Improving Transformers with Dynamically Composable Multi-Head Attention

• URL: http://arxiv.org/abs/2405.08553v2
• Date: Tue, 4 Jun 2024 14:49:36 GMT
• Title: Improving Transformers with Dynamically Composable Multi-Head Attention
• Authors: Da Xiao, Qingye Meng, Shengping Li, Xingyuan Yuan,
• Abstract summary: Multi-Head Attention (MHA) is a key component of Transformer. We propose Dynamically Composable Multi-Head Attention (DCMHA) as a parameter and computation efficient attention architecture. DCMHA can be used as a drop-in replacement of MHA in any transformer architecture to obtain the corresponding DCFormer.
• Score: 0.4999814847776097
• License: http://creativecommons.org/licenses/by/4.0/
• Abstract: Multi-Head Attention (MHA) is a key component of Transformer. In MHA, attention heads work independently, causing problems such as low-rank bottleneck of attention score matrices and head redundancy. We propose Dynamically Composable Multi-Head Attention (DCMHA), a parameter and computation efficient attention architecture that tackles the shortcomings of MHA and increases the expressive power of the model by dynamically composing attention heads. At the core of DCMHA is a $\it{Compose}$ function that transforms the attention score and weight matrices in an input-dependent way. DCMHA can be used as a drop-in replacement of MHA in any transformer architecture to obtain the corresponding DCFormer. DCFormer significantly outperforms Transformer on different architectures and model scales in language modeling, matching the performance of models with ~1.7x-2.0x compute. For example, DCPythia-6.9B outperforms open source Pythia-12B on both pretraining perplexity and downstream task evaluation. The code and models are available at https://github.com/Caiyun-AI/DCFormer.

Related papers

• Transformers to SSMs: Distilling Quadratic Knowledge to Subquadratic Models [92.36510016591782]
  We present a method that is able to distill a pretrained Transformer architecture into alternative architectures such as state space models (SSMs) Our method, called MOHAWK, is able to distill a Mamba-2 variant based on the Phi-1.5 architecture using only 3B tokens and a hybrid version (Hybrid Phi-Mamba) using 5B tokens. Despite using less than 1% of the training data typically used to train models from scratch, Phi-Mamba boasts substantially stronger performance compared to all past open-source non-Transformer models.
  arXiv  Detail & Related papers  (2024-08-19T17:48:11Z)
• HAFormer: Unleashing the Power of Hierarchy-Aware Features for Lightweight Semantic Segmentation [11.334990474402915]
  We introduce HAFormer, a model that combines the hierarchical features extraction ability of CNNs with the global dependency modeling capability of Transformers. HAFormer achieves high performance with minimal computational overhead and compact model size.
  arXiv  Detail & Related papers  (2024-07-10T07:53:24Z)
• Hyper-Transformer for Amodal Completion [82.4118011026855]
  Amodal object completion is a complex task that involves predicting the invisible parts of an object based on visible segments and background information. We introduce a novel framework named the Hyper-Transformer Amodal Network (H-TAN) This framework utilizes a hyper transformer equipped with a dynamic convolution head to directly learn shape priors and accurately predict amodal masks.
  arXiv  Detail & Related papers  (2024-05-30T11:11:54Z)
• DAT++: Spatially Dynamic Vision Transformer with Deformable Attention [87.41016963608067]
  We present Deformable Attention Transformer ( DAT++), a vision backbone efficient and effective for visual recognition. DAT++ achieves state-of-the-art results on various visual recognition benchmarks, with 85.9% ImageNet accuracy, 54.5 and 47.0 MS-COCO instance segmentation mAP, and 51.5 ADE20K semantic segmentation mIoU.
  arXiv  Detail & Related papers  (2023-09-04T08:26:47Z)
• ViTAEv2: Vision Transformer Advanced by Exploring Inductive Bias for Image Recognition and Beyond [76.35955924137986]
  We propose a Vision Transformer Advanced by Exploring intrinsic IB from convolutions, i.e., ViTAE. ViTAE has several spatial pyramid reduction modules to downsample and embed the input image into tokens with rich multi-scale context. We obtain the state-of-the-art classification performance, i.e., 88.5% Top-1 classification accuracy on ImageNet validation set and the best 91.2% Top-1 accuracy on ImageNet real validation set.
  arXiv  Detail & Related papers  (2022-02-21T10:40:05Z)
• nnFormer: Interleaved Transformer for Volumetric Segmentation [50.10441845967601]
  We introduce nnFormer, a powerful segmentation model with an interleaved architecture based on empirical combination of self-attention and convolution. nnFormer achieves tremendous improvements over previous transformer-based methods on two commonly used datasets Synapse and ACDC.
  arXiv  Detail & Related papers  (2021-09-07T17:08:24Z)
• ViTAE: Vision Transformer Advanced by Exploring Intrinsic Inductive Bias [76.16156833138038]
  We propose a novel Vision Transformer Advanced by Exploring intrinsic IB from convolutions, ie, ViTAE. ViTAE has several spatial pyramid reduction modules to downsample and embed the input image into tokens with rich multi-scale context. In each transformer layer, ViTAE has a convolution block in parallel to the multi-head self-attention module, whose features are fused and fed into the feed-forward network.
  arXiv  Detail & Related papers  (2021-06-07T05:31:06Z)
• Cascaded Head-colliding Attention [28.293881246428377]
  Transformers have advanced the field of natural language processing (NLP) on a variety of important tasks. We present cascaded head-colliding attention (CODA) which explicitly models the interactions between attention heads through a hierarchical variational distribution.
  arXiv  Detail & Related papers  (2021-05-31T10:06:42Z)
• Stochastic Attention Head Removal: A simple and effective method for improving Transformer Based ASR Models [40.991809705930955]
  We propose to randomly remove attention heads during training and keep all attention heads at test time, thus the final model is an ensemble of models with different architectures. Our method gives consistent performance gains over strong baselines on the Wall Street Journal, AISHELL, Switchboard and AMI datasets.
  arXiv  Detail & Related papers  (2020-11-08T15:41:03Z)

This list is automatically generated from the titles and abstracts of the papers in this site.

This site does not guarantee the quality of this site (including all information) and is not responsible for any consequences.