A SURVEY ON META-LEARNING

PhD Qualifying Examination


Title: "A SURVEY ON META-LEARNING"

by

Mr. Weisen JIANG


Abstract:

Humans can extract knowledge and experience from historical tasks to accelerate
learning new tasks from few examples. However, deep networks are data-hungry,
and a large number of labeled samples are required for training. In order to
reduce the labor-intensive and time-consuming data labeling process,
meta-learning (or learning-to-learn) aims at extracting meta-knowledge from
seen tasks to accelerate learning on unseen tasks. This survey provides an
overview of meta-learning. We review popular meta-learning algorithms, which
are categorized into three groups: (i) optimization-based methods include
metainitialization and meta-regularization; (ii) metric-based methods developed
for few-shot classification; and (iii) memory-based methods using a memory
buffer or hypernetworks to store meta-knowledge. We further discuss
applications of meta-learning in natural language processing, including
prompting and in-context learning. Lastly, we present several directions for
future research.


Date:                   Wednesday, 16 August 2023

Time:                   10:00am - 12:00noon

Venue:                  Room 3494
                        Lifts 25/26

Committee Members:      Prof. James Kwok (Supervisor)
                        Dr. Brian Mak (Chairperson)
                        Dr. Minhao Cheng
                        Dr. Dan Xu


**** ALL are Welcome ****