Learn Continually, Generalize Rapidly: Lifelong Knowledge Accumulation for Few-shot Learning

By Xisen Jin, Bill Yuchen Lin, Mohammad Rostami, Xiang Ren. 2021

47 citations recorded by the source corpus.

The ability to continuously expand knowledge over time and utilize it to rapidly generalize to new tasks is a key feature of human linguistic intelligence. Existing models that pursue rapid generalization to new tasks (e.g., few-shot learning methods), however, are mostly trained in a single shot on fixed datasets, unable to dynamically expand their knowledge; while continual learning algorithms are not specifically designed for rapid generalization. We present a new learning setup, Continual Learning of Few-Shot Learners (CLIF), to address the challenges of both learning settings in a unified setup. CLIF assumes a model learns from a sequence of diverse NLP tasks arriving sequentially, accumulating knowledge for improved generalization to new tasks, while also retaining performance on the tasks learned earlier. We examine how the generalization ability is affected in the continual learning setup, evaluate a number of continual learning algorithms, and propose a novel regularized adapter generation approach. We find that catastrophic forgetting affects generalization ability to a less degree than performance on seen tasks; while continual learning algorithms can still bring considerable benefit to the generalization ability.

Subfields: Artificial Intelligence, Computer Vision and Pattern Recognition, Meta & Continual Learning, Statistical Learning Theory

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

1. Zixuan Ke
2. Hu Xu
3. Bing Liu
4. Lei Shu
5. Bin Liu
6. Jun Zhu
7. Hang Su
8. Xingxing Zhang

In Computer Vision and Pattern Recognition

1. Jun Zhu
2. Zixuan Ke
3. Hang Su
4. Bin Liu
5. Liyuan Wang
6. Xingxing Zhang
7. Hu Xu
8. Lei Shu

In Meta & Continual Learning

1. Zixuan Ke
2. Bin Liu
3. Hu Xu
4. Lei Shu
5. Liyuan Wang
6. Xingxing Zhang
7. Hang Su
8. Jun Zhu

In Statistical Learning Theory

1. Mohammad Rostami
2. Jun Zhu
3. Caiming Xiong
4. Jesse Thomason
5. Bin Liu
6. Zixuan Ke

Citations in the corpus

• A Comprehensive Survey of Continual Learning: Theory, Method and Application (2023, background): “For example, a pre-trained transformer can effectively accommodate downstream tasks by learning only a few task-specific parameters, including adaptor-tuning [165] with inserted fully-connected layers (CTR [208], B-CL [209], CLASSIC [207], CPT [205], CLIF [190], AdapterCL [282], ACM [499], ADA [111]”
• Continual Training of Language Models for Few-Shot Learning (2022, background): “Several researchers have studied this problem recently (An-toniou et al., 2020; Qin and Joty, 2021; Jin et al., 2021; Xia et al., 2021; Yin et al., 2022).”
• CLiMB: A Continual Learning Benchmark for Vision-and-Language Tasks (2022, background): “The CLIF-26 benchmark [Jin et al., 2021] is built for CL on the GLUE [Wang et al., 2019] language tasks.”
• Continual Sequence Generation with Adaptive Compositional Modules (2022, background): “Extra modules like attention module (Pfeiffer et al., 2020a), capsule network (Ke et al., 2021), and hypernetworks (Jin et al., 2021) are demonstrated beneficial for knowledge transfer, but they need to introduce extra parameters and fail to consider any reusable or compositional modules.”
• ConTinTin: Continual Learning from Task Instructions (2022, background): “Similar with (Xia et al., 2021; Jin et al., 2021), our work also focuses on low-resource continual learning (Solomonoff, 1989), sequential learning (McCloskey and Cohen, 1989), and never-ending learning (Carlson et al., 2010). learning; in contrast, our learning problem belongs to TCL while each…”
• LFS-GAN: Lifelong Few-Shot Image Generation (2023, background): “The above example assumes that the rank r =2.”