Training Question Answering Models From Synthetic Data (Research Paper Walkthrough)
#nlp #questionanswering #dataaugmentation
Can a Q/A model trained on just the Synthetic Data beat SOTA? This research focuses on improving question answering models by generating synthetic questions and answers when dealing with the limited amount of human-annotated data. Author’s achieve better performance on SQUAD1.1 question answering task solely with synthetic data compared to human-annotated questions from the training set. Watch to know more :-)
⏩ Abstract: Question and answer generation is a data augmentation method that aims to improve question answering (QA) models given the limited amount of human labeled data. However, a considerable gap remains between synthetic and human-generated question-answer pairs. This work aims to narrow this gap by taking advantage of large language models and explores several factors such as model size, quality of pretrained models, scale of data synthesized, and algorithmic choices. On the SQuAD1.1 question answering task, we achieve higher accuracy using solely synthetic questions and answers than when using the SQuAD1.1 training set questions alone. Removing access to real Wikipedia data, we synthesize questions and answers from a synthetic corpus generated by an 8.3 billion parameter GPT-2 model. With no access to human supervision and only access to other models, we are able to train state of the art question answering networks on entirely model-generated data that achieve 88.4 Exact Match (EM) and 93.9 F1 score on the SQuAD1.1 dev set. We further apply our methodology to SQuAD2.0 and show a 2.8 absolute gain on EM score compared to prior work using synthetic data.
Please feel free to share out the content and subscribe to my channel :)
⏩ Subscribe - https://youtube.com/channel/UCoz8NrwgL7U9535VNc0mRPA?sub_confirmation=1
⏩ OUTLINE:
0:00 - Abstract & Background
02:02 - Proposed Method (3-step pipeline)
03:31 - Algorithm (Pipeline for generating and evaluating synthetic data)
05:05 - Answer Generation - Equations
06:03 - DataFlow
07:30 - Question Generation
09:05 - Roundtrip filtration
09:37 - Wrap-up
⏩ Paper Title: Training Question Answering Models From Synthetic Data
⏩ Paper: https://arxiv.org/pdf/2002.09599.pdf
⏩ Author: Raul Puri, Ryan Spring, Mostofa Patwary, Mohammad Shoeybi, Bryan Catanzaro
⏩ Organisation: Nvidia, Rice University
Blog: https://towardsdatascience.com/training-question-answering-models-from-synthetic-data-research-paper-summary-2220186703f
Research Paper Summary Playlist: https://www.youtube.com/watch?v=ykClwtoLER8&list=PLsAqq9lZFOtWUz1WEoJ3GXw197LD7BxMc
BERT in NLP Playlist: https://www.youtube.com/watch?v=uhnKsGDyhEg&list=PLsAqq9lZFOtX-WN8lldIOI7p-p0lBzjtY
NLP Data Augmentation Playlist: https://www.youtube.com/watch?v=9O9scQb4sNo&list=PLsAqq9lZFOtUg63g_95OuV-R2GhV1UiIZ
*********************************************
If you want to support me financially which totally optional and voluntary :) ❤️
You can consider buying me chai ( because i don't drink coffee :) ) at https://www.buymeacoffee.com/TechvizCoffee
*********************************************
⏩ Youtube - https://www.youtube.com/c/TechVizTheDataScienceGuy
⏩ Blog - https://prakhartechviz.blogspot.com
⏩ LinkedIn - https://linkedin.com/in/prakhar21
⏩ Medium - https://medium.com/@prakhar.mishra
⏩ GitHub - https://github.com/prakhar21
⏩ Twitter - https://twitter.com/rattller
*********************************************
Tools I use for making videos :)
⏩ iPad - https://tinyurl.com/y39p6pwc
⏩ Apple Pencil - https://tinyurl.com/y5rk8txn
⏩ GoodNotes - https://tinyurl.com/y627cfsa
#techviz #datascienceguy #research #qa #arxiv
Видео Training Question Answering Models From Synthetic Data (Research Paper Walkthrough) канала TechViz - The Data Science Guy
Can a Q/A model trained on just the Synthetic Data beat SOTA? This research focuses on improving question answering models by generating synthetic questions and answers when dealing with the limited amount of human-annotated data. Author’s achieve better performance on SQUAD1.1 question answering task solely with synthetic data compared to human-annotated questions from the training set. Watch to know more :-)
⏩ Abstract: Question and answer generation is a data augmentation method that aims to improve question answering (QA) models given the limited amount of human labeled data. However, a considerable gap remains between synthetic and human-generated question-answer pairs. This work aims to narrow this gap by taking advantage of large language models and explores several factors such as model size, quality of pretrained models, scale of data synthesized, and algorithmic choices. On the SQuAD1.1 question answering task, we achieve higher accuracy using solely synthetic questions and answers than when using the SQuAD1.1 training set questions alone. Removing access to real Wikipedia data, we synthesize questions and answers from a synthetic corpus generated by an 8.3 billion parameter GPT-2 model. With no access to human supervision and only access to other models, we are able to train state of the art question answering networks on entirely model-generated data that achieve 88.4 Exact Match (EM) and 93.9 F1 score on the SQuAD1.1 dev set. We further apply our methodology to SQuAD2.0 and show a 2.8 absolute gain on EM score compared to prior work using synthetic data.
