{"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/deeparchitect-automatically-designing-and","title":"DeepArchitect: Automatically Designing and Training Deep Architectures","arxiv_id":"1704.08792","date":"2017-04-28","proceeding":"ICLR 2018 1","authors":["Renato Negrinho","Geoff Gordon"],"abstract":"In deep learning, performance is strongly affected by the choice of\narchitecture and hyperparameters. While there has been extensive work on\nautomatic hyperparameter optimization for simple spaces, complex spaces such as\nthe space of deep architectures remain largely unexplored. As a result, the\nchoice of architecture is done manually by the human expert through a slow\ntrial and error process guided mainly by intuition. In this paper we describe a\nframework for automatically designing and training deep models. We propose an\nextensible and modular language that allows the human expert to compactly\nrepresent complex search spaces over architectures and their hyperparameters.\nThe resulting search spaces are tree-structured and therefore easy to traverse.\nModels can be automatically compiled to computational graphs once values for\nall hyperparameters have been chosen. We can leverage the structure of the\nsearch space to introduce different model search algorithms, such as random\nsearch, Monte Carlo tree search (MCTS), and sequential model-based optimization\n(SMBO). We present experiments comparing the different algorithms on CIFAR-10\nand show that MCTS and SMBO outperform random search. In addition, these\nexperiments show that our framework can be used effectively for model\ndiscovery, as it is possible to describe expressive search spaces and discover\ncompetitive models without much effort from the human expert. Code for our\nframework and experiments has been made publicly available.","url_abs":"http://arxiv.org/abs/1704.08792v1","url_pdf":"http://arxiv.org/pdf/1704.08792v1.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":"deeparchitect-automatically-designing-and","repo_url":"https://github.com/negrinho/deep_architect_legacy","is_official":0,"mentioned_in_paper":0,"mentioned_in_github":1,"framework":"tf","reach":{"status":"ok","spdx":"MIT"}}],"tasks":[{"task_slug":"hyperparameter-optimization","task_name":"Hyperparameter Optimization"},{"task_slug":"model-discovery","task_name":"Model Discovery"}],"methods":[],"datasets_introduced":[],"methods_introduced":[],"results":[],"syntology":{"atlas_url":"https://app.syntology.ai/?focus=1704.08792","mcp":null,"developers":"https://syntology.ai/developers"},"arxiv_metadata":null,"syntology_extracted_results":null}