BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//pretalx//pretalx.com//pydata-amsterdam2026//talk//9GR8L8
BEGIN:VTIMEZONE
TZID:CET
BEGIN:STANDARD
DTSTART:20001029T040000
RRULE:FREQ=YEARLY;BYDAY=-1SU;BYMONTH=10
TZNAME:CET
TZOFFSETFROM:+0200
TZOFFSETTO:+0100
END:STANDARD
BEGIN:DAYLIGHT
DTSTART:20000326T030000
RRULE:FREQ=YEARLY;BYDAY=-1SU;BYMONTH=3
TZNAME:CEST
TZOFFSETFROM:+0100
TZOFFSETTO:+0200
END:DAYLIGHT
END:VTIMEZONE
BEGIN:VEVENT
UID:pretalx-pydata-amsterdam2026-9GR8L8@pretalx.com
DTSTART;TZID=CET:20260910T133000
DTEND;TZID=CET:20260910T141500
DESCRIPTION:Representation learning has transformed AI for language and vis
 ion. Large foundation models learn abstract\, reusable features that can t
 ransfer to arbitrary downstream tasks. But what about other real world pro
 blems ? \nMany practitioners work with inputs that are high-dimensional co
 ntaining complex non-linear patterns: weather or satellite maps\, sensor n
 etworks\, graph data or medical scans. In contrast\, prediction targets ar
 e often simple and low-dimensional. Learning directly from raw high-dimens
 ional inputs to low-dimensional targets is tempting\, but often fails in p
 ractice: limited labeled data forces models to simultaneously discover use
 ful representations *and* solve the prediction task.\n\nThis talk presents
  the *embed first\, predict later* paradigm and shows that it works just a
 s powerfully for other domains as it does for text and images\, using ener
 gy forecasting from weather data as our running example. We show how a pre
 -trained open source foundation model learns abstract representations of w
 eather such as storms or cloud systems\, and how these embeddings can be f
 ed into a multitude of lightweight downstream tasks. The result is a modul
 ar and reusable ML pipeline that generalizes well.\n\nYou'll leave with th
 ree concrete things: (1) the building blocks of an embedding-first pipelin
 e\, illustrated through a real-world implementation\, (2) practical tips a
 nd tricks for visualizing and inspecting high-dimensional embedding spaces
  and (3) an intuition for when this paradigm can be applied to your own do
 main and pointers on how to get started.
DTSTAMP:20260710T164112Z
LOCATION:Room 2 (350)
SUMMARY:Embed First\, Predict Later: Energy forecasting from weather embedd
 ings - Kai Jeggle
URL:https://pretalx.com/pydata-amsterdam2026/talk/9GR8L8/
END:VEVENT
END:VCALENDAR