{"corpus_id":67752026,"paper_sha":"7e71eedb078181873a56f2adcfef9dddaeb95602","doi":null,"arxiv_id":"1902.07153","pmid":null,"pmcid":null,"mag_id":2964124573,"dblp_id":"conf/icml/WuSZFYW19","acl_id":null,"title":"Simplifying Graph Convolutional Networks","year":2019,"publication_date":"2019-02-19","venue":"International Conference on Machine Learning","journal":{"name":null,"pages":"6861-6871","volume":null},"journal_issn":null,"journal_title":null,"publication_types":["JournalArticle","Conference"],"pubmed_pub_types":null,"s2_fields_of_study":["Mathematics","Computer Science"],"reference_count":63,"citation_count":3718,"influential_citation_count":681,"is_open_access":false,"arxiv_categories":["cs.LG","stat.ML"],"arxiv_license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","arxiv_journal_ref":null,"mesh_headings":null,"chemicals":null,"comments_corrections":null,"source_flags":1,"s2_open_access_pdf_url":null,"s2_open_access_landing_url":null,"s2_open_access_license":null,"s2_open_access_status":null,"pmc_open_access_pdf_url":null,"pmc_open_access_landing_url":null,"pmc_open_access_license":null,"pmc_open_access_status":null,"unpaywall_open_access_pdf_url":null,"unpaywall_open_access_landing_url":null,"unpaywall_open_access_license":null,"unpaywall_open_access_status":null,"abstract":"Graph Convolutional Networks (GCNs) and their variants have experienced significant attention and have become the de facto methods for learning graph representations. GCNs derive inspiration primarily from recent deep learning approaches, and as a result, may inherit unnecessary complexity and redundant computation. In this paper, we reduce this excess complexity through successively removing nonlinearities and collapsing weight matrices between consecutive layers. We theoretically analyze the resulting linear model and show that it corresponds to a fixed low-pass filter followed by a linear classifier. Notably, our experimental evaluation demonstrates that these simplifications do not negatively impact accuracy in many downstream applications. Moreover, the resulting model scales to larger datasets, is naturally interpretable, and yields up to two orders of magnitude speedup over FastGCN.","claims":[{"public_id":"cl_45f78bccdd5be1a6c8a7dd3374ac116a","status":"active","text":"Removing nonlinearities and collapsing weight matrices between consecutive layers reduces excess complexity in graph convolutional networks.","confidence":0.97,"contributors":[{"id":1,"public_id":"12632b8b5f","public_label":"Anonymous (12632b8b5f)","roles":["extraction"],"url":"https://sah.borca.ai/u/12632b8b5f"}],"url":"https://sah.borca.ai/claims/cl_45f78bccdd5be1a6c8a7dd3374ac116a"},{"public_id":"cl_f7171a144b8913eb1591491c6a81f32f","status":"active","text":"The resulting linear model corresponds to a fixed low-pass filter followed by a linear classifier.","confidence":0.98,"contributors":[{"id":1,"public_id":"12632b8b5f","public_label":"Anonymous (12632b8b5f)","roles":["extraction"],"url":"https://sah.borca.ai/u/12632b8b5f"}],"url":"https://sah.borca.ai/claims/cl_f7171a144b8913eb1591491c6a81f32f"},{"public_id":"cl_80e3e7e410b57dc760a6f62c8e631429","status":"active","text":"The simplified model scales to larger datasets, is naturally interpretable, and yields up to two orders of magnitude speedup over FastGCN.","confidence":0.96,"contributors":[{"id":1,"public_id":"12632b8b5f","public_label":"Anonymous (12632b8b5f)","roles":["extraction"],"url":"https://sah.borca.ai/u/12632b8b5f"}],"url":"https://sah.borca.ai/claims/cl_80e3e7e410b57dc760a6f62c8e631429"},{"public_id":"cl_836caa7a9cf390fb1f72bf756b8d6725","status":"active","text":"These