7:00 pm-9:00 pm
Haven’t we all heard of TFs major upgrade? But do we really now what changes have been introduced & how they’re affecting and hopefully enhancing our workflows? We’ve an expert on the topic—Romeo is actually writing a book on the topic and as IBM Watson’s chief data scientist, he’s also pretty badass.
Second talk is by Ilja, who’s a machine learning engineer/data scientist— actually a medical doctor by formation. He’s been knee deep into Flow Cytometry used for cells phenotyping, which is a great example of how powerful Python can be for data wrangling & classification in a variety of fields.
*PyData is a grassroots effort: please help us with networking for interesting speakers of all experience-levels & cross-disciplinary backgrounds: Serving a community where newbies, pros & everything inbetween find joy & new inputs—while sharing knowledge and a love for open source & Python*
*Contribute your contacts, demands, ideas preferably via https://github.com/pydatazrh/meetup/issues, but also feel free to walk up to us over beers or ping us via meetup.com PM*
Towards de-facto standard in AI: What’s new in TF 2.0
Despite the hype around TF, there have been complaints on usability as well: The fact that debugging was only possible after construction of the static execution graph. Also, neural networks needed to be expressed as a set of linear algebra operations—considered as too low level by many practitioners.
PyTorch and Keras addressed many of the flaws in TensorFlow & gained a lot of ground.
TF 2.0 successfully addressed those complaints & promises to become the go-to framework for many AI problems.
We’ll introduce the most prominent changes in TF 2.0 & how you can use these new features successfully in your projects. We’ll cover eager execution, parallelisation strategies, the advantages of the tight high level Keras integration, live neural network training monitoring using TensorBoard, automated hyper parameter optimization, Model serving with TF Service, TF.js & TF Lite. We’ll finalise with an outlook on TFX – where Google is planning to open source it’s complete AI pipeline and will contrast it with existing de-facto standard frameworks like Apache Spark.
Bio: Romeo Kienzler is Chief Data Scientist at the IBM, with specialisation in Information Systems, Bioinformatics and Applied Statistics & teaching in Bern, his research focusing on cloud-scale machine learning and deep learning using & contributing to open source technologies including TensorFlow, Keras, DeepLearning4J, Apache SystemML and the Apache Spark stack.
Cell Phenotyping with Python
This talk will be about Flow Cytometry used for cells phenotyping, which is how the cells of immune system are scanned. In broad strokes, in this hybrid talk/tutorial we will read some fcs files with immune fluorescence data & cluster them in to cell categories
Since we’ll be dealing with some pretty complex error surfaces we’ll use some principles of evolutionary algorithms to avoid gradient descent in complex loss function with lots of local minima. This is quite a technical solution but with biological principals. The approach that will be described is CMA-ES, which stands for covariance matrix adaptation evolution strategy.
Bio: Ilja Rasin is Physician, University Lecturer in Mathematics, Programmer – now unifying all this as a Data Scientist focusing on making the world a better (ideally more predictable) place, currently at IBM Watson Health as deep learning engineer and data scientist.
Sign up (optional, recommended): https://www.eventbrite.com/e/tensorflow-20-whats-new-cell-phenotyping-with-python-tickets-63120694697
Please register beforehand w/ full name & save us all time getting you badges!
IBM Schweiz AG Vulkanstrasse 106 · 8048 Zurich