{"601897":{"#nid":"601897","#data":{"type":"news","title":"Combining Deep Learning with Prior Knowledge and Graph Structure to Create a New Type of Algorithm ","body":[{"value":"\u003Cp\u003ESeveral Georgia Tech faculty members and Ph.D. students recently took part in the thirty-first annual conference on \u003Ca href=\u0022https:\/\/nips.cc\/Conferences\/2017\u0022\u003ENeural Information Processing Systems\u003C\/a\u003E (NIPS2017), the premier conference for machine learning (ML) and computational neuroscience\u003Cstrong\u003E.\u003C\/strong\u003E\u003C\/p\u003E\r\n\r\n\u003Cp\u003ESchool of Computational Science and Engineering (CSE) Associate Professor and \u003Ca href=\u0022http:\/\/ml.gatech.edu\/\u0022\u003ECenter for Machine Learning\u003C\/a\u003E at Georgia Tech (ML@GT) Associate Director\u0026nbsp;\u003Ca href=\u0022https:\/\/www.cc.gatech.edu\/~lsong\/\u0022\u003E\u003Cstrong\u003ELe Song\u003C\/strong\u003E\u003C\/a\u003E and CSE Ph.D. students \u003Ca href=\u0022https:\/\/www.cc.gatech.edu\/~hdai8\/\u0022\u003E\u003Cstrong\u003EHanjun Dai\u003C\/strong\u003E\u003C\/a\u003E and \u003Ca href=\u0022https:\/\/sites.google.com\/site\/daibohr\/\u0022\u003E\u003Cstrong\u003EBo Dai\u003C\/strong\u003E\u003C\/a\u003E, along with \u003Cstrong\u003ESteven Skiena \u003C\/strong\u003Eand \u003Cstrong\u003EYingtao Tian\u003C\/strong\u003E of Stony Brook University, won the best paper award at the NIPS2017 workshop, \u003Ca href=\u0022http:\/\/www.quantum-machine.org\/workshops\/nips2017\/\u0022\u003EMachine Learning for Molecules and Materials\u003C\/a\u003E, for their paper titled \u003Ca href=\u0022http:\/\/www.quantum-machine.org\/workshops\/nips2017\/assets\/pdf\/sdvae_workshop_camera_ready.pdf\u0022\u003E\u003Cem\u003ESyntax-Directed Variational Autoencoder for Molecule Generation.\u003C\/em\u003E\u003C\/a\u003E\u003C\/p\u003E\r\n\r\n\u003Cp\u003EWhen discussing the award-winning paper, Song said, \u0026ldquo;Although this is a molecule and material workshop, we have an extended version of this study for computer programs. The extended version has been accepted to the premier deep learning conference, \u003Ca href=\u0022https:\/\/iclr.cc\/\u0022\u003EICLR 2018\u003C\/a\u003E.\u0026rdquo;\u003C\/p\u003E\r\n\r\n\u003Cp\u003E\u0026ldquo;Much in the same that molecules can be valid or invalid for testing, programs too can be valid and invalid. This is why syntax and semantics are valid or not \u0026ndash; and the same technique that is generated for the molecules can be applied to both.\u0026rdquo;\u003C\/p\u003E\r\n\r\n\u003Cp\u003E\u0026ldquo;We build upon deep learning models, something that has no prior knowledge and does not use logic \u0026ndash; it simply has an input and output. We then augment deep learning model with graph structure, prior knowledge, and the ability to reason with logic,\u0026ldquo; said Song.\u003C\/p\u003E\r\n\r\n\u003Cp\u003EThis approach is proven to be significantly better when compared to other current methods for solving the issue of capturing representations for discrete structures with formal grammars and semantics. In other words, this approach uses a deep learning model with the combined methods of prior knowledge with graph structure to create a more efficient algorithm that requires less training data \u0026ndash; the likes of which have seldom been combined effectively before.\u003C\/p\u003E\r\n\r\n\u003Cp\u003ESong became the associate director for ML@GT in 2016 and an associate professor for CSE in 2017. His primary research focuses on embedding, dynamic processes over networks, large scale machine learning, and interdisciplinary problems. He also co-authored and submitted \u0026ndash; alongside several Georgia Tech faculty members and students \u0026ndash; five papers to the main conference this year:\u0026nbsp;\u003C\/p\u003E\r\n\r\n\u003Cul\u003E\r\n\t\u003Cli\u003E\u003Ca href=\u0022http:\/\/papers.nips.cc\/paper\/6917-wasserstein-learning-of-deep-generative-point-process-models\u0022\u003E\u003Cem\u003EWasserstein Learning of Deep Generative Point Process Models\u003C\/em\u003E\u003C\/a\u003E\u003C\/li\u003E\r\n\t\u003Cli\u003E\u003Ca href=\u0022http:\/\/papers.nips.cc\/paper\/6984-deep-hyperspherical-learning\u0022\u003E\u003Cem\u003EDeep