Papers by Joshua Ainslie
ReadTwice: Reading Very Large Documents with Memories (2021.naacl-main)
Copied to clipboard
| Challenge: | Existing approaches to model long-range dependencies in text are limited to 512 tokens . however, the amount of compute in attention depends quadratically on the number of tokens in an input text passage. |
| Approach: | They propose a technique that summarises text into a memory table to be used in a second read of the text. |
| Outcome: | The proposed method outperforms models of comparable size on several question answering datasets and sets a new state of the art on the NarrativeQA task, with questions about entire books. |
Generate-and-Retrieve: Use Your Predictions to Improve Retrieval for Semantic Parsing (2022.coling-1)
Copied to clipboard
Yury Zemlyanskiy, Michiel de Jong, Joshua Ainslie, Panupong Pasupat, Peter Shaw, Linlu Qiu, Sumit Sanghai, Fei Sha
| Challenge: | Existing retrieval techniques for semantic parsing use similarity of query and exemplar inputs . Existing work suggests that appending training samples to training samples improves performance . |
| Approach: | They propose a retrieval procedure that retrieves exemplars for which outputs are similar . existing retrieval techniques are based on similarity of query and exemplar inputs . |
| Outcome: | Existing retrieval techniques rely on similarity of query and exemplar inputs . they retrieve exemplars with similar outputs and generate a final prediction . |
Sparse Mixers: Combining MoE and Mixing to build a more efficient BERT (2022.findings-emnlp)
Copied to clipboard
| Challenge: | Sparse Mixer encoder model outperforms BERT on GLUE and SuperGLUE, trains 65% faster and runs inference 61% faster. |
| Approach: | They combine the capacity of sparsely gated Mixture-of-Experts (MoE) with the speed and stability of linear, mixing transformations to design the Sparse Mixer encoder model. |
| Outcome: | The proposed model outperforms BERT on GLUE and SuperGLUE but trains and runs twice as fast. |
FormNet: Structural Encoding beyond Sequential Modeling in Form Document Information Extraction (2022.acl-long)
Copied to clipboard
Chen-Yu Lee, Chun-Liang Li, Timothy Dozat, Vincent Perot, Guolong Su, Nan Hua, Joshua Ainslie, Renshen Wang, Yasuhisa Fujii, Tomas Pfister
| Challenge: | Form-like document understanding is a surging research topic due to its practical applications . form documents have unique challenges stemming from their structural characteristics . |
| Approach: | They propose a structure-aware sequence model that leverages spatial relationships between tokens in a form for more precise attention score calculation. |
| Outcome: | The proposed model outperforms existing methods with a more compact model size and less pre-training data. |
FNet: Mixing Tokens with Fourier Transforms (2022.naacl-main)
Copied to clipboard
| Challenge: | Using simple linear transformations, Transformer encoders can be sped up with limited accuracy costs by replacing the self-attention sublayers with simple linear mixing mechanisms. |
| Approach: | They propose to replace the self-attention sublayer with a linear transformation that "mixes" input tokens. |
| Outcome: | The proposed model outperforms the “efficient Transformers” on the GLUE benchmark at longer input lengths and on smaller models with a light memory footprint. |
FiDO: Fusion-in-Decoder optimized for stronger performance and faster inference (2023.findings-acl)
Copied to clipboard
Michiel de Jong, Yury Zemlyanskiy, Joshua Ainslie, Nicholas FitzGerald, Sumit Sanghai, Fei Sha, William Cohen
| Challenge: | Fusion-in-Decoder (FiD) is a powerful retrieval-augmented language model . however, the architecture used for FiD was not designed for retrieval augmented models . |
| Approach: | They propose to make FiD a modified retrieval-augmented language model with a large decoder and memory bandwidth constraints to alleviate memory bandwidth limitations. |
| Outcome: | The proposed architecture outperforms existing models on knowledge-intensive tasks even on large models on many knowledge-based tasks. |
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. |
Improving Compositional Generalization in Classification Tasks via Structure Annotations (2021.acl-short)
Copied to clipboard
| Challenge: | Compositional generalization is the ability to generalize systematically to a new data distribution by combining known components. |
| Approach: | They propose to convert a natural language sequence-to-sequence dataset into a classification dataset that requires compositional generalization. |
| Outcome: | The proposed model can generalize compositionally by providing hints on the structure of the input. |
GQA: Training Generalized Multi-Query Transformer Models from Multi-Head Checkpoints (2023.emnlp-main)
