{"url":"/method/forward-gradient","slug":"forward-gradient","name":"Forward gradient","full_name":"Forward gradient","full_name_withheld":false,"description_markdown":"Forward gradients are unbiased estimators of the gradient $\\nabla f(\\theta)$ for a function $f: \\mathbb{R}^n \\rightarrow \\mathbb{R}$, given by $g(\\theta) = \\langle \\nabla f(\\theta) , v \\rangle v$. \r\n\r\nHere $v = (v_1, \\ldots, v_n)$ is a random vector, which must satisfy the following conditions in order for $g(\\theta)$ to be an unbiased estimator of $\\nabla f(\\theta)$\r\n\r\n* $v_i \\perp v_j$ for all $i \\neq j$\r\n* $\\mathbb{E}[v_i] = 0$ for all $i$\r\n* $\\mathbb{V}[v_i] = 1$ for all $i$\r\n\r\nForward gradients can be computed with a single jvp (Jacobian Vector Product), which enables the use of the forward mode of autodifferentiation instead of the usual reverse mode, which has worse computational characteristics.","description_state":"present","introduced_year":null,"introduced_by":{"title":"Gradients without Backpropagation","paper":"/paper/gradients-without-backpropagation","first_author":"Atılım Güneş Baydin","n_authors":5,"url_abs":null,"archive_paper_url":"https://paperswithcode.com/paper/gradients-without-backpropagation"},"source":{"url":"https://arxiv.org/abs/2202.08587v1","title":"Gradients without Backpropagation","url_on_a_paper_host":true},"code_snippet_url":null,"code_snippet_url_on_a_code_host":false,"categories":[{"area":"General","area_id":"general","collection":"Stochastic Optimization","url":"/methods/category/stochastic-optimization","pwc_aliases":[]}],"n_papers_tagged":12,"archive_num_papers":12,"papers_newest_first":[{"paper":"/paper/a-scalable-hybrid-training-approach-for","title":"A Scalable Hybrid Training Approach for Recurrent Spiking Neural Networks","date":"2025-06-17","arxiv_id":"2506.14464","n_code_links":1,"syntology":null},{"paper":null,"title":"Flexible and Efficient Surrogate Gradient Modeling with Forward Gradient Injection","date":"2024-05-31","arxiv_id":"2406.00177","n_code_links":0,"syntology":null},{"paper":null,"title":"Projected Forward Gradient-Guided Frank-Wolfe Algorithm via Variance Reduction","date":"2024-03-19","arxiv_id":"2403.12511","n_code_links":0,"syntology":null},{"paper":"/paper/revisiting-zeroth-order-optimization-for","title":"Revisiting Zeroth-Order Optimization for Memory-Efficient LLM Fine-Tuning: A Benchmark","date":"2024-02-18","arxiv_id":"2402.11592","n_code_links":1,"syntology":{"ran":8,"of":13,"unverified":5,"pointer_only":13}},{"paper":"/paper/odice-revealing-the-mystery-of-distribution","title":"ODICE: Revealing the Mystery of Distribution Correction Estimation via Orthogonal-gradient Update","date":"2024-02-01","arxiv_id":"2402.00348","n_code_links":1,"syntology":{"ran":8,"of":8,"unverified":0,"pointer_only":8}},{"paper":null,"title":"Convergence guarantees for forward gradient descent in the linear regression model","date":"2023-09-26","arxiv_id":"2309.15001","n_code_links":0,"syntology":null},{"paper":null,"title":"Accelerated On-Device Forward Neural Network Training with Module-Wise Descending Asynchronism","date":"2023-09-21","arxiv_id":null,"n_code_links":0,"syntology":null},{"paper":"/paper/can-forward-gradient-match-backpropagation","title":"Can Forward Gradient Match Backpropagation?","date":"2023-06-12","arxiv_id":"2306.06968","n_code_links":1,"syntology":null},{"paper":"/paper/directional-direct-feedback-alignment","title":"Low-Variance Forward Gradients using Direct Feedback Alignment and Momentum","date":"2022-12-14","arxiv_id":"2212.07282","n_code_links":1,"syntology":null},{"paper":"/paper/scaling-forward-gradient-with-local-losses","title":"Scaling Forward Gradient With Local Losses","date":"2022-10-07","arxiv_id":"2210.03310","n_code_links":1,"syntology":null},{"paper":"/paper/optimization-without-backpropagation","title":"Optimization without Backpropagation","date":"2022-09-13","arxiv_id":"2209.06302","n_code_links":1,"syntology":{"ran":0,"of":8,"unverified":8,"pointer_only":0}},{"paper":"/paper/gradients-without-backpropagation","title":"Gradients without Backpropagation","date":"2022-02-17","arxiv_id":"2202.08587","n_code_links":3,"syntology":{"ran":2,"of":4,"unverified":2,"pointer_only":0}}],"papers_shown":12,"tasks":[{"task":"/task/benchmarking","name":"Benchmarking","papers":1},{"task":"/task/imitation-learning","name":"Imitation Learning","papers":1},{"task":"/task/memorization","name":"Memorization","papers":1},{"task":"/task/model-optimization","name":"Model Optimization","papers":1},{"task":"/task/offline-rl","name":"Offline RL","papers":1},{"task":"/task/reinforcement-learning-1","name":"Reinforcement Learning (RL)","papers":1},{"task":"/task/regression-1","name":"regression","papers":1}],"tasks_shown":7,"n_tasks":7,"usage_by_year":[{"year":"2022","papers":4},{"year":"2023","papers":3},{"year":"2024","papers":4},{"year":"2025","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/forward-gradient"},"syntology_read_at":"2026-09-24T18:15:14+00:00"}