Jin-woo Lee, Junhwa Choi, Bongkyu Hwang, Jinho Choo, Bogun Kim, Jeongseon Yi, Joonseok Lee, DongYoung Jung, Jaeseon Park, Kyoungwon Park, Suk-hoon Jung
| Challenge: | Recent discussions suggest that further progress will come from scaling the right structure, not merely parameters or data, while preserving acquired knowledge. |
| Approach: | They propose a width upscaling architecture that inserts lightweight expansions into linear modules while freezing all pre-trained parameters. |
| Outcome: | The proposed architecture reduces severe forgetting while learning new knowledge on a controlled synthetic biography benchmark. |
Similar Papers
Breaking Language Barriers: Cross-Lingual Continual Pre-Training at Scale (2024.emnlp-main)
Copied to clipboard
| Challenge: | Large Language Models (LLMs) have made significant strides towards Artificial General Intelligence, but training them from scratch is prohibitively expensive. |
| Approach: | They propose to continuously pre-train LLMs from existing pre-trained LLM models by using a set of parameters instead of randomly initializing them. |
| Outcome: | The proposed approach saves significant resources and accelerates convergence and performance. |
Continual Learning of Large Language Models (2025.emnlp-tutorials)
Copied to clipboard
| Challenge: | This tutorial explores the challenges of continual learning in large language models . participants will learn strategies to mitigate forgetting and manage data and evaluation pipelines . |
| Approach: | This tutorial offers a comprehensive exploration of continual learning in the context of large language models. |
| Outcome: | This tutorial explores the challenges of continual learning in large language models . participants will learn how to manage data and evaluation pipelines and adapt responsibly . |
A Survey of Post-Training Scaling in Large Language Models (2025.acl-long)
Copied to clipboard
Hanyu Lai, Xiao Liu, Junjie Gao, Jiale Cheng, Zehan Qi, Yifan Xu, Shuntian Yao, Dan Zhang, Jinhua Du, Zhenyu Hou, Xin Lv, Minlie Huang, Yuxiao Dong, Jie Tang
| Challenge: | Large language models (LLMs) have demonstrated proficiency in understanding and generating human natural languages. |
| Approach: | They propose a framework for scaling large language models using supervised fine-tuning, RLxF and test-time compute methodologies. |
| Outcome: | The proposed model can be used to understand and generate human natural languages. |
Recyclable Tuning for Continual Pre-training (2023.findings-acl)
Copied to clipboard
Yujia Qin, Cheng Qian, Xu Han, Yankai Lin, Huadong Wang, Ruobing Xie, Zhiyuan Liu, Maosong Sun, Jie Zhou
| Challenge: | Continual pre-training is the paradigm where pre-trained language models acquire fresh knowledge and gradually get upgraded. |
| Approach: | They propose to use adapted weights to recycle old PLMs for continual pre-training . they propose to combine initialization and distillation methods to achieve better performance . |
| Outcome: | The proposed method improves the convergence and performance of the upgraded PLM. |
Unlocking Continual Learning Abilities in Language Models (2024.findings-emnlp)
Copied to clipboard
| Challenge: | Existing approaches to learning models (LMs) incorporate old task data or task-wise inductive bias into LMs, but old data and accurate task information are often unavailable or costly to collect. |
| Approach: | They propose a rehearsal-free method that updates model parameters with large magnitudes . they found that the L1-normalized magnitude distribution is different when different task data is used . |
| Outcome: | The proposed method improves accuracy and performance on four CL benchmarks. |
Disentangling Continued Pre-Training: Attention-Driven Routing and Semantic Hub Preservation in Language Adaptation (2026.findings-acl)
Copied to clipboard
| Challenge: | Continued Pre-Training (CPT) enables Large Language Models (LLMs) to acquire second-language capabilities, yet the mechanisms underlying CPT remain poorly understood. |
| Approach: | They investigate how CPT adapts model representations across diverse language families and scripts, model sizes, and architectures. |
| Outcome: | The proposed model can be surgically transferred between base and CPT models with minimal loss. |
Language Scaling for Universal Suggested Replies Model (2021.naacl-industry)
Copied to clipboard
Qianlan Ying, Payal Bajaj, Budhaditya Deb, Yu Yang, Wei Wang, Bojia Lin, Milad Shokouhi, Xia Song, Yang Yang, Daxin Jiang
| Challenge: | We consider scaling automated suggested replies (SR) to multiple languages for a commercial email application. |
| Approach: | They propose a multi-lingual multi-task continual learning framework with auxiliary tasks and language adapters to train universal language representation across regions. |
| Outcome: | The proposed model reduces catastrophic forgetting and improves cross-lingual transfer across languages while reducing training costs. |
RECALL: REpresentation-aligned Catastrophic-forgetting ALLeviation via Hierarchical Model Merging (2025.emnlp-main)
Copied to clipboard
Bowen Wang, Haiyuan Wan, Liwen Shi, Chen Yang, Peng He, Yue Ma, Haochen Han, Wenhao Li, Tiao Tan, Yongjian Li, Fangming Liu, Gong Yifan, Sheng Zhang
| Challenge: | Existing models that require task labels or performance trade-offs are susceptible to catastrophic forgetting. |
| Approach: | They propose a representation-aware model merging framework for continual learning without access to historical data. |
| Outcome: | The proposed framework outperforms baselines in knowledge retention and generalization across five NLP tasks and multiple continual learning scenarios. |
Overcoming Catastrophic Forgetting in Massively Multilingual Continual Learning (2023.findings-acl)
Copied to clipboard
Genta Winata, Lingjue Xie, Karthik Radhakrishnan, Shijie Wu, Xisen Jin, Pengxiang Cheng, Mayank Kulkarni, Daniel Preotiuc-Pietro
| Challenge: | Existing methods to handle catastrophic forgetting fail to retain knowledge learnt in the past when sudden shifts occur in training data distributions. |
| Approach: | They propose a learning rate scheduling method that preserves new information without strongly overwriting past knowledge. |
| Outcome: | The proposed method preserves new information without overwriting past knowledge in a multilingual continuous learning framework. |
Rethinking Long Context Generation from the Continual Learning Perspective (2025.coling-main)
Copied to clipboard
| Challenge: | Large Language Models (LLMs) struggle with processing long contexts due to the limited context window. |
| Approach: | They propose to combine a limited context window with a continual learning perspective to improve LLMs' efficiency in processing long contexts. |
| Outcome: | The proposed models improve the performance of Large Language Models (LLMs) by integrating learning strategies with existing approaches. |