{"url":"/method/sru","slug":"sru","name":"SRU","full_name":"SRU","full_name_withheld":false,"description_markdown":"**SRU**, or **Simple Recurrent Unit**, is a recurrent neural unit with a light form of recurrence. SRU exhibits the same level of parallelism as [convolution](https://paperswithcode.com/method/convolution) and [feed-forward nets](https://paperswithcode.com/methods/category/feedforward-networks). This is achieved by balancing sequential dependence and independence: while the state computation of SRU is time-dependent, each state dimension is independent. This simplification enables CUDA-level optimizations that parallelize the computation across hidden dimensions and time steps, effectively using the full capacity of modern GPUs. \r\n\r\nSRU also replaces the use of convolutions (i.e., ngram filters), as in [QRNN](https://paperswithcode.com/method/qrnn) and KNN, with more recurrent connections. This retains modeling capacity, while using less computation (and hyper-parameters). Additionally, SRU improves the training of deep recurrent models by employing [highway connections](https://paperswithcode.com/method/highway-layer) and a parameter initialization scheme tailored for gradient propagation in deep architectures.\r\n\r\nA single layer of SRU involves the following computation:\r\n\r\n$$\r\n\\mathbf{f}\\_{t} =\\sigma\\left(\\mathbf{W}\\_{f} \\mathbf{x}\\_{t}+\\mathbf{v}\\_{f} \\odot \\mathbf{c}\\_{t-1}+\\mathbf{b}\\_{f}\\right) \r\n$$\r\n\r\n$$\r\n\\mathbf{c}\\_{t} =\\mathbf{f}\\_{t} \\odot \\mathbf{c}\\_{t-1}+\\left(1-\\mathbf{f}\\_{t}\\right) \\odot\\left(\\mathbf{W} \\mathbf{x}\\_{t}\\right) \\\\\r\n$$\r\n\r\n$$\r\n\\mathbf{r}\\_{t} =\\sigma\\left(\\mathbf{W}\\_{r} \\mathbf{x}\\_{t}+\\mathbf{v}\\_{r} \\odot \\mathbf{c}\\_{t-1}+\\mathbf{b}\\_{r}\\right) \\\\\r\n$$\r\n\r\n$$\r\n\\mathbf{h}\\_{t} =\\mathbf{r}\\_{t} \\odot \\mathbf{c}\\_{t}+\\left(1-\\mathbf{r}\\_{t}\\right) \\odot \\mathbf{x}\\_{t}\r\n$$\r\n\r\nwhere $\\mathbf{W}, \\mathbf{W}\\_{f}$ and $\\mathbf{W}\\_{r}$ are parameter matrices and $\\mathbf{v}\\_{f}, \\mathbf{v}\\_{r}, \\mathbf{b}\\_{f}$ and $\\mathbf{b}_{v}$ are parameter vectors to be learnt during training. The complete architecture decomposes to two sub-components: a light recurrence and a highway network,\r\n\r\nThe light recurrence component successively reads the input vectors $\\mathbf{x}\\_{t}$ and computes the sequence of states $\\mathbf{c}\\_{t}$ capturing sequential information. The computation resembles other recurrent networks such as [LSTM](https://paperswithcode.com/method/lstm), [GRU](https://paperswithcode.com/method/gru) and RAN. Specifically, a forget gate $\\mathbf{f}\\_{t}$ controls the information flow and the state vector $\\mathbf{c}\\_{t}$ is determined by adaptively averaging the previous state $\\mathbf{c}\\_{t-1}$ and the current observation $\\mathbf{W} \\mathbf{x}_{+}$according to $\\mathbf{f}\\_{t}$.","description_state":"present","introduced_year":null,"introduced_by":{"title":"Simple Recurrent Units for Highly Parallelizable Recurrence","paper":"/paper/simple-recurrent-units-for-highly","first_author":"Tao Lei","n_authors":5,"url_abs":null,"archive_paper_url":"https://paperswithcode.com/paper/simple-recurrent-units-for-highly"},"source":{"url":"http://arxiv.org/abs/1709.02755v5","title":"Simple Recurrent Units for Highly Parallelizable Recurrence","url_on_a_paper_host":true},"code_snippet_url":null,"code_snippet_url_on_a_code_host":false,"categories":[{"area":"Sequential","area_id":"sequential","collection":"Recurrent Neural Networks","url":"/methods/category/recurrent-neural-networks","pwc_aliases":[]}],"n_papers_tagged":16,"archive_num_papers":16,"papers_newest_first":[{"paper":"/paper/selective-stereo-adaptive-frequency","title":"Selective-Stereo: Adaptive Frequency Information Selection for Stereo Matching","date":"2024-03-01","arxiv_id":"2403.00486","n_code_links":1,"syntology":{"ran":9,"of":13,"unverified":4,"pointer_only":0}},{"paper":"/paper/on-the-effectiveness-of-unlearning-in-session","title":"On the Effectiveness of Unlearning in Session-Based Recommendation","date":"2023-12-22","arxiv_id":"2312.14447","n_code_links":1,"syntology":{"ran":1,"of":1,"unverified":0,"pointer_only":1}},{"paper":"/paper/neural-machine-translation-models-with","title":"Neural Machine Translation Models with Attention-Based Dropout Layer","date":"2023-05-01","arxiv_id":null,"n_code_links":1,"syntology":null},{"paper":"/paper/scconv-spatial-and-channel-reconstruction","title":"SCConv: