BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//Talks.cam//talks.cam.ac.uk//
X-WR-CALNAME:Talks.cam
BEGIN:VEVENT
SUMMARY:The second generation of meta-learning methods - Massimiliano (Max
 ) Patacchiola\, University of Cambridge
DTSTART:20221103T130000Z
DTEND:20221103T140000Z
UID:TALK189554@talks.cam.ac.uk
CONTACT:Microsoft Research Cambridge Talks Admins
DESCRIPTION:In the last few years\, several methods have been developed to
  tackle the few-shot setting that is\, learning from a limited amount of d
 ata belonging to a specific task. Most approaches have followed a meta-lea
 rning paradigm that is\, learning-to-learn through exposure to large numbe
 rs of training tasks (episodic training) with the aim of generalizing to n
 ew unseen tasks at test time. Meta-learning has been accomplished in diffe
 rent ways\, via rapid adaption in a few gradient steps (e.g. MAML)\, explo
 iting learned metrics (e.g. ProtoNets)\, or through probabilistic approach
 es (e.g. Deep Kernel Transfer\, CNAPs). However\, most meta-learning metho
 ds are affected by severe issues such as training instability and poor sca
 ling that hinder their performance. In addition\, there has been a growing
  body of empirical research showing that simpler fine-tuning routines are 
 very effective at few-shot image classification\, while being much easier 
 to train and deploy. In this talk I will start from these empirical findin
 gs\, comparing strengths and weaknesses of meta-learners and fine-tuners. 
 I will then introduce technical solutions that could be used as building b
 locks for a second generation of meta-learners. In particular\, I will des
 cribe our recent method called LITE (Bronskill et al.\, NeurIPS 2021) that
  allows meta-training efficiently on large images\, and a new adaptive blo
 ck called CaSE (Patacchiola et al.\, NeurIPS 2022) that allows fast adapta
 tion of pretrained models on a context set. I will provide strong empirica
 l evidence showing that methods based on LITE and CaSE are able to achieve
  state-of-the-art performance on a variety of tasks\, including real-world
  personalization benchmarks such as the recently proposed ORBIT.
LOCATION:Small lecture room\, Microsoft Research Ltd\, 21 Station Road\, C
 ambridge\, CB1 2FB
END:VEVENT
END:VCALENDAR