Timeseries Forecasting: From Probabilistic Models to Winning Solutions

Decision Intelligence: Data Science, Forecasting & AI in Heidelberg- hosted by paretos!

Our third AI Forecasting Academy is all about probabilistic models and winning solutions. Juan Camilo Orduz will focus on Probabilistic Time Series Forecasting and present the new potentials that come with it such as a causal treatment of time series data and quantitative uncertainty estimations. On the other hand, Jakub Figura shares insights about his winning solution of the recent international VN1 Forecasting Competition. Come join us in our office or online (Hybrid-Event!) to learn from Experts in the field, network and get hands-on knowledge. Sign-up today and Stay tuned!

For updates please check below!

Our Agenda

18:00 Doors open
18:30 Welcome from the hosts
18:45 Dr. Juan Camilo Orduz - Probabilistic Time Series Forecasting
19:15 Break: Networking with pizza and beverages
20:00 Jakub Figura - VN1 Forecasting Competition: A Winning Method Overview and Submission Process
20:30 Lightning Talks
20:45 Networking with snacks and beverages
21:30 End

Want to participate in our Lightning Talks? Just shoot us a message at [email protected] - Subject: Lightning Talk.

If you are more spontaneous, no worries - meet us onsite and take the stage!

We are looking forward to meeting you!

---

Join remotely on YouTube:
Link will follow

---

Our speakers:

Juan Camilo Orduz: Principal Data Scientist - PyMC Labs
In this talk, we discuss the opportunities of custom probabilistic forecasting models in real applications. For instance,
(1) Extending classical forecasting models through hierarchies to allow information to be shared across groups (partial pooling).
(2) Forecasting unseen demand using a censoring likelihood.
(3) Calibrating forecasting models with observational data through additional likelihood components.
We will discuss the concepts and methods behind such examples and provide some code tricks to scale and reproduce the simulations presented in https://juanitorduz.github.io

Jakub Figura, Data Scientist Consultant - Forecasting, Lingaro
In this talk, we speak about:
- Exploratory Data Analysis (EDA) techniques to uncover hidden patterns and insights.
- Thoughtful feature engineering to enhance model performance.
- Designing robust cross-validation strategies for reliable model evaluation.
- Ensembling different models to achieve superior forecasting accuracy.
- Step-by-step walkthrough of the submission process and key takeaways from our winning approach in the VN1 Forecasting competition.