Spatial and Channel Reconstruction Convolution for Feature Redundancy","date":"2023-01-01","arxiv_id":null,"n_code_links":1,"syntology":null},{"paper":null,"title":"A Robust Approach for the Decomposition of High-Energy-Consuming Industrial Loads with Deep Learning","date":"2022-03-11","arxiv_id":"2203.07075","n_code_links":0,"syntology":null},{"paper":null,"title":"MOHAQ: Multi-Objective Hardware-Aware Quantization of Recurrent Neural Networks","date":"2021-08-02","arxiv_id":"2108.01192","n_code_links":0,"syntology":null},{"paper":null,"title":"Intelligent Reflecting Surface Enhanced Indoor Robot Path Planning: A Radio Map based Approach","date":"2020-09-27","arxiv_id":"2009.12804","n_code_links":0,"syntology":null},{"paper":"/paper/asapp-asr-multistream-cnn-and-self-attentive","title":"ASAPP-ASR: Multistream CNN and Self-Attentive SRU for SOTA Speech Recognition","date":"2020-05-21","arxiv_id":"2005.10469","n_code_links":0,"syntology":null},{"paper":null,"title":"Multistream CNN for Robust Acoustic Modeling","date":"2020-05-21","arxiv_id":"2005.10470","n_code_links":0,"syntology":null},{"paper":null,"title":"Utterance-level Sequential Modeling For Deep Gaussian Process Based Speech Synthesis Using Simple Recurrent Unit","date":"2020-04-22","arxiv_id":"2004.10823","n_code_links":0,"syntology":null},{"paper":"/paper/wavecrn-an-efficient-convolutional-recurrent","title":"WaveCRN: An Efficient Convolutional Recurrent Neural Network for End-to-end Speech Enhancement","date":"2020-04-06","arxiv_id":"2004.04098","n_code_links":5,"syntology":null},{"paper":"/paper/economy-statistical-recurrent-units-for-1","title":"Economy Statistical Recurrent Units For Inferring Nonlinear Granger Causality","date":"2019-11-22","arxiv_id":"1911.09879","n_code_links":1,"syntology":{"ran":1,"of":1,"unverified":0,"pointer_only":0}},{"paper":"/paper/fastfusionnet-new-state-of-the-art-for","title":"FastFusionNet: New State-of-the-Art for DAWNBench SQuAD","date":"2019-02-28","arxiv_id":"1902.11291","n_code_links":2,"syntology":null},{"paper":null,"title":"Single Stream Parallelization of Recurrent Neural Networks for Low Power and Fast Inference","date":"2018-03-30","arxiv_id":"1803.11389","n_code_links":0,"syntology":null},{"paper":"/paper/training-rnns-as-fast-as-cnns","title":"Training RNNs as Fast as CNNs","date":"2018-01-01","arxiv_id":null,"n_code_links":2,"syntology":null},{"paper":"/paper/simple-recurrent-units-for-highly","title":"Simple Recurrent Units for Highly Parallelizable Recurrence","date":"2017-09-08","arxiv_id":"1709.02755","n_code_links":11,"syntology":{"ran":1,"of":1,"unverified":0,"pointer_only":0}}],"papers_shown":16,"tasks":[{"task":"/task/speech-recognition","name":"Speech Recognition","papers":4},{"task":"/task/speech-recognition-1","name":"speech-recognition","papers":4},{"task":"/task/translation","name":"Translation","papers":3},{"task":"/task/data-augmentation","name":"Data Augmentation","papers":2},{"task":"/task/classification","name":"General Classification","papers":2},{"task":"/task/language-modeling","name":"Language Modeling","papers":2},{"task":"/task/language-modelling","name":"Language Modelling","papers":2},{"task":"/task/machine-translation","name":"Machine Translation","papers":2},{"task":"/task/question-answering","name":"Question Answering","papers":2},{"task":null,"name":"CPU","papers":1},{"task":"/task/denoising","name":"Denoising","papers":1},{"task":"/task/diversity","name":"Diversity","papers":1},{"task":"/task/nmt","name":"NMT","papers":1},{"task":"/task/quantization","name":"Quantization","papers":1},{"task":"/task/reading-comprehension","name":"Reading Comprehension","papers":1},{"task":"/task/retrieval","name":"Retrieval","papers":1},{"task":"/task/robot-navigation","name":"Robot Navigation","papers":1},{"task":"/task/session-based-recommendations","name":"Session-Based Recommendations","papers":1},{"task":"/task/speech-denoising","name":"Speech Denoising","papers":1},{"task":"/task/speech-enhancement","name":"Speech Enhancement","papers":1}],"tasks_shown":20,"n_tasks":27,"usage_by_year":[{"year":"2017","papers":1},{"year":"2018","papers":2},{"year":"2019","papers":2},{"year":"2020","papers":5},{"year":"2021","papers":1},{"year":"2022","papers":1},{"year":"2023","papers":3},{"year":"2024","papers":1}],"row_source":"methods_table","archive":{"source":"pwc-archive (Hugging Face), CC BY-SA 4.0","snapshot":"2025-07-28","archive_url":"https://paperswithcode.com/method/sru"},"syntology_read_at":"2026-09-24T18:15:14+00:00"}