Deep pretrained language models have achieved great success in the way of pretraining first and then fine-tuning. But such a sequential transfer learning paradigm often confronts the catastrophic forgetting problem and leads to sub-optimal performance. To fine-tune with less forgetting, we propose a recall and learn mechanism, which adopts the idea of multi-task learning and jointly learns pretraining tasks and downstream tasks. Specifically, we propose a Pretraining Simulation mechanism to recall the knowledge from pretraining tasks without data, and an Objective Shifting mechanism to focus the learning on downstream tasks gradually. Experiments show that our method achieves state-of-the-art performance on the GLUE benchmark. Our method also enables BERT-base to achieve better performance than directly fine-tuning of BERT-large. Further, we provide the open-source RecAdam optimizer, which integrates the proposed mechanisms into Adam optimizer, to facility the NLP community.
RECALL: REpresentation-aligned Catastrophic-forgetting ALLeviation via Hierarchical Model Merging
Shenzhen International Graduate School, Tsinghua University（清华大学深圳国际研究生院）；Peng Cheng Laboratory（鹏城实验室）；Huazhong University of Science and Technology（华中科技大学）；Xiamen University（厦门大学）；The Hong Kong University of Science and Technology, Guangzhou（香港科技大学（广州））；School of Biomedical Engineering, Tsinghua University（清华大学生物医学工程学院）
Pre-trained language models (LM) such as BERT, DistilBERT, and RoBERTa can be tuned for different domains (domain-tuning) by continuing the pre-training phase on a new target domain corpus. This simple domain tuning (SDT) technique has been widely used to create domain-tuned models such as BioBERT, SciBERT and ClinicalBERT. However, during the pretraining phase on the target domain, the LM models may catastrophically forget the patterns learned from their source domain. In this work, we study the effects of catastrophic forgetting on domain-tuned LM models and investigate methods that mitigate…
Westlake University（西湖大学）；Technical University of Darmstadt（达姆施塔特工业大学）
Pretrained language models (PLM) have recently advanced graph-to-text generation, where the input graph is linearized into a sequence and fed into the PLM to obtain its representation. However, efficiently encoding the graph structure in PLMs is challenging because such models were pretrained on natural language, and modeling structured data may lead to catastrophic forgetting of distributional knowledge. In this paper, we propose StructAdapt, an adapter method to encode graph structure into PLMs. Contrary to prior work, StructAdapt effectively models interactions among the nodes based on the …
University of Chicago（芝加哥大学）；University of Southern California（南加州大学）
Fine-tuning over large pretrained language models (PLMs) has established many state-of-the-art results. Despite its superior performance, such fine-tuning can be unstable, resulting in significant variance in performance and potential risks for practical applications. Previous works have attributed such instability to the catastrophic forgetting problem in the top layers of PLMs, which indicates iteratively that fine-tuning layers in a top-down manner is a promising solution. In this paper, we first point out that this method does not always work out due to the different convergence speeds of …
Large-scale pretrained language models are surprisingly good at recalling factual knowledge presented in the training corpus. In this paper, we present preliminary studies on how factual knowledge is stored in pretrained Transformers by introducing the concept of knowledge neurons. Specifically, we examine the fill-in-the-blank cloze task for BERT. Given a relational fact, we propose a knowledge attribution method to identify the neurons that express the fact. We find that the activation of such knowledge neurons is positively correlated to the expression of their corresponding facts. In our c…
University of Cambridge（剑桥大学）；University of Glasgow（格拉斯哥大学）
Acquiring factual knowledge with Pretrained Language Models (PLMs) has attracted increasing attention, showing promising performance in many knowledge-intensive tasks. Their good performance has led the community to believe that the models do possess a modicum of reasoning competence rather than merely memorising the knowledge. In this paper, we conduct a comprehensive evaluation of the learnable deductive (also known as explicit) reasoning capability of PLMs. Through a series of controlled experiments, we posit two main findings. (i) PLMs inadequately generalise learned logic rules and perfor…
When Forgetting is not Catastrophic: On the Mechanics of Spurious Forgetting
MPI for Intelligent Systems, Tübingen（马克斯·普朗克智能系统研究所（蒂宾根））；Jinesis Lab, University of Toronto & Vector Institute（Jinesis实验室，多伦多大学与向量研究所）；EuroSafeAI；Hector Foundation（赫克托基金会）；Stanford University（斯坦福大学）；ELLIS Institute Tübingen（ELLIS研究所（蒂宾根））
National Technical University of Athens（雅典国家技术大学）；Behavioral Signal Technologies（行为信号技术公司）
