Subscribe to the gold package and get unlimited access to Shamra Academy
Register a new userIncorporating Residual and Normalization Layers into Analysis of Masked Language Models
دمج طبقات المتبقية والتطبيع في تحليل نماذج اللغة الملثمين
260
0 0
0.0
(
0
)
Created by
Shamra Editor
Ask ChatGPT about the research
أصبحت بنية المحولات في كل مكان في مجال معالجة اللغات الطبيعية.لتفسير النماذج القائمة على المحولات، تم تحليل أنماط اهتمامها على نطاق واسع.ومع ذلك، فإن بنية المحولات لا تتكون فقط من الاهتمام متعدد الأطراف؛يمكن أن تسهم مكونات أخرى أيضا في الأداء التدريجي المحولات.في هذه الدراسة، مددنا نطاق تحليل المحولات من أنماط الانتباه فقط إلى كتلة الاهتمام بأكمله، أي اهتمام متعدد الأطراف، والاتصال المتبقي، وتطبيع الطبقة.يوضح تحليل النماذج اللغوية المقصودة للمحولات أن التفاعل الرمزي إلى الرمز المميز الذي يؤديه عن طريق الاهتمام له تأثير أقل على التمثيل الوسيط مما كان مفترض سابقا.توفر هذه النتائج تفسيرات جديدة بديهية للتقارير القائمة؛على سبيل المثال، تم تجاهل أنماط الانتباه المستفادة لا تؤثر سلبا على الأداء.رموز تجاربنا متاحة للجمهور.
Transformer architecture has become ubiquitous in the natural language processing field. To interpret the Transformer-based models, their attention patterns have been extensively analyzed. However, the Transformer architecture is not only composed of the multi-head attention; other components can also contribute to Transformers' progressive performance. In this study, we extended the scope of the analysis of Transformers from solely the attention patterns to the whole attention block, i.e., multi-head attention, residual connection, and layer normalization. Our analysis of Transformer-based masked language models shows that the token-to-token interaction performed via attention has less impact on the intermediate representations than previously assumed. These results provide new intuitive explanations of existing reports; for example, discarding the learned attention patterns tends not to adversely affect the performance. The codes of our experiments are publicly available.
References used
https://aclanthology.org/
rate research
Read More
In natural language generation tasks, a neural language model is used for generating a sequence of words forming a sentence. The topmost weight matrix of the language model, known as the classification layer, can be viewed as a set of vectors, each r
epresenting a target word from the target dictionary. The target word vectors, along with the rest of the model parameters, are learned and updated during training. In this paper, we analyze the properties encoded in the target vectors and question the necessity of learning these vectors. We suggest to randomly draw the target vectors and set them as fixed so that no weights updates are being made during training. We show that by excluding the vectors from the optimization, the number of parameters drastically decreases with a marginal effect on the performance. We demonstrate the effectiveness of our method in image-captioning and machine-translation.
In this paper, we propose a knowledge infusion mechanism to incorporate domain knowledge into language transformers. Weakly supervised data is regarded as the main source for knowledge acquisition. We pre-train the language models to capture masked k
nowledge of focuses and aspects and then fine-tune them to obtain better performance on the downstream tasks. Due to the lack of publicly available datasets for multi-label classification of Chinese medical questions, we crawled questions from medical question/answer forums and manually annotated them using eight predefined classes: persons and organizations, symptom, cause, examination, disease, information, ingredient, and treatment. Finally, a total of 1,814 questions with 2,340 labels. Each question contains an average of 1.29 labels. We used Baidu Medical Encyclopedia as the knowledge resource. Two transformers BERT and RoBERTa were implemented to compare performance on our constructed datasets. Experimental results showed that our proposed model with knowledge infusion mechanism can achieve better performance, no matter which evaluation metric including Macro F1, Micro F1, Weighted F1 or Subset Accuracy were considered.
Large pretrained language models using the transformer neural network architecture are becoming a dominant methodology for many natural language processing tasks, such as question answering, text classification, word sense disambiguation, text comple
tion and machine translation. Commonly comprising hundreds of millions of parameters, these models offer state-of-the-art performance, but at the expense of interpretability. The attention mechanism is the main component of transformer networks. We present AttViz, a method for exploration of self-attention in transformer networks, which can help in explanation and debugging of the trained models by showing associations between text tokens in an input sequence. We show that existing deep learning pipelines can be explored with AttViz, which offers novel visualizations of the attention heads and their aggregations. We implemented the proposed methods in an online toolkit and an offline library. Using examples from news analysis, we demonstrate how AttViz can be used to inspect and potentially better understand what a model has learned.
Previous work has indicated that pretrained Masked Language Models (MLMs) are not effective as universal lexical and sentence encoders off-the-shelf, i.e., without further task-specific fine-tuning on NLI, sentence similarity, or paraphrasing tasks u
sing annotated task data. In this work, we demonstrate that it is possible to turn MLMs into effective lexical and sentence encoders even without any additional data, relying simply on self-supervision. We propose an extremely simple, fast, and effective contrastive learning technique, termed Mirror-BERT, which converts MLMs (e.g., BERT and RoBERTa) into such encoders in 20-30 seconds with no access to additional external knowledge. Mirror-BERT relies on identical and slightly modified string pairs as positive (i.e., synonymous) fine-tuning examples, and aims to maximise their similarity during identity fine-tuning''. We report huge gains over off-the-shelf MLMs with Mirror-BERT both in lexical-level and in sentence-level tasks, across different domains and different languages. Notably, in sentence similarity (STS) and question-answer entailment (QNLI) tasks, our self-supervised Mirror-BERT model even matches the performance of the Sentence-BERT models from prior work which rely on annotated task data. Finally, we delve deeper into the inner workings of MLMs, and suggest some evidence on why this simple Mirror-BERT fine-tuning approach can yield effective universal lexical and sentence encoders.
Recent studies have revealed a security threat to natural language processing (NLP) models, called the Backdoor Attack. Victim models can maintain competitive performance on clean samples while behaving abnormally on samples with a specific trigger w
ord inserted. Previous backdoor attacking methods usually assume that attackers have a certain degree of data knowledge, either the dataset which users would use or proxy datasets for a similar task, for implementing the data poisoning procedure. However, in this paper, we find that it is possible to hack the model in a data-free way by modifying one single word embedding vector, with almost no accuracy sacrificed on clean samples. Experimental results on sentiment analysis and sentence-pair classification tasks show that our method is more efficient and stealthier. We hope this work can raise the awareness of such a critical security risk hidden in the embedding layers of NLP models. Our code is available at https://github.com/lancopku/Embedding-Poisoning.
suggested questions
Log in to be able to interact and post comments
comments
Fetching comments
Sign in to be able to follow your search criteria
