TensorFlow Buenos Aires T02E02

Segunda charla de TF Buenos Aires del año! Nos vemos en las oficinas de OLX y vamos a tener dos charlas muy interesantes. Recuerden confirmar asistencia sólo si van a poder venir para no dejar sin lugar a otras personas.

Agenda

6:30
Nos encontramos en OLX. Catering de bienvenida!

7:00
Deep pre-trained language representations
Melanie Sclar, Lead Data Scientist @ BrightSector

Self-attention, transformers y embeddings contextuales. ¿Qué hay detrás de los modelos que están provocando un cambio de paradigma en NLP y cuáles son sus aplicaciones?

7:30
Moving around my dirt pile: an introduction to Optimal Transport.
Pedro Carossi, Project Leader @ Despegar

In recent times some classical results on optimal transportation and
allocation of resources have been rediscovered under the context of DNN's, with WGAN'S as a notorious use case. In this talk we are going to give an introduction to the topic and present some interesting uses.

8:00
Unified ML model serving for production
Mouloud LOUNACI - OLX

In this talk, I propose to introduce Mleap (not to confuse with Spark MLlib) and ONNX, unified libraries that makes it possible to serve machine learning pipelines (Spark, tensorflow or scikit-learn) with the same code and reduce the prediction latency to match production SLAs. I will explain how they work, what they can/can’t do and what to expect for the future. I will also share our experiences (good and bad) with Mleap for serving spark trained ranking model and ONNX for serving tensorflow for Item to Item recommender.

8:45
Networking y comida