A growing number of state-of-the-art transfer learning methods employ language models pretrained on large generic corpora. In this paper we present a conceptually simple and effective transfer learning approach that addresses the problem of catastrophic forgetting. Specifically, we combine the task-specific optimization function with an auxiliary language model objective, which is adjusted during the training process. This preserves language regularities captured by language models, while enabling sufficient adaptation for solving the target task. Our method does not require pretraining or fin…
National University of Singapore（新加坡国立大学）；DAMO Academy, Alibaba Group（阿里巴巴集团达摩院）；Nanyang Technological University（南洋理工大学）；Singapore University of Technology and Design（新加坡科技设计大学）
Adapter-based tuning has recently arisen as an alternative to fine-tuning. It works by adding light-weight adapter modules to a pretrained language model (PrLM) and only updating the parameters of adapter modules when learning on a downstream task. As such, it adds only a few trainable parameters per new task, allowing a high degree of parameter sharing. Prior studies have shown that adapter-based tuning often achieves comparable results to fine-tuning. However, existing work only focuses on the parameter-efficient aspect of adapter-based tuning while lacking further investigation on its effec…
Bar Ilan University（巴伊兰大学）；Allen Institute for Artificial Intelligence（艾伦人工智能研究所）
The notion of "in-domain data" in NLP is often over-simplistic and vague, as textual data varies in many nuanced linguistic aspects such as topic, style or level of formality. In addition, domain labels are many times unavailable, making it challenging to build domain-specific systems. We show that massive pre-trained language models implicitly learn sentence representations that cluster by domains without supervision -- suggesting a simple data-driven definition of domains in textual data. We harness this property and propose domain data selection methods based on such models, which require o…
Center for Information and Language Processing (CIS)（信息与语言处理中心（CIS））；LMU Munich（慕尼黑大学）
Khandelwal et al. (2020) use a k-nearest-neighbor (kNN) component to improve language model performance. We show that this idea is beneficial for open-domain question answering (QA). To improve the recall of facts encountered during training, we combine BERT (Devlin et al., 2019) with a traditional information retrieval step (IR) and a kNN search over a large datastore of an embedded text collection. Our contributions are as follows: i) BERT-kNN outperforms BERT on cloze-style QA by large margins without any further training. ii) We show that BERT often identifies the correct response category…
College of Computer Science and Technology（计算机科学与技术学院）；Nanjing University of Aeronautics and Astronautics（南京航空航天大学）；MIIT Key Laboratory of Pattern Analysis and Machine Intelligence（工业和信息化部模式分析与机器智能重点实验室）
The task of Chinese Spelling Check (CSC) is aiming to detect and correct spelling errors that can be found in the text. While manually annotating a high-quality dataset is expensive and time-consuming, thus the scale of the training dataset is usually very small (e.g., SIGHAN15 only contains 2339 samples for training), therefore supervised-learning based models usually suffer the data sparsity limitation and over-fitting issue, especially in the era of big language models. In this paper, we are dedicated to investigating the \textbf{unsupervised} paradigm to address the CSC problem and we prop…
MLP Memory: A Retriever-Pretrained Memory for Large Language Models
Shanghai Jiao Tong University（上海交通大学）；Shanghai AI Laboratory（上海人工智能实验室）；Tsinghua University（清华大学）；SJTU Paris Elite Institute of Technology（上海交通大学巴黎卓越工程师学院）
Time-Incremental Continued Pretraining of LLMs: Knowledge Updates Without Catastrophic Forgetting
Concordia University（康考迪亚大学）；Mila – Quebec AI Institute（米拉-魁北克人工智能研究所）；Université de Montréal（蒙特利尔大学）；Laval University（拉瓦尔大学）；University of Tübingen（蒂宾根大学）；Tübingen AI Center（蒂宾根人工智能中心）
Keyphrase extraction is the process of automatically selecting a small set of most relevant phrases from a given text. Supervised keyphrase extraction approaches need large amounts of labeled training data and perform poorly outside the domain of the training data. In this paper, we present PatternRank, which leverages pretrained language models and part-of-speech for unsupervised keyphrase extraction from single documents. Our experiments show PatternRank achieves higher precision, recall and F1-scores than previous state-of-the-art approaches. In addition, we present the KeyphraseVectorizers…
Memory Decoder: A Pretrained, Plug-and-Play Memory for Large Language Models
LUMIA Lab（LUMIA实验室）；School of Artificial Intelligence（人工智能学院）；Shanghai Jiao Tong University（上海交通大学）；Shanghai AI Laboratory（上海人工智能实验室）；Tsinghua University（清华大学）；SJTU Paris Elite Institute of Technology（上海交通大学巴黎精英技术学院）
ShanghaiTech University（上海科技大学）；Shanghai Engineering Research Center of Intelligent Vision and Imaging（上海智能视觉与成像工程研究中心）
Ultra-fine entity typing (UFET) predicts extremely free-formed types (e.g., president, politician) of a given entity mention (e.g., Joe Biden) in context. State-of-the-art (SOTA) methods use the cross-encoder (CE) based architecture. CE concatenates the mention (and its context) with each type and feeds the pairs into a pretrained language model (PLM) to score their relevance. It brings deeper interaction between mention and types to reach better performance but has to perform N (type set size) forward passes to infer types of a single mention. CE is therefore very slow in inference when the t…
