{"id":"https://openalex.org/W7125961589","doi":"https://doi.org/10.13016/wez2-5y9v","title":"ON THE EMERGENT STRUCTURES OF DEEP NEURAL NETWORKS","display_name":"ON THE EMERGENT STRUCTURES OF DEEP NEURAL NETWORKS","publication_year":2025,"publication_date":"2025-01-01","ids":{"openalex":"https://openalex.org/W7125961589","doi":"https://doi.org/10.13016/wez2-5y9v"},"language":"en","primary_location":{"id":"pmh:oai:drum.lib.umd.edu:1903/35123","doi":null,"is_oa":false,"landing_page_url":"https://doi.org/10.13016/wez2-5y9v","pdf_url":null,"source":{"id":"https://openalex.org/S4306401518","display_name":"University Libraries (University of Maryland)","issn_l":null,"issn":null,"is_oa":false,"is_in_doaj":false,"is_core":false,"listed_in":[],"host_organization":"https://openalex.org/I66946132","host_organization_name":"University of Maryland, College Park","host_organization_lineage":["https://openalex.org/I66946132"],"host_organization_lineage_names":[],"type":"repository"},"license":null,"license_id":null,"version":"submittedVersion","is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"Dissertation"},"volume":null,"type":"dissertation","indexed_in":["datacite"],"open_access":{"is_oa":true,"oa_status":"green","oa_url":"https://doi.org/10.13016/wez2-5y9v","any_repository_has_fulltext":true},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5125051359","display_name":"Tianyu He","orcid":null,"observed_orcids":[]},"institutions":[],"countries":[],"is_corresponding":true,"raw_author_name":"He, Tianyu","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]}],"institutions":[],"countries_distinct_count":0,"institutions_distinct_count":0,"corresponding_author_ids":["https://openalex.org/A5125051359"],"corresponding_institution_ids":[],"apc_list":null,"apc_paid":null,"fwci":null,"has_fulltext":false,"cited_by_count":0,"citation_normalized_percentile":null,"cited_by_percentile_year":null,"biblio":{"volume":null,"issue":null,"first_page":null,"last_page":null},"is_retracted":false,"is_paratext":false,"is_xpac":false,"primary_topic":{"id":"https://openalex.org/T10320","display_name":"Neural Networks and Applications","score":0.8144999742507935,"subfield":{"id":"https://openalex.org/subfields/1702","display_name":"Artificial Intelligence"},"field":{"id":"https://openalex.org/fields/17","display_name":"Computer Science"},"domain":{"id":"https://openalex.org/domains/3","display_name":"Physical Sciences"}},"topics":[{"id":"https://openalex.org/T10320","display_name":"Neural Networks and Applications","score":0.8144999742507935,"subfield":{"id":"https://openalex.org/subfields/1702","display_name":"Artificial Intelligence"},"field":{"id":"https://openalex.org/fields/17","display_name":"Computer Science"},"domain":{"id":"https://openalex.org/domains/3","display_name":"Physical Sciences"}},{"id":"https://openalex.org/T11612","display_name":"Stochastic Gradient Optimization Techniques","score":0.17030000686645508,"subfield":{"id":"https://openalex.org/subfields/1702","display_name":"Artificial Intelligence"},"field":{"id":"https://openalex.org/fields/17","display_name":"Computer Science"},"domain":{"id":"https://openalex.org/domains/3","display_name":"Physical Sciences"}},{"id":"https://openalex.org/T11307","display_name":"Domain Adaptation and Few-Shot Learning","score":0.0035000001080334187,"subfield":{"id":"https://openalex.org/subfields/1702","display_name":"Artificial Intelligence"},"field":{"id":"https://openalex.org/fields/17","display_name":"Computer Science"},"domain":{"id":"https://openalex.org/domains/3","display_name":"Physical Sciences"}}],"keywords":[{"id":"https://openalex.org/keywords/deep-neural-network","display_name":"deep neural network","score":0.7329999804496765},{"id":"https://openalex.org/keywords/signal-propagation","display_name":"signal propagation","score":0.718999981880188},{"id":"https://openalex.org/keywords/infinite-width-limit","display_name":"infinite-width