Hyperspherical Learning\u003C\/em\u003E\u003C\/a\u003E\u003C\/li\u003E\r\n\t\u003Cli\u003E\u003Ca href=\u0022http:\/\/papers.nips.cc\/paper\/7135-on-the-complexity-of-learning-neural-networks\u0022\u003E\u003Cem\u003EOn the Complexity of Learning Neural Networks\u0026nbsp;\u003C\/em\u003E\u0026nbsp;\u003C\/a\u003E\u003C\/li\u003E\r\n\t\u003Cli\u003E\u003Ca href=\u0022http:\/\/papers.nips.cc\/paper\/6762-predicting-user-activity-level-in-point-processes-with-mass-transport-equation\u0022\u003E\u003Cem\u003EPredicting User Activity Level in Point Processes With Mass Transport Equation\u003C\/em\u003E\u0026nbsp;\u0026nbsp;\u003C\/a\u003E\u003C\/li\u003E\r\n\t\u003Cli\u003E\u003Ca href=\u0022http:\/\/papers.nips.cc\/paper\/7214-learning-combinatorial-optimization-algorithms-over-graphs\u0022\u003E\u003Cem\u003ELearning Combinatorial Optimization Algorithms over Graphs\u003C\/em\u003E\u0026nbsp;\u003C\/a\u003E\u003C\/li\u003E\r\n\u003C\/ul\u003E\r\n\r\n\u003Cp\u003E\u003Cem\u003ELearning Combinatorial Optimization Algorithms over Graphs\u003C\/em\u003E, creates a framework for using deep learning to develop learning optimization algorithms. These frameworks are already being used by global industry titans. Tencent, China\u0026rsquo;s technology and investment holding conglomerate, which owns three of the world\u0026rsquo;s five biggest social networks, is using this algorithm for advertising placement to match ads to their most effective target audience. Similarly, Alibaba, China\u0026rsquo;s biggest online commerce company, is also using this algorithm for fraud detection.\u003C\/p\u003E\r\n\r\n\u003Cp\u003EHowever, this framework has uses ranging beyond that of fraud detection or social media marketing. Song said, \u0026ldquo;Many real-world problems are constrained on a graph, such as the traveling salesman problem appearing in logistics, and the assignment of computing tasks to a network of computing nodes.\u0026rdquo;\u003C\/p\u003E\r\n\r\n\u003Cp\u003E\u0026ldquo;We want to learn an intelligent algorithm which can become smarter as it solves more problems. [This applies to a wide variety of applications] For instance, we have a new algorithm for improving robot planning which can learn and become smarter.\u0026rdquo;\u003C\/p\u003E\r\n","summary":null,"format":"limited_html"}],"field_subtitle":"","field_summary":"","field_summary_sentence":[{"value":"CSE Associate Professor Le Song and PhD Students Create New Algorithm that Requires Less Training Data. "}],"uid":"34540","created_gmt":"2018-02-05 19:57:34","changed_gmt":"2018-02-09 15:37:37","author":"Kristen Perez","boilerplate_text":"","field_publication":"","field_article_url":"","dateline":{"date":"2018-02-05T00:00:00-05:00","iso_date":"2018-02-05T00:00:00-05:00","tz":"America\/New_York"},"extras":[],"hg_media":{"583584":{"id":"583584","type":"image","title":"Le Song_Klaus","body":null,"created":"1478540785","gmt_created":"2016-11-07 17:46:25","changed":"1478540785","gmt_changed":"2016-11-07 17:46:25","alt":"","file":{"fid":"222481","name":"Le Song.jpg","image_path":"\/sites\/default\/files\/images\/Le%20Song.jpg","image_full_path":"http:\/\/hg.gatech.edu\/\/sites\/default\/files\/images\/Le%20Song.jpg","mime":"image\/jpeg","size":386090,"path_740":"http:\/\/hg.gatech.edu\/sites\/default\/files\/styles\/740xx_scale\/public\/images\/Le%20Song.jpg?itok=2VZUb659"}}},"media_ids":["583584"],"groups":[{"id":"47223","name":"College of Computing"},{"id":"50877","name":"School of Computational Science and Engineering"}],"categories":[{"id":"134","name":"Student and Faculty"},{"id":"8862","name":"Student Research"},{"id":"135","name":"Research"}],"keywords":[{"id":"173446","name":"CSE. School of Computational Science and Engineering"},{"id":"127171","name":"Le Song"},{"id":"177019","name":"prior knowledge"},{"id":"3167","name":"algorithm"},{"id":"177020","name":"graph structure"}],"core_research_areas":[],"news_room_topics":[],"event_categories":[],"invited_audience":[],"affiliations":[],"classification":[],"areas_of_expertise":[],"news_and_recent_appearances":[],"phone":[],"contact":[],"email":["kristen.perez@cc.gatech.edu"],"slides":[],"orientation":[],"userdata":""}}}