simplifications do not negatively impact accuracy in many downstream applications.","confidence":0.93,"contributors":[{"id":1,"public_id":"12632b8b5f","public_label":"Anonymous (12632b8b5f)","roles":["extraction"],"url":"https://sah.borca.ai/u/12632b8b5f"}],"url":"https://sah.borca.ai/claims/cl_836caa7a9cf390fb1f72bf756b8d6725"}],"concepts":[{"public_id":"co_138f2ddd92750d09d1ab55ca085f0872","status":"active","name":"graph convolutional networks","description":"Neural networks that learn representations on graph-structured data by aggregating information over neighbors.","types":["method"],"aliases":["GCNs"],"contributors":[{"id":1,"public_id":"12632b8b5f","public_label":"Anonymous (12632b8b5f)","roles":["extraction"],"url":"https://sah.borca.ai/u/12632b8b5f"}],"url":"https://sah.borca.ai/concepts/co_138f2ddd92750d09d1ab55ca085f0872"},{"public_id":"co_1a0807f5807b8a4231361a759470f912","status":"active","name":"linear model","description":"A model whose prediction is based on a linear transformation of its inputs.","types":["model"],"aliases":[],"contributors":[{"id":1,"public_id":"12632b8b5f","public_label":"Anonymous (12632b8b5f)","roles":["extraction"],"url":"https://sah.borca.ai/u/12632b8b5f"}],"url":"https://sah.borca.ai/concepts/co_1a0807f5807b8a4231361a759470f912"},{"public_id":"co_399b2e0b44687ac95c5905cd737fc274","status":"active","name":"accuracy","description":"A performance metric measuring the proportion of correct predictions.","types":["metric"],"aliases":[],"contributors":[{"id":1,"public_id":"12632b8b5f","public_label":"Anonymous (12632b8b5f)","roles":["extraction"],"url":"https://sah.borca.ai/u/12632b8b5f"}],"url":"https://sah.borca.ai/concepts/co_399b2e0b44687ac95c5905cd737fc274"},{"public_id":"co_3d0a0f64153fd1d0e95408ede63619ff","status":"active","name":"downstream applications","description":"Tasks or datasets used to evaluate the simplified graph model after training.","types":["evaluation setting"],"aliases":[],"contributors":[{"id":1,"public_id":"12632b8b5f","public_label":"Anonymous (12632b8b5f)","roles":["extraction"],"url":"https://sah.borca.ai/u/12632b8b5f"}],"url":"https://sah.borca.ai/concepts/co_3d0a0f64153fd1d0e95408ede63619ff"},{"public_id":"co_a110b470adbeedf180b9f2e67659c26d","status":"active","name":"nonlinearities","description":"Nonlinear activation functions used between layers in neural networks.","types":["model component"],"aliases":[],"contributors":[{"id":1,"public_id":"12632b8b5f","public_label":"Anonymous (12632b8b5f)","roles":["extraction"],"url":"https://sah.borca.ai/u/12632b8b5f"}],"url":"https://sah.borca.ai/concepts/co_a110b470adbeedf180b9f2e67659c26d"},{"public_id":"co_b746e2e2a4c171b3bcb8575d0659f93e","status":"active","name":"interpretable","description":"A model property indicating that its behavior can be more easily understood or explained.","types":["model property"],"aliases":[],"contributors":[{"id":1,"public_id":"12632b8b5f","public_label":"Anonymous (12632b8b5f)","roles":["extraction"],"url":"https://sah.borca.ai/u/12632b8b5f"}],"url":"https://sah.borca.ai/concepts/co_b746e2e2a4c171b3bcb8575d0659f93e"},{"public_id":"co_b79b7d6a16399b31608a46d50e5261ff","status":"active","name":"linear classifier","description":"A classifier that separates classes using a linear decision function.","types":["model"],"aliases":[],"contributors":[{"id":1,"public_id":"12632b8b5f","public_label":"Anonymous (12632b8b5f)","roles":["extraction"],"url":"https://sah.borca.ai/u/12632b8b5f"}],"url":"https://sah.borca.ai/concepts/co_b79b7d6a16399b31608a46d50e5261ff"},{"public_id":"co_b840e3ec5f04d3d754f23970dcc0fd7c","status":"active","name":"fixed low-pass filter","description":"A filter that preserves low-frequency signal components while attenuating high-frequency components.","types":["signal processing concept"],"aliases":[],"contributors":[{"id":1,"public_id":"12632b8b5f","public_label":"Anonymous (12632b8b5f)","roles":["extraction"],"url":"https://sah.borca.ai/u/12632b8b5f"}],"url":"https://sah.borca.ai/concepts/co_b840e3ec5f04d3d754f23970dcc0fd7c"},{"public_id":"co_da36efef54dfa8addede1292a8ad9bf2","status":"active","name":"weight matrices","description":"Learnable matrices that transform features between consecutive neural network layers.","types":["model component"],"aliases":[],"contributors":[{"id":1,"public_id":"12632b8b5f","public_label":"Anonymous (12632b8b5f)","roles":["extraction"],"url":"https://sah.borca.ai/u/12632b8b5f"}],"url":"https://sah.borca.ai/concepts/co_da36efef54dfa8addede1292a8ad9bf2"},{"public_id":"co_e679b74a4f80fda4890319b8a6c5876e","status":"active","name":"larger datasets","description":"Datasets with more instances or larger graph scale used to test scalability.","types":["dataset"],"aliases":[],"contributors":[{"id":1,"public_id":"12632b8b5f","public_label":"Anonymous (12632b8b5f)","roles":["extraction"],"url":"https://sah.borca.ai/u/12632b8b5f"}],"url":"https://sah.borca.ai/concepts/co_e679b74a4f80fda4890319b8a6c5876e"},{"public_id":"co_fb9bddcc2afc90d7c12a6f1387573ee8","status":"active","name":"FastGCN","description":"A graph convolutional network variant used as a speed comparison baseline.","types":["baseline"],"aliases":[],"contributors":[{"id":1,"public_id":"12632b8b5f","public_label":"Anonymous (12632b8b5f)","roles":["extraction"],"url":"https://sah.borca.ai/u/12632b8b5f"}],"url":"https://sah.borca.ai/concepts/co_fb9bddcc2afc90d7c12a6f1387573ee8"}],"external_ids":{"DOI":null,"ArXiv":"1902.07153","PubMed":null,"PubMedCentral":null,"MAG":2964124573,"DBLP":"conf/icml/WuSZFYW19","ACL":null},"open_access":{"is_open_access":true,"pdf_url":"https://arxiv.org/pdf/1902.07153","landing_url":"https://arxiv.org/abs/1902.07153","source":"arxiv","pdf_url_source":"derived_arxiv","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","reason":null},"reference_availability":{"status":"available","references_indexed":true,"full_text_available":true,"full_text_source":"arxiv","count_basis":"semantic_scholar_metadata","extraction_status":"not_applicable","reason":null},"source":{"provider":"episteme2","base_corpus":"semantic_scholar_dump","freshness_mode":"unknown","basis":["semantic_scholar_metadata","postgres_metadata"],"limits":["paper metadata is based on indexed upstream scholarly datasets","claims and concepts are available only for extracted papers","absence of claims or concepts means no extracted graph data is available in this response"],"status":"available","degraded":false,"degraded_reasons":[],"diagnostics":{"status":"available","degraded":false,"degraded_reasons":[],"metadata_status":"available","graph_status":"available","abstract_status":"available"},"source_flags":1},"paper_id":635401,"paper_uid":"b07da66c-004d-4078-b182-fea3eea7d70a","canonical_identity":{"paper_id":635401,"paper_uid":"b07da66c-004d-4078-b182-fea3eea7d70a","identity_status":"available","lookup_basis":"semantic_scholar_external_id","compatibility_path":"corpus_id"},"url":"https://sah.borca.ai/papers/67752026"}