Subscribe to the gold package and get unlimited access to Shamra Academy
Register a new userExposure Bias versus Self-Recovery: Are Distortions Really Incremental for Autoregressive Text Generation?
التعرض التحيز مقابل الاسترداد الذاتي: هل تشوهات تدريجية حقا لتوليد النص التلقائي؟
127
0 0
0.0
(
0
)
Created by
Shamra Editor
Ask ChatGPT about the research
تم اعتبار تحيز التعرض مشكلة مركزية لنماذج اللغة التراجعية التلقائية (LM). وهذا يدعي أن المعلم يجبر سيؤدي إلى تشويه جيل وقت الاختبار تدريجيا بسبب تباين توليد التدريب. على الرغم من أن الكثير من الخوارزميات قد اقترحت تجنب التغلب على المعلم وبالتالي تخفيف تحيز التعرض، إلا أن هناك ضئيل من العمل الذي يظهر مدى خطورة مشكلة تحيز التعرض في الواقع. في هذا العمل، نركز على مهمة توليد اللغة المفتوح العضوية، اقترح المقاييس لتحديد تأثير تحيز التعرض في جوانب الجودة والتنوع والاتساق. الحدس الرئيسي الخاص بنا هو أنه إذا قمت بإطعام بادئات بيانات الحقيقة الأرضية (بدلا من البادئات الناتجة عن النموذج نفسه) في النموذج واطلب من مواصلة الجيل، يجب أن يصبح الأداء أفضل بكثير لأن التناقض في توليد التدريب في البادئة إزالة. يتم إجراء التقييمات التلقائية والإنسانية في تجاربنا. على عكس الاعتقاد الشعبي بتحيز التعرض، نجد أن التشويه الناجم عن تناقض البادئة محدود، ولا يبدو أنه تدريجي خلال الجيل. علاوة على ذلك، يكشف تحليلنا عن قدرة مثيرة للاهتمام للانكماش الذاتي ل LM، والتي نفترض أنها تكافؤها لتكون مواجهة الآثار الضارة من تحيز التعرض.
Exposure bias has been regarded as a central problem for auto-regressive language models (LM). It claims that teacher forcing would cause the test-time generation to be incrementally distorted due to the training-generation discrepancy. Although a lot of algorithms have been proposed to avoid teacher forcing and therefore alleviate exposure bias, there is little work showing how serious the exposure bias problem actually is. In this work, we focus on the task of open-ended language generation, propose metrics to quantify the impact of exposure bias in the aspects of quality, diversity, and consistency. Our key intuition is that if we feed ground-truth data prefixes (instead of prefixes generated by the model itself) into the model and ask it to continue the generation, the performance should become much better because the training-generation discrepancy in the prefix is removed. Both automatic and human evaluations are conducted in our experiments. On the contrary to the popular belief in exposure bias, we find that the the distortion induced by the prefix discrepancy is limited, and does not seem to be incremental during the generation. Moreover, our analysis reveals an interesting self-recovery ability of the LM, which we hypothesize to be countering the harmful effects from exposure bias.
References used
https://aclanthology.org/
rate research
Read More
Internet search affects people's cognition of the world, so mitigating biases in search results and learning fair models is imperative for social good. We study a unique gender bias in image search in this work: the search images are often gender-imb
alanced for gender-neutral natural language queries. We diagnose two typical image search models, the specialized model trained on in-domain datasets and the generalized representation model pre-trained on massive image and text data across the internet. Both models suffer from severe gender bias. Therefore, we introduce two novel debiasing approaches: an in-processing fair sampling method to address the gender imbalance issue for training models, and a post-processing feature clipping method base on mutual information to debias multimodal representations of pre-trained models. Extensive experiments on MS-COCO and Flickr30K benchmarks show that our methods significantly reduce the gender bias in image search models.
In this paper, we present the joint contribution of Unbabel and IST to the WMT 2021 Metrics Shared Task. With this year's focus on Multidimensional Quality Metric (MQM) as the ground-truth human assessment, our aim was to steer COMET towards higher c
orrelations with MQM. We do so by first pre-training on Direct Assessments and then fine-tuning on z-normalized MQM scores. In our experiments we also show that reference-free COMET models are becoming competitive with reference-based models, even outperforming the best COMET model from 2020 on this year's development data. Additionally, we present COMETinho, a lightweight COMET model that is 19x faster on CPU than the original model, while also achieving state-of-the-art correlations with MQM. Finally, in the QE as a metric'' track, we also participated with a QE model trained using the OpenKiwi framework leveraging MQM scores and word-level annotations.
Abstract Text classification is a widely studied problem and has broad applications. In many real-world problems, the number of texts for training classification models is limited, which renders these models prone to overfitting. To address this prob
lem, we propose SSL-Reg, a data-dependent regularization approach based on self-supervised learning (SSL). SSL (Devlin et al., 2019a) is an unsupervised learning approach that defines auxiliary tasks on input data without using any human-provided labels and learns data representations by solving these auxiliary tasks. In SSL-Reg, a supervised classification task and an unsupervised SSL task are performed simultaneously. The SSL task is unsupervised, which is defined purely on input texts without using any human- provided labels. Training a model using an SSL task can prevent the model from being overfitted to a limited number of class labels in the classification task. Experiments on 17 text classification datasets demonstrate the effectiveness of our proposed method. Code is available at https://github.com/UCSD-AI4H/SSReg.
Abstract ⚠ This paper contains prompts and model outputs that are offensive in nature. When trained on large, unfiltered crawls from the Internet, language models pick up and reproduce all kinds of undesirable biases that can be found in the data: Th
ey often generate racist, sexist, violent, or otherwise toxic language. As large models require millions of training examples to achieve good performance, it is difficult to completely prevent them from being exposed to such content. In this paper, we first demonstrate a surprising finding: Pretrained language models recognize, to a considerable degree, their undesirable biases and the toxicity of the content they produce. We refer to this capability as self-diagnosis. Based on this finding, we then propose a decoding algorithm that, given only a textual description of the undesired behavior, reduces the probability of a language model producing problematic text. We refer to this approach as self-debiasing. Self-debiasing does not rely on manually curated word lists, nor does it require any training data or changes to the model's parameters. While we by no means eliminate the issue of language models generating biased text, we believe our approach to be an important step in this direction.1
Due to efficient end-to-end training and fluency in generated texts, several encoder-decoder framework-based models are recently proposed for data-to-text generations. Appropriate encoding of input data is a crucial part of such encoder-decoder model
s. However, only a few research works have concentrated on proper encoding methods. This paper presents a novel encoder-decoder based data-to-text generation model where the proposed encoder carefully encodes input data according to underlying structure of the data. The effectiveness of the proposed encoder is evaluated both extrinsically and intrinsically by shuffling input data without changing meaning of that data. For selecting appropriate content information in encoded data from encoder, the proposed model incorporates attention gates in the decoder. With extensive experiments on WikiBio and E2E dataset, we show that our model outperforms the state-of-the models and several standard baseline systems. Analysis of the model through component ablation tests and human evaluation endorse the proposed model as a well-grounded system.
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