Copied to clipboard
| Challenge: | Multi-query attention (MQA) can lead to quality degradation and training instability . it may not be feasible to train separate models optimized for quality and inference. |
| Approach: | They propose a recipe for uptraining existing multi-head language model checkpoints into models with MQA using 5% of original training compute. |
| Outcome: | The proposed model achieves comparable quality to multi-head attention with comparable speed. |
ETC: Encoding Long and Structured Inputs in Transformers (2020.emnlp-main)
Copied to clipboard
Joshua Ainslie, Santiago Ontanon, Chris Alberti, Vaclav Cvicek, Zachary Fisher, Philip Pham, Anirudh Ravula, Sumit Sanghai, Qifan Wang, Li Yang
| Challenge: | Existing models for natural language processing (NLP) have been challenging to scale attention to longer inputs. |
| Approach: | They propose an extended Transformer construction architecture that scales attention to longer inputs by combining global-local attention with relative position encodings and a "Contrastive Predictive Coding" objective. |
| Outcome: | The proposed architecture scales attention to longer inputs and encodes structured inputs. |
A Suite of Generative Tasks for Multi-Level Multimodal Webpage Understanding (2023.emnlp-main)
Copied to clipboard
Andrea Burns, Krishna Srinivasan, Joshua Ainslie, Geoff Brown, Bryan Plummer, Kate Saenko, Jianmo Ni, Mandy Guo
| Challenge: | Existing datasets for webpages contain only fragments of webpages . generative tasks like page description generation and section summarization are often left unstudied . |
| Approach: | They introduce a Wikipedia Webpage suite that contains 2M pages with all associated image, text, and structure data. |
| Outcome: | The proposed approach performs better than full attention with lower computational complexity. |
Making Transformers Solve Compositional Tasks (2022.acl-long)
Copied to clipboard
| Challenge: | Several studies have reported the inability of Transformer models to generalize compositionally . a key aspect of natural language is the ability to learn basic primitives . |
| Approach: | They propose to use Transformers to generalize compositionally in a large range of tasks . they find that Transformers generalize significantly better than previous models . |
| Outcome: | The proposed models generalize compositionally significantly better than previous models . a set of 12 datasets shows that the proposed models can be improved . |
RealFormer: Transformer Likes Residual Attention (2021.findings-acl)
Copied to clipboard
| Challenge: | Existing techniques to create Residual Attention Layer Transformer networks outperform the canonical Transformer on a wide spectrum of tasks. |
| Approach: | They propose a technique to create Residual Attention Layer Transformer networks that outperform the canonical Transformer on a wide spectrum of tasks. |
| Outcome: | The proposed technique outperforms the canonical Transformer on a wide spectrum of tasks including Masked Language Modeling, GLUE, SQUAD, Neural Machine Translation, WikiHop, HotpotQA, Natural Questions, and OpenKP. |
FormNetV2: Multimodal Graph Contrastive Learning for Form Document Information Extraction (2023.acl-long)
Copied to clipboard
Chen-Yu Lee, Chun-Liang Li, Hao Zhang, Timothy Dozat, Vincent Perot, Guolong Su, Xiang Zhang, Kihyuk Sohn, Nikolay Glushnev, Renshen Wang, Joshua Ainslie, Shangbang Long, Siyang Qin, Yasuhisa Fujii, Nan Hua, Tomas Pfister
| Challenge: | Existing approaches that extend the mask language modeling to other modalities require careful multi-task tuning, complex reconstruction target designs, or additional pre-training data. |
| Approach: | They propose a centralized multimodal graph contrastive learning strategy to unify self-supervised pre-training for all modalities in one loss. |
| Outcome: | The proposed model achieves state-of-the-art performance on FUNSD, CORD, SROIE and Payment benchmarks with a more compact model size. |
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. |
MEMORY-VQ: Compression for Tractable Internet-Scale Memory (2024.naacl-short)
Copied to clipboard
Yury Zemlyanskiy, Michiel de Jong, Luke Vilnis, Santiago Ontanon, William Cohen, Sumit Sanghai, Joshua Ainslie
| Challenge: | Memory-based methods like LUMEN pre-compute token representations for retrieved passages to speed up inference. |
| Approach: | They propose a method to reduce storage requirements of memory-augmented models . they use a vector quantization variational autoencoder to compress token representations . |
| Outcome: | The proposed method achieves 16x compression rate with comparable performance on KILT benchmark. |
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. |