Thailand

This paper presents a novel pre-trained language models (PLM) compression approach based on the matrix product operator (short as MPO) from quantum many-body physics. It can decompose an original matrix into central tensors (containing the core information) and auxiliary tensors (with only a small proportion of parameters). With the decomposed MPO structure, we propose a novel fine-tuning strategy by only updating the parameters from the auxiliary tensors, and design an optimization algorithm for MPO-based approximation over stacked network architectures. Our approach can be applied to the original or the compressed PLMs in a general way, which derives a lighter network and significantly reduces the parameters to be fine-tuned. Extensive experiments have demonstrated the effectiveness of the proposed approach in model compression, especially the reduction in fine-tuning parameters (91% reduction on average). The code to reproduce the results of this paper can be found at https://github.com/RUCAIBox/MPOP.

ACL-IJCNLP 2021

Enabling Lightweight Fine-tuning for Pre-trained Language Model Compression based on Matrix Product Operators

lightweight fine-tuning

natural language processing

model compression

**Welcome to ACL-IJCNLP 2021!**

The great event is jointly organized by the Association for Computational Linguistics (ACL) and Asian Federation of Natural Language Processing (AFNLP). 

As in previous years, the program of the conference includes a poster session, tutorials, workshops and demonstrations in addition to the main conference.


We were able to keep the registration fees similar to those charged for the virtual ACL 2020. The one fee allows attendance at the main conference and any/all tutorials and workshops. These fees would be $125 Regular Early and $175 Regular Late; $50 Student Early and $75 Student Late. Early registration closes at midnight July 11, 2021 (Eastern Daylight Time).

**Reminder:** It is ACL’s policy that at least one author of each accepted paper (including ACL Finding papers) must register for the conference.

**Reminder2:** Underline site will open closer to the event. If you already registered you will receive access detail

Registration is now open

poster

The importance of parameter selection in supervised learning is well known. However, due to the many parameter combinations, an incomplete or an insufficient procedure is often applied. This situation may cause misleading or confusing conclusions. In this opinion paper, through an intriguing example we point out that the seriousness goes beyond what is generally recognized. In the topic of multilabel classification for medical code prediction, one influential paper conducted a proper parameter selection on a set, but when moving to a subset of frequently occurring labels, the authors used the same parameters without a separate tuning. The set of frequent labels became a popular benchmark in subsequent studies, which kept pushing the state of the art. However, we discovered that most of the results in these studies cannot surpass the approach in the original paper if a parameter tuning had been conducted at the time. Thus it is unclear how much progress the subsequent developments have actually brought. The lesson clearly indicates that without enough attention on parameter selection, the research progress in our field can be uncertain or even illusive.

Parameter Selection: Why We Should Pay More Attention to It

In this work, we demonstrate that the contextualized word vectors derived from pretrained masked language model-based encoders share a common, perhaps undesirable pattern across layers. Namely, we find cases of persistent outlier neurons within BERT and RoBERTa's hidden state vectors that consistently bear the smallest or largest values in said vectors. In an attempt to investigate the source of this information, we introduce a neuron-level analysis method, which reveals that the outliers are closely related to information captured by positional embeddings. We also pre-train the RoBERTa-base models from scratch and find that the outliers disappear without using positional embeddings. These outliers, we find, are the major cause of anisotropy of encoders' raw vector spaces, and clipping them leads to increased similarity across vectors. We demonstrate this in practice by showing that clipped vectors can more accurately distinguish word senses, as well as lead to better sentence embeddings when mean pooling. In three supervised tasks, we find that clipping does not affect the performance.

Positional Artefacts Propagate Through Masked Language Model Embeddings

We propose an alternate approach to quantifying how well language models learn natural language: we ask how well they match the statistical tendencies of natural language. To answer this question, we analyze whether text generated from language models exhibits the statistical tendencies present in the human-generated text on which they were trained. We provide a framework--paired with significance tests--for evaluating the fit of language models to these trends. We find that neural language models appear to learn only a subset of the tendencies considered, but align much more closely with empirical trends than proposed theoretical distributions (when present). Further, the fit to different distributions is highly-dependent on both model architecture and generation strategy. As concrete examples, text generated under the nucleus sampling scheme adheres more closely to the type--token relationship of natural language than text produced using standard ancestral sampling; text from LSTMs reflects the natural language distributions over length, stopwords, and symbols surprisingly well.

Language Model Evaluation Beyond Perplexity

The importance of explaining the outcome of a machine learning model, especially a black-box model, is widely acknowledged. Recent approaches explain an outcome by identifying the contributions of input features to this outcome. In environments involving large black-box models or complex inputs, this leads to computationally demanding algorithms. Further, these algorithms often suffer from low stability, with explanations varying significantly across similar examples. In this paper, we propose a Learning to Explain (L2E) approach that learns the behaviour of an underlying explanation algorithm simultaneously from all training examples. Once the explanation algorithm is distilled into an explainer network, it can be used to explain new instances. Our experiments on three classification tasks, which compare our approach to six explanation algorithms, show that L2E is between 5 and 7.5×10^4 times faster than these algorithms, while generating more stable explanations, and having comparable faithfulness to the black-box model.

