Imagine new horizons
In the Industry Intelligence team we develop new algorithmic approaches to information extraction. We develop generic solutions in partnership with customers. We sell them internally in the form of solution prototypes that make sense and create value for the industry. We support the applications once deployed.
Hyper-parameters selection in embeddings
The Challenges ahead
DS analytics solutions rely on embedding techniques to vectorize data and remove biases from original representations of a variety of objects. Since we primarily sell software, not service, we rule out manual, adhoc optimization. So the goal is to identify hyper-parameters of the embedding models with high impact on embedding quality, and high dependence on input data, and approaches to automatically set them.
This may apply to embedding of multiple types of data: text, semi structured documents, graphs, and time series. The goal is to improve approaches that we have already at least in R&D.
Your Key success factors
Profile: software engineer (Python), data scientist (deep, semantics, signal)
Scientific mindset, ability to deliver, team work.