November Meetup

[jakeret]

Let's define the PyData meetup in November:

Speakers

• Hendrik Schöneberg (@hensb)
• Joel Akeret (@jakeret)
• ?

Date

• 1 November 2018

Location

• Zühlke Engineering, Schlieren

[jakeret]

@lukaszbinden quite some time has passed since you offered to give a talk. Do you happen to have time on the 1st of november?

@hensb could you provide an abstract for your talk (20-30min) that we can put online on our meetup page?

[lukaszbinden]

@jakeret let me discuss with my collegue and respond till Monday

[jakeret]

From PoC to Prod: Operationalization of machine learning

We see a wide adoption of Data Analytics and Machine Learning in various industry sectors. Many very promising model prototypes are being developed but only few manage to move from research-like projects into productive systems, causing frustration both on business and data scientist side. In my talk I will explore possible reasons for this and discuss why machine learning brings additional complexity to the common challenges of software engineering. I will show how we use MLFlow and Apache Airflow to address some of the difficulties in the machine learning lifecycle.

[hensb]

Data Analytics Platform in the Cloud

The rise of Data Analytics applications requires new architectures enabling the work of Data Scientists while being capable of processing large volumes and varieties of data. Building such a platform in the cloud comes with many challenges: There is a plethora of decisions to make, vendors to compare and design principles, services and products to choose from.

In this talk I will illustrate a data analytics platforms working on top of a data lake in the Google Cloud Platform, show the architectural foundation, the core concepts, and highlight important learnings.

The example is based on a marketing customer which leverages large-scale Apache Beam pipelines on Google Cloud Dataflow to crunch billions of records for their business intelligence use cases.

[lukaszbinden]

From chest to hand X-rays: Transfer learning for skeletal age prediction

The transfer learning method in machine learning aims at applying and leveraging knowledge gained from solving one task to a different but related task. In our experiment we study and try to exploit transfer learning from chest X-rays to hand X-rays for skeletal age prediction. The deep learning model is first trained on the large chest X-rays dataset and then transferred for training on the much smaller pediatric hand X-rays dataset. Finally, the model is used for age prediction on unseen hand X-rays to compete with radiologists.

In this talk we will outline the deep learning experiments performed and conclude with insights obtained from this research area.

[jakeret]

@lukaszbinden @hensb could you guys upload your slides to our repo