* The most important thing is that you are excited about this project, and like the mission of our company.
* Highly knowledgeable about machine learning and neural networks. Both practical and theoretical.
* Good computer skills, and general tech savvy: you will do many tasks like setting up a rasberry pi to record datasets, creating small python programs for dataset labelling and management, manipulating large image datasets, and many others.
* Experience with image processing and convolutional networks
* Computer programming experience in other areas is helpful
Promising drone-delivery startup is hiring a Dataset and Machine Learning Manager to work remotely. You will build the Datasets and Neural Networks that fly our autonomous delivery drones. These drones will then deliver millions of meals and other packages in the real world.
Just like any real life machine learning project, you will spend most of your time on infrastructure and dataset management, followed by a small amount of time where you actually train and optimise the neural networks.
These are the 3 most important neural networks that you will build:
• Customer finder: This neural network takes the images from the drone camera and predicts where the customers balcony is. To create this dataset, you will build a simplified mini version of our drone, and equip it with cameras and a rasberry pi for recording. You will then go to a miniature park to record a dataset set. You will also work together with our game developer to export a synthetic dataset from our simulator. You will then mix those datasets together, and train our neural network.
• Depth estimator: This one takes the drone cameras as input, and generates a simplified 3d map of the surroundings You will install a 3d camera on one of hour drones, and record a couple hours of footage, to have 3d ground truth matched with the default cameras of our drone. You will then work with our game developer again to export more synthetic data from our simulator. Mix them together, and train the network.
• Control network: This network takes the 3d depth map, the customer location, and some other inputs and creates the steering commands that control the drone. First you will perform some deliveries by hand to have a dataset for imitation learning. After our network has reached a decent level through imitation learning, you will work together with our game developer to do millions of deliveries in our simulator and optimise it with an unsupervised learning technique like reinforcement learning or genetic learning.
There will also be periods where you will help set up the computer infrastructure for these machine learning processes.