Please feel free to share out the content and subscribe to my channel :)
⏩ Subscribe - https://youtube.com/channel/UCoz8NrwgL7U9535VNc0mRPA?sub_confirmation=1
⏩ OUTLINE:
0:00 - Abstract & Background
02:02 - Proposed Method (3-step pipeline)
03:31 - Algorithm (Pipeline for generating and evaluating synthetic data)
05:05 - Answer Generation - Equations
06:03 - DataFlow
07:30 - Question Generation
09:05 - Roundtrip filtration
09:37 - Wrap-up
⏩ Paper Title: Training Question Answering Models From Synthetic Data
⏩ Paper: https://arxiv.org/pdf/2002.09599.pdf
⏩ Author: Raul Puri, Ryan Spring, Mostofa Patwary, Mohammad Shoeybi, Bryan Catanzaro
⏩ Organisation: Nvidia, Rice University
Blog: https://towardsdatascience.com/training-question-answering-models-from-synthetic-data-research-paper-summary-2220186703f
Research Paper Summary Playlist: https://www.youtube.com/watch?v=ykClwtoLER8&list=PLsAqq9lZFOtWUz1WEoJ3GXw197LD7BxMc
BERT in NLP Playlist: https://www.youtube.com/watch?v=uhnKsGDyhEg&list=PLsAqq9lZFOtX-WN8lldIOI7p-p0lBzjtY
NLP Data Augmentation Playlist: https://www.youtube.com/watch?v=9O9scQb4sNo&list=PLsAqq9lZFOtUg63g_95OuV-R2GhV1UiIZ
*********************************************
If you want to support me financially which totally optional and voluntary :) ❤️
You can consider buying me chai ( because i don't drink coffee :) ) at https://www.buymeacoffee.com/TechvizCoffee
*********************************************
⏩ Youtube - https://www.youtube.com/c/TechVizTheDataScienceGuy
⏩ Blog - https://prakhartechviz.blogspot.com
⏩ LinkedIn - https://linkedin.com/in/prakhar21
⏩ Medium - https://medium.com/@prakhar.mishra
⏩ GitHub - https://github.com/prakhar21
⏩ Twitter - https://twitter.com/rattller
*********************************************
Tools I use for making videos :)
⏩ iPad - https://tinyurl.com/y39p6pwc
⏩ Apple Pencil - https://tinyurl.com/y5rk8txn
⏩ GoodNotes - https://tinyurl.com/y627cfsa
#techviz #datascienceguy #research #qa #arxiv
Видео Training Question Answering Models From Synthetic Data (Research Paper Walkthrough) канала TechViz - The Data Science Guy
Показать
Комментарии отсутствуют
Информация о видео
29 мая 2021 г. 12:27:33
00:10:16
Другие видео канала
![Entity-level Factual Consistency of Abstractive Text Summarization (Research Paper Walkthrough)](https://i.ytimg.com/vi/P9wr8IBfDQs/default.jpg)
![Detecting Hallucinated Content in Conditional Neural Sequence Generation (NLP Paper Walkthrough)](https://i.ytimg.com/vi/fD2g9s1Isi4/default.jpg)
![REALM: Retrieval-Augmented Language Model Pre-Training (Research Paper Walkthrough)](https://i.ytimg.com/vi/F1naDPJpdY4/default.jpg)
![What makes a good life? Lessons from the longest study on happiness | Robert Waldinger](https://i.ytimg.com/vi/8KkKuTCFvzI/default.jpg)
![Auto Generating Python code by editing Spreadsheet using Mito](https://i.ytimg.com/vi/A0IxLGi5xC0/default.jpg)
![Thieves on Sesame Street! Model Extraction of BERT-based APIs (Research Paper Walkthrough)](https://i.ytimg.com/vi/ueC2a3hlBVs/default.jpg)
![Guide to Retraining Machine Learning Models (Blog Walkthrough)](https://i.ytimg.com/vi/-53lR0LajHI/default.jpg)
![Foundation Models | On the opportunities and risks of calling pre-trained models “Foundation Models”](https://i.ytimg.com/vi/4Cxz4rnnZ7Q/default.jpg)
![BERT4Rec: Sequential Recommendation with Bidirectional Encoder Representations from Transformer](https://i.ytimg.com/vi/4pYHEzwTa78/default.jpg)
![Generalization through Memorization: Nearest Neighbor Language Models (Research Paper Walkthrough)](https://i.ytimg.com/vi/nJaekQb6DwU/default.jpg)
![How To Write A Research Proposal For A Dissertation Or Thesis (With Examples)](https://i.ytimg.com/vi/eALzUfkQJRU/default.jpg)
![Inside a Google data center](https://i.ytimg.com/vi/XZmGGAbHqa0/default.jpg)
![How to escape education's death valley | Sir Ken Robinson](https://i.ytimg.com/vi/wX78iKhInsc/default.jpg)
![The beauty of data visualization | David McCandless](https://i.ytimg.com/vi/pLqjQ55tz-U/default.jpg)
![How To Read A Research Paper ?](https://i.ytimg.com/vi/pNIe5Uf_nAE/default.jpg)
![Life Lessons From 100-Year-Olds](https://i.ytimg.com/vi/9AThycGCakk/default.jpg)
![Deduplicating Training Data Makes Language Models Better (Research Paper Walkthrough)](https://i.ytimg.com/vi/V8VPSLxgDV4/default.jpg)
![Bisecting K-Means Algorithm (Clustering in Machine Learning) #Shorts](https://i.ytimg.com/vi/syGLao32R5Q/default.jpg)
![Questions No One Knows the Answers to (Full Version)](https://i.ytimg.com/vi/7SWvDHvWXok/default.jpg)
![Masked Language Models Vs Causal Language Models in NLP #Shorts](https://i.ytimg.com/vi/pjej9nlsV3E/default.jpg)