Transfer learning

In practical applications, the annotation cost for many tasks is high, making it difficult to obtain sufficient training data. In such cases, transfer learning can be used. Suppose A and B are two related tasks. If task A has abundant training data, we can transfer some of the generalized knowledge learned from task A to task B. Transfer learning has several categories, including domain adaptation and domain generalization 1 2 3 4 5.

The paradigm of transfer learning in LLMs is pre-trained on massive data (multi-task learning or making the multi-task data into a unified form) and fine-tuned for specific tasks.

Domain Adaptation

Some libraries:

• Transfer-Learning-Library
• deep-transfer-learning: a PyTorch library for deep transfer learning
• salad: a Semi-supervised Adaptive Learning Across Domains
• Dassl: a PyTorch toolbox for domain adaptation and semi-supervised learning
• joliGEN: an integrated framework for training custom generative AI image-to-image models

Domain adaptation deals with scenarios where we have a source domain (A) and a target domain (B). The goal is to train a model using data from the source domain and make it perform well on the target domain, even though labeled data from the target domain is scarce or unavailable.

A basic idea for domain adaptation is to train a feature extractor to eliminate the different part and keep the collective part/the essence of the source and target domain.

A famous work is 1505.07818 Domain-Adversarial Training of Neural Networks.

The problem is that we want the domain classifier to split the data of two domain and meanwhile the feature extractor to get the collective feature. It seems a little tricky.

Other works:

• 1909.13231 Test-Time Training with Self-Supervision for Generalization under Distribution Shifts
• Test-Time Training Project Website
• CVPR 2019 Open Access Repository
• 1712.02560 Maximum Classifier Discrepancy for Unsupervised Domain Adaptation
• 1802.08735 A DIRT-T Approach to Unsupervised Domain Adaptation
• Accelerating Deep Unsupervised Domain Adaptation with Transfer Channel Pruning IJCNN
• Adversarial Adaptation of Scene Graph Models for Understanding Civic Issues WWW2019
• Cross-Dataset Adaptation for Visual Question Answering CVPR2018
• Cross-domain fault diagnosis through optimal transport for a CSTR process DYCOPS2022 Code

Domain generalization/Learn a domain-agnostic model

Domain generalization focuses on a more challenging setting. Here, we have access to multiple related domains during training (e.g., sketches, cartoons, real images), but the goal is to build a model that generalizes well to unseen testing domains.

• jindongwang/transferlearning: Transfer learning / domain adaptation / domain generalization / multi-task learning etc. Papers, codes, datasets, applications, tutorials.-迁移学习
• 2003.13216 Learning to Learn Single Domain Generalization
• Domain Generalization with Adversarial Feature Learning | IEEE Conference Publication | IEEE Xplore