Making Deep Learning Work on Messy Sensor Data
The smartphone is perhaps the most powerful machine in human history. People take it with them everywhere they go, and it provides essentially limitless entertainment, knowledge, and utility. Almost every phone comes equipped with sensors that passively generate enormous amounts of data, such as the GPS, IMU, magnetometer and barometer. In order to take advantage of the complex higher order relationships in this sensor data, we turn to deep learning. But making deep learning algorithms work on noisy, unreliable and poorly labeled smartphone sensor data can be tricky.
Dan is a Data Scientist at TrueMotion, where he builds machine learning algorithms that use smartphone sensors to understand and score driving behaviors. Dan leads TrueMotion's efforts on developing smartphone IMU algorithms to detect hard brakes and distracted driving. Dan is also a guest speaker at the NYC Data Science Academy. In the past, he has worked as a neurosurgery researcher at Rhode Island Hospital, as a Digital Humanities Programmer at the Brown University Library, and as a Computational Biology Software Consultant for the Weinreich Lab at Brown University. Dan graduated from Brown University in 2015.