{"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/deep-convolutional-decision-jungle-for-image","title":"Deep Convolutional Decision Jungle for Image Classification","arxiv_id":"1706.02003","date":"2017-06-06","proceeding":null,"authors":["Seungryul Baek","Kwang In Kim","Tae-Kyun Kim"],"abstract":"We propose a novel method called deep convolutional decision jungle (CDJ) and\nits learning algorithm for image classification. The CDJ maintains the\nstructure of standard convolutional neural networks (CNNs), i.e. multiple\nlayers of multiple response maps fully connected. Each response map-or node-in\nboth the convolutional and fully-connected layers selectively respond to class\nlabels s.t. each data sample travels via a specific soft route of those\nactivated nodes. The proposed method CDJ automatically learns features, whereas\ndecision forests and jungles require pre-defined feature sets. Compared to\nCNNs, the method embeds the benefits of using data-dependent discriminative\nfunctions, which better handles multi-modal/heterogeneous data; further,the\nmethod offers more diverse sparse network responses, which in turn can be used\nfor cost-effective learning/classification. The network is learnt by combining\nconventional softmax and proposed entropy losses in each layer. The entropy\nloss,as used in decision tree growing, measures the purity of data activation\naccording to the class label distribution. The back-propagation rule for the\nproposed loss function is derived from stochastic gradient descent (SGD)\noptimization of CNNs. We show that our proposed method outperforms\nstate-of-the-art methods on three public image classification benchmarks and\none face verification dataset. We also demonstrate the use of auxiliary data\nlabels, when available, which helps our method to learn more discriminative\nrouting and representations and leads to improved classification.","url_abs":"http://arxiv.org/abs/1706.02003v2","url_pdf":"http://arxiv.org/pdf/1706.02003v2.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":[],"tasks":[{"task_slug":"classification-1","task_name":"Classification"},{"task_slug":"face-verification","task_name":"Face Verification"},{"task_slug":"classification","task_name":"General Classification"},{"task_slug":"image-classification","task_name":"Image Classification"},{"task_slug":"image-classification","task_name":"image-classification"}],"methods":[{"method_slug":"softmax","method_name":"Softmax"}],"datasets_introduced":[],"methods_introduced":[],"results":[{"leaderboard":"/sota/image-classification-on-cifar-100","task":"Image Classification","dataset":"CIFAR-100","model":"NiN+Superclass+CDJ","rank_in_archive_order":179,"of":211,"metrics":{"Percentage correct":"69"},"uses_additional_data":true}],"syntology":{"atlas_url":"https://app.syntology.ai/?focus=1706.02003","mcp":null,"developers":"https://syntology.ai/developers"},"arxiv_metadata":null,"syntology_extracted_results":null}