University of Sussex（萨塞克斯大学）；Naver Labs Europe（Naver欧洲实验室）
Machine learning is shifting towards general-purpose pretrained generative models, trained in a self-supervised manner on large amounts of data, which can then be applied to solve a large number of tasks. However, due to their generic training methodology, these models often fail to meet some of the downstream requirements (e.g., hallucinations in abstractive summarization or style violations in code generation). This raises the important question of how to adapt pre-trained generative models to meet all requirements without destroying their general capabilities ("catastrophic forgetting"). Re…
Preserving Knowledge in Large Language Model with Model-Agnostic Self-Decompression
College of Computer Science and Technology, Zhejiang University（浙江大学计算机科学与技术学院）；Binjiang Research Institute of Zhejiang University（浙江大学滨江研究院）；Linker Technology Research Co. Ltd（领科技术研究有限公司）
Institute of Computing Technology, Chinese Academy of Sciences（中国科学院计算技术研究所）；University of Chinese Academy of Sciences（中国科学院大学）；Tencent（腾讯）
This paper considers continual learning of large-scale pretrained neural machine translation model without accessing the previous training data or introducing model separation. We argue that the widely used regularization-based methods, which perform multi-objective learning with an auxiliary loss, suffer from the misestimate problem and cannot always achieve a good balance between the previous and new tasks. To solve the problem, we propose a two-stage training method based on the local features of the real loss. We first search low forgetting risk regions, where the model can retain the perf…
The forgetting curve has been extensively explored by psychologists, educationalists and cognitive scientists alike. In the context of Intelligent Tutoring Systems, modelling the forgetting curve for each user and knowledge component (e.g. vocabulary word) should enable us to develop optimal revision strategies that counteract memory decay and ensure long-term retention. In this study we explore a variety of forgetting curve models incorporating psychological and linguistic features, and we use these models to predict the probability of word recall by learners of English as a second language. …
Information Sciences Institute（信息科学研究所）；University of Southern California（南加州大学）
We study the power of cross-attention in the Transformer architecture within the context of transfer learning for machine translation, and extend the findings of studies into cross-attention when training from scratch. We conduct a series of experiments through fine-tuning a translation model on data where either the source or target language has changed. These experiments reveal that fine-tuning only the cross-attention parameters is nearly as effective as fine-tuning all parameters (i.e., the entire translation model). We provide insights into why this is the case and observe that limiting f…
MIT（麻省理工学院）；Facebook（脸书）；New York University（纽约大学）
In this work, we study how the finetuning stage in the pretrain-finetune framework changes the behavior of a pretrained neural language generator. We focus on the transformer encoder-decoder model for the open-domain dialogue response generation task. Our major finding is that after standard finetuning, the model forgets some of the important language generation skills acquired during large-scale pretraining. We demonstrate the forgetting phenomenon through a set of detailed behavior analysis from the perspectives of knowledge transfer, context sensitivity, and function space projection. As a …
New York University（纽约大学）；University of Colorado Boulder（科罗拉多大学博尔德分校）
While pretrained models such as BERT have shown large gains across natural language understanding tasks, their performance can be improved by further training the model on a data-rich intermediate task, before fine-tuning it on a target task. However, it is still poorly understood when and why intermediate-task training is beneficial for a given target task. To investigate this, we perform a large-scale study on the pretrained RoBERTa model with 110 intermediate-target task combinations. We further evaluate all trained models with 25 probing tasks meant to reveal the specific skills that drive…
Data and Web Science Group, University of Mannheim（曼海姆大学数据与网络科学组）；Wluper；Ubiquitous Knowledge Processing (UKP) Lab, TU Darmstadt（达姆施塔特工业大学普适知识处理（UKP）实验室）
Following the major success of neural language models (LMs) such as BERT or GPT-2 on a variety of language understanding tasks, recent work focused on injecting (structured) knowledge from external resources into these models. While on the one hand, joint pretraining (i.e., training from scratch, adding objectives based on external knowledge to the primary LM objective) may be prohibitively computationally expensive, post-hoc fine-tuning on external knowledge, on the other hand, may lead to the catastrophic forgetting of distributional knowledge. In this work, we investigate models for complem…
