Learning What and Where to Transfer

By Yunhun Jang, Hankook Lee, Sung Ju Hwang, Jinwoo Shin. 2019

165 citations recorded by the source corpus.

As the application of deep learning has expanded to real-world problems with insufficient volume of training data, transfer learning recently has gained much attention as means of improving the performance in such small-data regime. However, when existing methods are applied between heterogeneous architectures and tasks, it becomes more important to manage their detailed configurations and often requires exhaustive tuning on them for the desired performance. To address the issue, we propose a novel transfer learning approach based on meta-learning that can automatically learn what knowledge to transfer from the source network to where in the target network. Given source and target networks, we propose an efficient training scheme to learn meta-networks that decide (a) which pairs of layers between the source and target networks should be matched for knowledge transfer and (b) which features and how much knowledge from each feature should be transferred. We validate our meta-transfer approach against recent transfer learning methods on various datasets and network architectures, on which our automated scheme significantly outperforms the prior baselines that find "what and where to transfer" in a hand-crafted manner.

Subfields: Artificial Intelligence, Computer Vision and Pattern Recognition

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Artificial Intelligence

In Computer Vision and Pattern Recognition

Citations in the corpus

Transfer without Forgetting (2022, background): “Alternatively, transfer can be induced via multi-level Knowledge Distillation, guided by meta-learning [30], attention [66] or higher-level descriptions of the flow of information within the model [67].”
Knowledge Distillation: A Survey (2020, methodology): “Recently, an increasing number of KD variants, which are based on the lifelong learning, have been developed (Jang et al., 2019; Flennerhag et al., 2019; Peng et al., 2019b; Liu et al., 2019e; Lee et al., 2019b; Zhai et al., 2019; Zhou et al., 2020; Shmelkov et al., 2017; Li and Hoiem, 2017).”
Freeze the Discriminator: a Simple Baseline for Fine-Tuning GANs (2020, background): “Investigating more advanced techniques ( e.g ., [1, 18, 34]) would be an interesting and promising future direction.”
TranSlider: Transfer Ensemble Learning from Exploitation to Exploration (2020, background): “Yet more recent studies [11, 16, 26, 27, 33, 41] attempt to simultaneously learn a transfer strategy and the parameters for the target model.”
Learning to Adapt to Evolving Domains (2020, background): “Through an explicit learning-to-learn paradigm, the meta-learner is able to learn transferable features for downstream tasks [13, 29], which is critical in few-shot learning and transfer learning.”
Pay Attention to Features, Transfer Learn Faster CNNs (2020, methodology): “Instead of manually deciding the regularization penalties and what to regularize as in the previous approaches, Jang et al. (2019) used meta-learning to automatically learn what knowledge to transfer from the teacher and to where in the student model.”
Show, Attend and Distill: Knowledge Distillation via Attention-based Feature Matching (2021, methodology): “To compensate for the limitation, Jang et al. (Jang et al. 2019) apply a meta-networks, “learning to transfer (L2T)”, automatically determining the links.”
Cross-Layer Distillation with Semantic Calibration (2020, unknown): “Based on the transfer learning framework, a newly solution is to learn association weights by a meta-network given only feature maps of the source network (Jang et al. 2019), while our proposed approach incorporates more information from teacher-student layer pairs.”
Self-Distillation: Towards Efficient and Compact Neural Networks (2021, background): “Moreover, Jang et al. propose L2T, which exploits a meta neural network to learn the best layers and weights for knowledge transferring [45].”