{"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/dnn-architecture-for-high-performance","title":"DNN Architecture for High Performance Prediction on Natural Videos Loses Submodule's Ability to Learn Discrete-World Dataset","arxiv_id":"1904.07969","date":"2019-04-16","proceeding":null,"authors":["Lana Sinapayen","Atsushi Noda"],"abstract":"Is cognition a collection of loosely connected functions tuned to different\ntasks, or can there be a general learning algorithm? If such an hypothetical\ngeneral algorithm did exist, tuned to our world, could it adapt seamlessly to a\nworld with different laws of nature? We consider the theory that predictive\ncoding is such a general rule, and falsify it for one specific neural\narchitecture known for high-performance predictions on natural videos and\nreplication of human visual illusions: PredNet. Our results show that PredNet's\nhigh performance generalizes without retraining on a completely different\nnatural video dataset. Yet PredNet cannot be trained to reach even mediocre\naccuracy on an artificial video dataset created with the rules of the Game of\nLife (GoL). We also find that a submodule of PredNet, a Convolutional Neural\nNetwork trained alone, reaches perfect accuracy on the GoL while being mediocre\nfor natural videos, showing that PredNet's architecture itself is responsible\nfor both the high performance on natural videos and the loss of performance on\nthe GoL. Just as humans cannot predict the dynamics of the GoL, our results\nsuggest that there might be a trade-off between high performance on sensory\ninputs with different sets of rules.","url_abs":"http://arxiv.org/abs/1904.07969v1","url_pdf":"http://arxiv.org/pdf/1904.07969v1.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":"dnn-architecture-for-high-performance","repo_url":"https://github.com/LanaSina/prednet_gol","is_official":1,"mentioned_in_paper":1,"mentioned_in_github":0,"framework":"tf","reach":null}],"tasks":[],"methods":[],"datasets_introduced":[],"methods_introduced":[],"results":[],"syntology":{"atlas_url":null,"mcp":null,"developers":"https://syntology.ai/developers"},"arxiv_metadata":null,"syntology_extracted_results":null}