Meta-Learning with Context-Agnostic Initialisations
Toby Perrett (University of Bristol)*, Alessandro Masullo (University of Bristol), Tilo Burghardt (University of Bristol), Majid Mirmehdi (University of Bristol), Dima Damen (University of Bristol)
Keywords: Deep Learning for Computer Vision
Abstract:
Meta-learning approaches have addressed few-shot problems by finding initialisations suited for fine-tuning to target tasks. Often there are additional properties within training data (which we refer to as context), not relevant to the target task, which act as a distractor to meta-learning, particularly when the target task contains examples from a novel context not seen during training. We address this oversight by incorporating a context-adversarial component into the meta-learning process. This produces an initialisation which is both context-agnostic and task-generalised. We evaluate our approach on three commonly used meta-learning algorithms and four case studies. We demonstrate our context-agnostic meta-learning improves results in each case. First, we report few-shot character classification on the Omniglot dataset, using alphabets as context. An average improvement of 4.3% is observed across methods and tasks when classifying characters from an unseen alphabet. Second, we perform few-shot classification on Mini-ImageNet, obtaining context from the label hierarchy, with an average improvement of 2.8%. Third, we perform few-shot classification on CUB, with annotation metadata as context, and demonstrate an average improvement of 1.9%. Fourth, we evaluate on a dataset for personalised energy expenditure predictions from video, using participant knowledge as context. We demonstrate that context-agnostic meta-learning decreases the average mean square error by 30%.
SlidesLive
Similar Papers
OpenGAN: Open Set Generative Adversarial Networks
Luke Ditria (Monash University), Benjamin J. Meyer (Monash University)*, Tom Drummond (Monash University)

Horizontal Flipping Assisted Disentangled Feature Learning for Semi-Supervised Person Re-Identification
Gehan Hao ( University of Electronic Science and Technology of China), Yang Yang (Institute of Automation, Chinese Academy of Sciences), Xue Zhou (University of Electronic Science and Technology of China)*, Guanan Wang (CASIA), Zhen Lei (NLPR, CASIA, China)

Background Learnable Cascade for Zero-Shot Object Detection
Ye Zheng (Institute of Computing Technology, Chinese Academy of Sciences)*, Ruoran Huang (Institute of Computing Technology, Chinese Academy of Sciences), Chuanqi Han (Institute of Computing Technology, Chinese Academy of Sciences), Xi Huang (Institute of computing technology of the Chinese Academy of Sciences), Li Cui ( Institute of computing technology of the Chinese Academy of Sciences)