Learning to Explain: Generating Stable Explanations Fast

Deep learning models have achieved great success on the task of Natural Language Inference (NLI), though only a few attempts try to explain their behaviors. Existing explanation methods usually pick prominent features such as words or phrases from the input text. However, for NLI, alignments among words or phrases are more enlightening clues to explain the model. To this end, this paper presents AREC, a post-hoc approach to generate alignment rationale explanations for co-attention based models in NLI. The explanation is based on feature selection, which keeps few but sufficient alignments while maintaining the same prediction of the target model. Experimental results show that our method is more faithful and human-readable compared with many existing approaches. We further study and re-evaluate three typical models through our explanation beyond accuracy, and propose a simple method that greatly improves the model robustness.

Alignment Rationale for Natural Language Inference

This work investigates and reveals the relationship between two closely related machine learning disciplines, namely Active Learning (AL) and Curriculum Learning (CL), from the lens of several novel curricula. This paper also introduces Active Curriculum Learning (ACL) which improves AL by combining AL with CL to benefit from the dynamic nature of the AL informativeness concept as well as the human insights used in the design of the curriculum heuristics. Comparison of the performance of ACL and AL on two public datasets for the Named Entity Recognition (NER) task shows the effectiveness of combining AL and CL using our proposed framework.

Active Curriculum Learning

We investigate if a model can learn natural language with minimal linguistic input through interaction. Addressing this question, we design and implement an interactive language learning game that learns logical semantic representations compositionally. Our game allows us to explore the benefits of logical inference for natural language learning. Evaluation shows that the model can accurately narrow down potential logical representations for words over the course of the game, suggesting that our model is able to learn lexical mappings from scratch successfully.

SHAPELURN: An Interactive Language Learning Game with Logical Inference

Easy access, variety of content, and fast widespread interactions are some of the reasons that have made social media increasingly popular in today's society. However, this has also enabled the widespread propagation of fake news, text that is published with an intent to spread misinformation and sway beliefs. Detecting fake news is important to prevent misinformation and maintain a healthy society. 

While prior works have tackled this problem by building supervised learning systems, automatedly modeling the social media landscape that enables the spread of fake news is challenging. On the contrary, having humans fact check all news is not scalable. Thus, in this paper, we propose to approach this problem interactively, where human insight can be continually combined with an automated system, enabling better social media representation quality. Our experiments show performance improvements in this setting.

Tackling Fake News Detection by Interactively Learning Representations using Graph Neural Networks

Timeline Summarisation (TLS) aims to generate a concise, time-ordered list of events described in sources such as news articles. However, current systems do not provide an adequate way to adapt to new domains nor to focus on the aspects of interest to a particular user. Therefore, we propose a method for interactively learning abstractive TLS using Reinforcement Learning (RL). We define a compound reward function and use RL to fine-tune an abstractive Multi-document Summarisation (MDS) model, which avoids the need to train using reference summaries. One of the sub-reward functions will be learned interactively from user feedback to ensure the consistency between users’ demands and the generated timeline. The other sub-reward functions contribute to topical coherence and linguistic fluency. We plan experiments to evaluate whether our approach could generate accurate and precise timelines tailored for each user.

A Proposal: Interactively Learning to Summarise Timelines by Reinforcement Learning

We investigate the question of how adaptive feedback from a virtual agent impacts the linguistic input of the user in a shared world game environment. To do so, we carry out an exploratory pilot study to observe how individualized linguistic feedback affects the user's speech input. We introduce a speech-controlled game, Apple Core-dination, in which an agent learns complex tasks using a base knowledge of simple actions. 
The agent is equipped with a learning mechanism for mapping new commands to sequences of simple actions, as well as the ability to incorporate user input into written responses. The agent repeatedly shares its internal knowledge state by responding to what it knows and does not know about language meaning and the shared environment. 
We focus on the linguistic feedback loop in order to analyze the nature of user input. Feedback from the agent is provided in the form of visual movement and written linguistic responses. Particular attention is given to incorporating user input into agent responses and updating the speech-to-action mappings based on commands provided by the user. Through our pilot study, we analyze task success and compare the lexical features of user input. Results show variation in input length and lexical variety across users, suggesting a correlation between the two that can be studied further.

Enabling Lightweight Fine-tuning for Pre-trained Language Model Compression based on Matrix Product Operators

Downloads

Next from ACL-IJCNLP 2021

Parameter Selection: Why We Should Pay More Attention to It

Similar lecture

BinaryBERT: Pushing the Limit of BERT Quantization

Stay up to date with the latest Underline news!

PRESENTATIONS

CONFERENCES

COMPANY

RESOURCES