BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//Talks.cam//talks.cam.ac.uk//
X-WR-CALNAME:Talks.cam
BEGIN:VEVENT
SUMMARY:Scaling laws for large time-series models: More data\, more parame
 ters - James Alvey (KICC)
DTSTART:20241028T160000Z
DTEND:20241028T170000Z
UID:TALK222820@talks.cam.ac.uk
CONTACT:65128
DESCRIPTION:Scaling laws for large language models (LLMs) offer valuable i
 nsights into how increasing model size and training data leads to predicta
 ble performance improvements. Time series forecasting\, which shares a seq
 uential structure similar to language\, is well-suited to large-scale tran
 sformer architectures. In this talk\, I will demonstrate that foundational
  decoder-only time series transformer models exhibit scaling behaviour ana
 logous to LLMs. I will begin with a general introduction to scaling laws a
 nd how they can inform efficient\, optimised model training. I will then f
 ocus on their specific application to time series data\, highlighting the 
 emergence of power law behaviour. Finally\, I will discuss the broader imp
 lications of these findings\, and potential scientific applications.\n\nRe
 lated papers:\nhttps://arxiv.org/abs/2001.08361\nhttps://arxiv.org/abs/240
 5.13867
LOCATION:Martin Ryle Seminar Room\, KICC
END:VEVENT
END:VCALENDAR