Papers by David Uthus
Memory Augmented Language Models through Mixture of Word Experts (2024.naacl-long)
Copied to clipboard
| Challenge: | Increasing the parameter count of language models has been a primary driver of improved model quality, but increasing the model size also increases the cost of training and serving the model. |
| Approach: | They propose to decouple learning capacity and FLOPs by using a mixture-of-experts approach with large knowledge-rich vocabulary based routing functions. |
| Outcome: | The proposed model outperforms the T5 family of models with similar number of FLOPs on knowledge intensive tasks and similar performance to memory augmented approaches. |
Augmenting Poetry Composition with Verse by Verse (2022.naacl-industry)
Copied to clipboard
| Challenge: | a new approach to poetry generation has been developed that allows an AI to generate a full poem by itself, thus writing in a closed system. |
| Approach: | They describe an AI poet that offers suggestions while a user is composing a poem . they use a generative model and a dual encoder model to offer the suggestions . |
| Outcome: | The proposed system can offer suggestions generated lines of verse while a user is composing a poem. |
RISE: Leveraging Retrieval Techniques for Summarization Evaluation (2023.findings-acl)
Copied to clipboard
| Challenge: | Summarization evaluation approaches have relied on ROUGE for summarization, but they fall short of human evaluations. |
| Approach: | They propose a new approach to evaluate summaries by leveraging retrieval techniques . they use a dual-encoder retrieval setup to train a retrieval task . |
| Outcome: | The proposed method outperforms existing methods on two document summarization benchmarks and a long document summmarization test. |
LongT5: Efficient Text-To-Text Transformer for Long Sequences (2022.findings-naacl)
Copied to clipboard
| Challenge: | Recent work has shown that increasing the input length or increasing model size can improve the performance of Transformer-based neural models. |
| Approach: | They propose a model that integrates attention ideas from long-input transformers and adopts pre-training strategies from summarization pre-train into the scalable T5 architecture. |
| Outcome: | The proposed model outperforms the original T5 models on several summarization and question answering tasks and achieves state-of-the-art results. |
TextSETTR: Few-Shot Text Style Extraction and Tunable Targeted Restyling (2021.acl-long)
Copied to clipboard
| Challenge: | Existing methods for text style transfer require style-labeled training data, but use only labeled data at inference time. |
| Approach: | They propose a method that uses readily-available unlabeled text to train style transfer . they use a style vector to condition a decoder to perform style transfer using unlabelled text . |
| Outcome: | The proposed method is competitive on sentiment transfer, even compared to models trained fully on labeled data. |
mLongT5: A Multilingual and Efficient Text-To-Text Transformer for Longer Sequences (2023.findings-emnlp)
Copied to clipboard
| Challenge: | a new text-to-text transformer is suitable for multilingual inputs . many of the current models are English-only, making them inapplicable to other languages. |
| Approach: | They propose to extend a multilingual text-to-text transformer to handle long inputs . they use the mC4 dataset to pretrain the model to handle multilingual data . |
| Outcome: | The proposed model performs well on multilingual summarization and question-answering tasks. |
CoLT5: Faster Long-Range Transformers with Conditional Computation (2023.emnlp-main)
Copied to clipboard
Joshua Ainslie, Tao Lei, Michiel de Jong, Santiago Ontanon, Siddhartha Brahma, Yury Zemlyanskiy, David Uthus, Mandy Guo, James Lee-Thorp, Yi Tay, Yun-Hsuan Sung, Sumit Sanghai
| Challenge: | Many natural language processing tasks require long inputs, but processing long documents with a Transformer model is expensive due to quadratic attention complexity and applying feedforward and attention projection layers to every input token. |
| Approach: | They propose a long-input Transformer model that builds on the intuition that some tokens are more important than others and uses conditional computation to devote more computation to important tokens. |
| Outcome: | The proposed model achieves stronger performance than LongT5 with faster training and inference, achieving SOTA on the long-input SCROLLS benchmark. |
Reconsidering Sentence-Level Sign Language Translation (2024.emnlp-main)
Copied to clipboard
| Challenge: | Historically, sign language machine translation is framed as a sentence-level task . however, there are known intersentential dependencies that are impossible to resolve in isolation. |
| Approach: | They propose a human baseline for sign language translation that substitutes a person into the machine learning task framing instead of providing the entire document as context. |
| Outcome: | The proposed human baseline for sign language translation shows that deaf signers can only understand key parts of the clip in light of additional discourse-level context. |