{"about":{"site":"https://codewithpapers.app","non_affiliation":"Code with Papers and Syntology are not affiliated with, endorsed by, or sponsored by Papers with Code, Meta, or the pwc-archive mirror.","licence":"CC BY-SA 4.0","licence_url":"https://creativecommons.org/licenses/by-sa/4.0/legalcode","attribution":"https://codewithpapers.app/attribution","modified":"archive material modified by Syntology; see the attribution page"},"url":"/paper/learning-attractor-dynamics-for-generative","title":"Learning Attractor Dynamics for Generative Memory","arxiv_id":"1811.09556","date":"2018-11-23","proceeding":"NeurIPS 2018 12","authors":["Yan Wu","Greg Wayne","Karol Gregor","Timothy Lillicrap"],"abstract":"A central challenge faced by memory systems is the robust retrieval of a\nstored pattern in the presence of interference due to other stored patterns and\nnoise. A theoretically well-founded solution to robust retrieval is given by\nattractor dynamics, which iteratively clean up patterns during recall. However,\nincorporating attractor dynamics into modern deep learning systems poses\ndifficulties: attractor basins are characterised by vanishing gradients, which\nare known to make training neural networks difficult. In this work, we avoid\nthe vanishing gradient problem by training a generative distributed memory\nwithout simulating the attractor dynamics. Based on the idea of memory writing\nas inference, as proposed in the Kanerva Machine, we show that a\nlikelihood-based Lyapunov function emerges from maximising the variational\nlower-bound of a generative memory. Experiments shows it converges to correct\npatterns upon iterative retrieval and achieves competitive performance as both\na memory model and a generative model.","url_abs":"http://arxiv.org/abs/1811.09556v1","url_pdf":"http://arxiv.org/pdf/1811.09556v1.pdf","source":{"archive":"pwc-archive (Hugging Face), CC BY-SA 4.0","snapshot":"2025-07-28","licence_url":"https://creativecommons.org/licenses/by-sa/4.0/legalcode","row_kind":"abstracts"},"code_links":[{"paper_slug":"learning-attractor-dynamics-for-generative","repo_url":"https://github.com/deepmind/dynamic-kanerva-machines","is_official":1,"mentioned_in_paper":1,"mentioned_in_github":0,"framework":"tf","reach":{"status":"ok","spdx":"Apache-2.0"}}],"tasks":[{"task_slug":"retrieval","task_name":"Retrieval"}],"methods":[],"datasets_introduced":[],"methods_introduced":[],"results":[],"syntology":{"syntology_url":"https://syntology.ai/paper/1811.09556","atlas_url":"https://app.syntology.ai/?focus=1811.09556","mcp":null,"developers":"https://syntology.ai/developers"},"arxiv_metadata":null,"syntology_extracted_results":null}