MLMU Prague is back! We are thrilled to announce that we managed to find 3 awesome speakers for our next ML meetup and since we have not seen our community for a while, we are really looking forward to seeing you all!
This time, the talks will focus on "ML in the wild". Following tutorials, reading papers and watching videos about various applications, training a cool model using the state of the art algorithms. But what about all that stuff no one actually talks about? Do you really need this state of the art models? Are you even tackling the right problem? And when you trained the model, how do you get it to the production? This and even more will be covered during the three talks of our October meetup.
Please register here by clicking RSVP Yes! The number of seats is limited.
18:00-18:30 Arrival of attendees
18:30-19:00 Radovan Parrák: Seamless Development >> Deployment >> Delivery of Data Products using Open-source tooling
19:00-19:30 Boril Šopov: TBA
19:30-20:00 Pablo Maldonado: Adopting machine learning models as part of the business
=== Seamless Development >> Deployment >> Delivery of Data Products using Open-source tooling ===
Abstract: So your brand new, all fine-tuned, nifty predictive model is ready? Have you found an interesting pattern in the data that might be worthwhile trying out in reality? Have you created a flashy visualization that might help business arm to steer decisions better? Or did you create a simple yet effective dashboard that your customer now wants to have available in real-time? Well, brace yourself now because the excruciating deployment and delivery process of your “baby” is about to start. All data scientists have been in this, or similar, situation by now... In this talk, I will demonstrate our take on seamless development, deployment, and delivery of Data Products, using a fully open-source tool stack. I will demonstrate that turning Data Insights into Data Products need not to be a headache for anyone involved (Data Scientists, ML DevOps, Data Science PMs) - even when you don’t want to tie your hands and wallet to an “all mighty, all integrated” commercial data science platform.
Bio: Rado is a Data Scientist and ML DevOps enthusiast in Credo.
=== In the past “Code as code”. More recently “infrastructure as code” and I say “model as a code”. ===
Abstract: In this presentation, I will go through several phases of a standard model management cycle. These phases will be shown in real working example of how to deploy ML model into production using modern tools such as Python, Docker, Gitlab Pipelines, Amazon AWS, Docker Swarm. All of it automatically with sound CI/CD and with provided code, which you can draw inspiration from.
Bio: Boril is a founder and a product leader in an insurtech startup Searpent.
=== Adopting machine learning models as part of the business ===
Abstract: A model that is not used by anyone is a useless model. Unfortunately, many companies invest in developing data science prototypes, but these often go nowhere beyond the "impress executives" phase. Disappointed business owners and bored data scientists are the only tangible product of these failed ventures. Can this sad outcome be avoided? Yes, it can! In this talk we will share some useful points to identify concrete business problems that can be attacked with machine learning and how to better communicate what ML models are doing to non-ML-experts. We will also explore some common features of healthy data science teams that do deliver business value as well as some red flags.
Bio: Pablo Maldonado is an applied mathematics and data science consultant based in Prague.