ML in practice - interaction between ML modeling and downstream design decisions

The theme of this talk is centered around Machine Learning from an engineering perspective. Here, accuracy is often the least interesting aspect of model selection, and very few real-world use cases depend on some 0.01% improved accuracy on ImageNet. This talk will demonstrate a particular project at Ericsson for predicting mobility in telecommunications networks. The use case will be used as a demonstrator in terms of the impact of model selection on the final task, and the sometimes very large implications a seemingly simple model choice will make on the final solution. If time (and audience) permits, the speaker will also go through a custom wait-free implementation of a sparse graph utilizing approximate updates implemented for this particular use case.

Speaker
Jesper Derehag, Sr. Machine Learning Engineer @ Ericsson AB

Schedule
17.00 - Doors open, and mingle
18.00 - Talk: Machine Learning in practice
19.30 - Food & mingle

Important info
You have to register at the Ericsson reception to be let in. The reception closes at 18, so you have to pass the gate before that time.