Towards Continual Knowledge Learning of Language Models

By Joel Jang, Seonghyeon Ye, Sohee Yang, Joongbo Shin, Janghoon Han, Gyeonghun Kim, Stanley Jungkyu Choi, Minjoon Seo. 2021

210 citations recorded by the source corpus.

Large Language Models (LMs) are known to encode world knowledge in their parameters as they pretrain on a vast amount of web corpus, which is often utilized for performing knowledge-dependent downstream tasks such as question answering, fact-checking, and open dialogue. In real-world scenarios, the world knowledge stored in the LMs can quickly become outdated as the world changes, but it is non-trivial to avoid catastrophic forgetting and reliably acquire new knowledge while preserving invariant knowledge. To push the community towards better maintenance of ever-changing LMs, we formulate a new continual learning (CL) problem called Continual Knowledge Learning (CKL). We construct a new benchmark and metric to quantify the retention of time-invariant world knowledge, the update of outdated knowledge, and the acquisition of new knowledge. We adopt applicable recent methods from literature to create several strong baselines. Through extensive experiments, we find that CKL exhibits unique challenges that are not addressed in previous CL setups, where parameter expansion is necessary to reliably retain and learn knowledge simultaneously. By highlighting the critical causes of knowledge forgetting, we show that CKL is a challenging and important problem that helps us better understand and train ever-changing LMs. The benchmark datasets, evaluation script, and baseline code to reproduce our results are available at https://github.com/joeljang/continual-knowledge-learning.

Subfields: Large Language Models, Meta & Continual Learning

How researchers see this paper

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

In Large Language Models

In Meta & Continual Learning

Citations in the corpus

A Comprehensive Survey of Continual Learning: Theory, Method and Application (2023, background): “learning in visual domains for two aspects: (1) the greater efforts in pre-training, including both upstream [149], [184], [399] and downstream [111], [454] continual learning; and”
ELLE: Efficient Lifelong Pre-training for Emerging Data (2022, background): “tual knowledge that changes over time (Dhingra et al., 2021; Jang et al., 2021).”
Factuality Enhanced Language Models for Open-Ended Text Generation (2022, background): “[40] Joel Jang, Seonghyeon Ye, Sohee Yang, Joongbo Shin, Janghoon Han, Gyeonghun Kim, Stanley Jungkyu Choi, and Minjoon Seo.”
TimeLMs: Diachronic Language Models from Twitter (2022, background): “Concerning the particular case of LMs, exposing them to new data and updating their parameters accordingly, also known as continual learning, is a promising direction, with an established tradition in machine learning (Lopez-Paz and Ranzato, 2017; Lewis et al., 2020; Lazaridou et al., 2021; Jang et ”
Fine-tuning Aligned Language Models Compromises Safety, Even When Users Do Not Intend To! (2023, background): “Other potential alternatives include regularized fine-tuning or continuous learning approaches (Jang et al., 2021; Kirkpatrick et al., 2017).”
RealTime QA: What's the Answer Right Now? (2022, background): “…and evaluation/deployment on many NLP tasks (Lazaridou et al., 2021; Röttger and Pierrehumbert, 2021; Luu et al., 2022; Onoe et al., 2022) and proposed mitigation methods for temporal adaptation (Huang and Paul, 2018, 2019; Dhingra et al., 2022; Jang et al., 2022a,b; Lee et al., 2022).”
Exploring AI Ethics of ChatGPT: A Diagnostic Analysis (2023, background): “Outdated Information: Outdated information is another type of incorrect information that may occur when a language model is trained on obsolete or inaccurate data [43].”
Entity Cloze By Date: What LMs Know About Unseen Entities (2022, background): “Newer cloze benchmarks (Dhingra et al., 2021; Jang et al., 2021) integrate temporal aspects to identify a time period when a cloze sentence is valid, but do not differentiate new and existing entities.”
Knowledge Unlearning for Mitigating Privacy Risks in Language Models (2022, background): “Another line of work simply quantifies how much knowledge is accumulated and forgotten during pretraining by extracting relational knowledge about the world (Petroni et al., 2019; Lazaridou et al., 2021; Jang et al., 2022b,a).”
StreamingQA: A Benchmark for Adaptation to New Knowledge over Time in Question Answering Models (2022, background): “Jang et al. (2021) propose new metrics for knowledge updates and establish strong baselines.”
Deep Class-Incremental Learning: A Survey (2023, unknown): “, semantic segmentation [164], [165], [166], [167], [174], object detection [175], [176], [177], [178], [179], [180], [181], video understanding [95], [133], medical surgery [182], [183], language model [184], [185], [186], [187], language-vision model [188], [189], etc.”
TemporalWiki: A Lifelong Benchmark for Training and Evaluating Ever-Evolving Language Models (2022, background): “, 2022) or (2) the update of factual information (Chen et al., 2021; Dhingra et al., 2021; Jang et al., 2022).”
Continual Learning with Foundation Models: An Empirical Study of Latent Replay (2022, background): “Additionally, language offers some interesting types of distribution shifts, such as e.g. outdated factual knowledge (Jang et al., 2021), which deserve a separate discussion.”
Knowledge Conflicts for LLMs: A Survey (2024, unknown): “Contin-8542 ual learning (CL) updates the internal knowledge through continual training on updated data (Lazari-dou et al., 2021; Jang et al., 2021, 2022).”
Lifelong Pretraining: Continually Adapting Language Models to Emerging Corpora (2021, background): “, 2021), where some of them specifically focus on methods to accumulate and update factual knowledge (Dhingra et al., 2021; Jang et al., 2021).”
A Survey on Large Language Models for Code Generation (2024, unknown): “LLMs have transformed code generation in software development as these models provide user assistance in a variety of coding tasks, such as code completion, code translation, etc.”