi's are really fast and very cool, so I can definitely recommend it!

Once you're at RDM, please follow these directions to get into the building:

https://drive.google.com/open?id=14M-xcPwYqZtb0CzUE2Hp9LKZN5e-CL15

I will make sure to have you guided from there!

Captain AI and https://www.meetup.com/RDMTech/ are sponsoring location, drinks and food.

• GeoAI: Automatic Detection of Zebra Crossings in Aerial Imagery by Maartje Holtslag & Niels van der Vaart (Esri Nederland)

Zebra crossings are an important feature on Dutch streets, as they give pedestrians the opportunity to safely cross roads. However, in the Netherlands the exact locations of zebra crossings are not known. In this project I created a method to automatically detect zebra crossings in aerial imagery, by using the TensorFlow Object Detection API to create a neural network model. The Scikit-Optimize library was used to optimize the hyper-parameters of this model. The final model was implemented into ArcGIS Pro to be able to detect zebra crossings in aerial imagery in a live environment.

Machine learning can be applied in many different fields. One of the relatively new fields in which machine learning is currently explored is the Geographic Information Systems (GIS) field. Spatial data cannot only be used to visualize results, but interesting analysis can be performed using this information. During this presentation an overview will be given of the possibilities when combining machine learning with GIS. A demonstration will show how an object detection framework is used to detect zebra crossings in aerial imagery.

• Deep Reinforcement Learning with Unity 3D by Gerard Simons (Captain AI)

Deep Reinforcement Learning (RL) is a very exciting relatively new field of machine learning, that has gained much notoriety thanks to the recent achievements of Deep Mind's Alpha Go. The idea behind it is to train a deep neural network to exhibit desirable behaviour by interacting with an environment. More specifically, it learns a series of actions, called a policy, that maximises a user-defined reward function given the current state of environment. Unity 3D is a popular game engine that has recently incorporated a new open-source plug-in called ML Agents that allows anyone to develop these RL agents inside a Unity game.

At Captain AI we have experimented with this plug-in to show the potential of RL for solving navigational tasks in a maritime environment. During the presentation we will discuss the theory and principles behind RL, how Unity 3D and the MLAgents plugin works and show you what all this looks like in an autonomous shipping scenario.