BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//Talks.cam//talks.cam.ac.uk//
X-WR-CALNAME:Talks.cam
BEGIN:VEVENT
SUMMARY:Variational Uncertainty Decomposition for In-Context Learning - Yi
 ngzhen Li (Imperial College London)
DTSTART:20250612T100000Z
DTEND:20250612T113000Z
UID:TALK233071@talks.cam.ac.uk
CONTACT:Xianda Sun
DESCRIPTION:As large language models (LLMs) gain popularity in conducting 
 prediction tasks in-context\, understanding the sources of uncertainty in 
 in-context learning becomes essential to ensuring reliability. The recent 
 hypothesis of in-context learning performing predictive Bayesian inference
  opens the avenue for Bayesian uncertainty estimation\, particularly for d
 ecomposing uncertainty into epistemic uncertainty due to lack of in-contex
 t data and aleatoric uncertainty inherent in the in-context prediction tas
 k. However\, the decomposition idea remains under-explored due to the intr
 actability of the latent parameter posterior from the underlying Bayesian 
 model. In this work\, we introduce a variational uncertainty decomposition
  framework for in-context learning without explicitly sampling from the la
 tent parameter posterior\, by optimising auxiliary inputs as probes to obt
 ain an upper bound to the aleatoric uncertainty of an LLM's in-context lea
 rning procedure. Through experiments on synthetic and real-world tasks\, w
 e show quantitatively and qualitatively that the decomposed uncertainties 
 obtained from our method exhibit desirable properties of epistemic and ale
 atoric uncertainty.
LOCATION:Cambridge University Engineering Department\, CBL Seminar room BE
 4-38.
END:VEVENT
END:VCALENDAR