Subscribe to the gold package and get unlimited access to Shamra Academy
Register a new userHow Does Fine-tuning Affect the Geometry of Embedding Space: A Case Study on Isotropy
كيف تؤثر الضبط الدقيق على هندسة مساحة التضمين: دراسة حالة عن ISOTROPY
354
0 0
0.0
(
0
)
Created by
Shamra Editor
Ask ChatGPT about the research
يتم قبولها على نطاق واسع أن نماذج اللغة المدربة مسبقا بشكل كبير توفر عادة تحسينات الأداء في مهام المصب. ومع ذلك، هناك دراسات محدودة حول الأسباب الكامنة وراء هذه الفعالية، لا سيما من وجهة نظر التغييرات الهيكلية في مساحة التضمين. في محاولة لملء هذه الفجوة، في هذه الورقة، نقوم بتحليل المدى الذي يتغير فيه iSotropy من مساحة التضمين بعد ضبط الدقيقة. نوضح ذلك، على الرغم من أن ISOTROPY هي خاصية هندسية مرغوبة، لا يؤدي الضبط بشكل جيد بالضرورة إلى تحسينات ISOTROPY. علاوة على ذلك، تخضع الهياكل المحلية في تمثيلات الكلمات السياقية المدربة مسبقا (CWRS)، مثل تلك الأنواع أو التردد المميز للترميز، من تغيير هائل أثناء الضبط الجمني. تظهر تجاربنا نموا دراماتيكيا في عدد الاتجاهات الممدودة في مساحة التضمين، والتي، على النقيض من CWRS المدربة مسبقا، تحمل المعرفة اللغوية الأساسية في مساحة التضمين الدقيقة، مما يجعل أساليب تعزيز ISOTROPY الحالية غير فعالة.
It is widely accepted that fine-tuning pre-trained language models usually brings about performance improvements in downstream tasks. However, there are limited studies on the reasons behind this effectiveness, particularly from the viewpoint of structural changes in the embedding space. Trying to fill this gap, in this paper, we analyze the extent to which the isotropy of the embedding space changes after fine-tuning. We demonstrate that, even though isotropy is a desirable geometrical property, fine-tuning does not necessarily result in isotropy enhancements. Moreover, local structures in pre-trained contextual word representations (CWRs), such as those encoding token types or frequency, undergo a massive change during fine-tuning. Our experiments show dramatic growth in the number of elongated directions in the embedding space, which, in contrast to pre-trained CWRs, carry the essential linguistic knowledge in the fine-tuned embedding space, making existing isotropy enhancement methods ineffective.
References used
https://aclanthology.org/
rate research
Read More
Recent work has raised concerns about the inherent limitations of text-only pretraining. In this paper, we first demonstrate that reporting bias, the tendency of people to not state the obvious, is one of the causes of this limitation, and then inves
tigate to what extent multimodal training can mitigate this issue. To accomplish this, we 1) generate the Color Dataset (CoDa), a dataset of human-perceived color distributions for 521 common objects; 2) use CoDa to analyze and compare the color distribution found in text, the distribution captured by language models, and a human's perception of color; and 3) investigate the performance differences between text-only and multimodal models on CoDa. Our results show that the distribution of colors that a language model recovers correlates more strongly with the inaccurate distribution found in text than with the ground-truth, supporting the claim that reporting bias negatively impacts and inherently limits text-only training. We then demonstrate that multimodal models can leverage their visual training to mitigate these effects, providing a promising avenue for future research.
Natural conversations are filled with disfluencies. This study investigates if and how BERT understands disfluency with three experiments: (1) a behavioural study using a downstream task, (2) an analysis of sentence embeddings and (3) an analysis of
the attention mechanism on disfluency. The behavioural study shows that without fine-tuning on disfluent data, BERT does not suffer significant performance loss when presented disfluent compared to fluent inputs (exp1). Analysis on sentence embeddings of disfluent and fluent sentence pairs reveals that the deeper the layer, the more similar their representation (exp2). This indicates that deep layers of BERT become relatively invariant to disfluency. We pinpoint attention as a potential mechanism that could explain this phenomenon (exp3). Overall, the study suggests that BERT has knowledge of disfluency structure. We emphasise the potential of using BERT to understand natural utterances without disfluency removal.
The problem of interpretation of knowledge learned by multi-head self-attention in transformers has been one of the central questions in NLP. However, a lot of work mainly focused on models trained for uni-modal tasks, e.g. machine translation. In th
is paper, we examine masked self-attention in a multi-modal transformer trained for the task of image captioning. In particular, we test whether the multi-modality of the task objective affects the learned attention patterns. Our visualisations of masked self-attention demonstrate that (i) it can learn general linguistic knowledge of the textual input, and (ii) its attention patterns incorporate artefacts from visual modality even though it has never accessed it directly. We compare our transformer's attention patterns with masked attention in distilgpt-2 tested for uni-modal text generation of image captions. Based on the maps of extracted attention weights, we argue that masked self-attention in image captioning transformer seems to be enhanced with semantic knowledge from images, exemplifying joint language-and-vision information in its attention patterns.
In cross-lingual language models, representations for many different languages live in the same space. Here, we investigate the linguistic and non-linguistic factors affecting sentence-level alignment in cross-lingual pretrained language models for 1
01 languages and 5,050 language pairs. Using BERT-based LaBSE and BiLSTM-based LASER as our models, and the Bible as our corpus, we compute a task-based measure of cross-lingual alignment in the form of bitext retrieval performance, as well as four intrinsic measures of vector space alignment and isomorphism. We then examine a range of linguistic, quasi-linguistic, and training-related features as potential predictors of these alignment metrics. The results of our analyses show that word order agreement and agreement in morphological complexity are two of the strongest linguistic predictors of cross-linguality. We also note in-family training data as a stronger predictor than language-specific training data across the board. We verify some of our linguistic findings by looking at the effect of morphological segmentation on English-Inuktitut alignment, in addition to examining the effect of word order agreement on isomorphism for 66 zero-shot language pairs from a different corpus. We make the data and code for our experiments publicly available.
Byte-pair encoding (BPE) is a ubiquitous algorithm in the subword tokenization process of language models as it provides multiple benefits. However, this process is solely based on pre-training data statistics, making it hard for the tokenizer to han
dle infrequent spellings. On the other hand, though robust to misspellings, pure character-level models often lead to unreasonably long sequences and make it harder for the model to learn meaningful words. To alleviate these challenges, we propose a character-based subword module (char2subword) that learns the subword embedding table in pre-trained models like BERT. Our char2subword module builds representations from characters out of the subword vocabulary, and it can be used as a drop-in replacement of the subword embedding table. The module is robust to character-level alterations such as misspellings, word inflection, casing, and punctuation. We integrate it further with BERT through pre-training while keeping BERT transformer parameters fixed--and thus, providing a practical method. Finally, we show that incorporating our module to mBERT significantly improves the performance on the social media linguistic code-switching evaluation (LinCE) benchmark.
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