limit","score":0.7480000257492065},{"id":"https://openalex.org/keywords/phase-diagram","display_name":"phase diagram","score":0.5370000004768372},{"id":"https://openalex.org/keywords/network-initialization","display_name":"network initialization","score":0.7730000019073486},{"id":"https://openalex.org/keywords/sharpness","display_name":"sharpness","score":0.8659999966621399},{"id":"https://openalex.org/keywords/gradient-descent-dynamics","display_name":"gradient descent dynamics","score":0.9559999704360962},{"id":"https://openalex.org/keywords/pseudorandom-number-generators","display_name":"pseudorandom number generators","score":0.8009999990463257},{"id":"https://openalex.org/keywords/dataset-diversity","display_name":"data diversity","score":0.527999997138977},{"id":"https://openalex.org/keywords/network-architecture","display_name":"network architecture","score":0.6370000243186951},{"id":"https://openalex.org/keywords/interpretability","display_name":"interpretability","score":0.42500001192092896},{"id":"https://openalex.org/keywords/vision-language-models","display_name":"vision-language models","score":0.6600000262260437}],"concepts":[{"id":"https://openalex.org/C114466953","wikidata":"https://www.wikidata.org/wiki/Q6034165","display_name":"Initialization","level":2,"score":0.6869999766349792},{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.6445000171661377},{"id":"https://openalex.org/C154945302","wikidata":"https://www.wikidata.org/wiki/Q11660","display_name":"Artificial intelligence","level":1,"score":0.6154000163078308},{"id":"https://openalex.org/C153258448","wikidata":"https://www.wikidata.org/wiki/Q1199743","display_name":"Gradient descent","level":3,"score":0.5855000019073486},{"id":"https://openalex.org/C50644808","wikidata":"https://www.wikidata.org/wiki/Q192776","display_name":"Artificial neural network","level":2,"score":0.49779999256134033},{"id":"https://openalex.org/C32848918","wikidata":"https://www.wikidata.org/wiki/Q845789","display_name":"Observable","level":2,"score":0.46779999136924744},{"id":"https://openalex.org/C176217482","wikidata":"https://www.wikidata.org/wiki/Q860554","display_name":"Metric (unit)","level":2,"score":0.4584999978542328},{"id":"https://openalex.org/C200331156","wikidata":"https://www.wikidata.org/wiki/Q506041","display_name":"Jacobian matrix and determinant","level":2,"score":0.44609999656677246},{"id":"https://openalex.org/C108583219","wikidata":"https://www.wikidata.org/wiki/Q197536","display_name":"Deep learning","level":2,"score":0.43560001254081726},{"id":"https://openalex.org/C173801870","wikidata":"https://www.wikidata.org/wiki/Q201413","display_name":"Heuristic","level":2,"score":0.4169999957084656},{"id":"https://openalex.org/C203616005","wikidata":"https://www.wikidata.org/wiki/Q620495","display_name":"Hessian matrix","level":2,"score":0.41519999504089355},{"id":"https://openalex.org/C14036430","wikidata":"https://www.wikidata.org/wiki/Q3736076","display_name":"Function (biology)","level":2,"score":0.3337000012397766},{"id":"https://openalex.org/C84269361","wikidata":"https://www.wikidata.org/wiki/Q179235","display_name":"Phenomenology (philosophy)","level":2,"score":0.3246000111103058},{"id":"https://openalex.org/C80444323","wikidata":"https://www.wikidata.org/wiki/Q2878974","display_name":"Theoretical computer science","level":1,"score":0.3116999864578247},{"id":"https://openalex.org/C119857082","wikidata":"https://www.wikidata.org/wiki/Q2539","display_name":"Machine learning","level":1,"score":0.29649999737739563},{"id":"https://openalex.org/C34947359","wikidata":"https://www.wikidata.org/wiki/Q665189","display_name":"Complex network","level":2,"score":0.28450000286102295},{"id":"https://openalex.org/C106301342","wikidata":"https://www.wikidata.org/wiki/Q4117933","display_name":"Entropy (arrow of time)","level":2,"score":0.27619999647140503},{"id":"https://openalex.org/C191795146","wikidata":"https://www.wikidata.org/wiki/Q3878446","display_name":"Norm (philosophy)","level":2,"score":0.27619999647140503},{"id":"https://openalex.org/C11413529","wikidata":"https://www.wikidata.org/wiki/Q8366","display_name":"Algorithm","level":1,"score":0.2745000123977661},{"id":"https://openalex.org/C2778334786","wikidata":"https://www.wikidata.org/wiki/Q1586270","display_name":"Variation (astronomy)","level":2,"score":0.271699994802475},{"id":"https://openalex.org/C11210021","wikidata":"https://www.wikidata.org/wiki/Q1520713","display_name":"Linearization","level":3,"score":0.2660999894142151},{"id":"https://openalex.org/C81363708","wikidata":"https://www.wikidata.org/wiki/Q17084460","display_name":"Convolutional neural network","level":2,"score":0.26409998536109924},{"id":"https://openalex.org/C2984842247","wikidata":"https://www.wikidata.org/wiki/Q197536","display_name":"Deep neural networks","level":3,"score":0.2603999972343445},{"id":"https://openalex.org/C137836250","wikidata":"https://www.wikidata.org/wiki/Q984063","display_name":"Optimization problem","level":2,"score":0.26030001044273376},{"id":"https://openalex.org/C136389625","wikidata":"https://www.wikidata.org/wiki/Q334384","display_name":"Supervised learning","level":3,"score":0.25999999046325684}],"mesh":[],"locations_count":2,"locations":[{"id":"pmh:oai:drum.lib.umd.edu:1903/35123","doi":null,"is_oa":false,"landing_page_url":"https://doi.org/10.13016/wez2-5y9v","pdf_url":null,"source":{"id":"https://openalex.org/S4306401518","display_name":"University Libraries (University of Maryland)","issn_l":null,"issn":null,"is_oa":false,"is_in_doaj":false,"is_core":false,"listed_in":[],"host_organization":"https://openalex.org/I66946132","host_organization_name":"University of Maryland, College Park","host_organization_lineage":["https://openalex.org/I66946132"],"host_organization_lineage_names":[],"type":"repository"},"license":null,"license_id":null,"version":"submittedVersion","is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"Dissertation"},{"id":"doi:10.13016/wez2-5y9v","doi":"https://doi.org/10.13016/wez2-5y9v","is_oa":true,"landing_page_url":"https://doi.org/10.13016/wez2-5y9v","pdf_url":null,"source":{"id":"https://openalex.org/S4306402644","display_name":"Digital Repository at the University of Maryland (University of Maryland College Park)","issn_l":null,"issn":null,"is_oa":false,"is_in_doaj":false,"is_core":false,"listed_in":[],"host_organization":"https://openalex.org/I66946132","host_organization_name":"University of Maryland, College Park","host_organization_lineage":["https://openalex.org/I66946132"],"host_organization_lineage_names":[],"type":"repository"},"license":null,"license_id":null,"version":null,"is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"Collection"}],"best_oa_location":{"id":"doi:10.13016/wez2-5y9v","doi":"https://doi.org/10.13016/wez2-5y9v","is_oa":true,"landing_page_url":"https://doi.org/10.13016/wez2-5y9v","pdf_url":null,"source":{"id":"https://openalex.org/S4306402644","display_name":"Digital Repository at the University of Maryland (University of Maryland College Park)","issn_l":null,"issn":null,"is_oa":false,"is_in_doaj":false,"is_core":false,"listed_in":[],"host_organization":"https://openalex.org/I66946132","host_organization_name":"University of Maryland, College Park","host_organization_lineage":["https://openalex.org/I66946132"],"host_organization_lineage_names":[],"type":"repository"},"license":null,"license_id":null,"version":null,"is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"Collection"},"sustainable_development_goals":[],"study_designs":[],"awards":[],"funders":[],"has_content":{"grobid_xml":false,"pdf":false},"content_urls":null,"referenced_works_count":0,"referenced_works":[],"related_works":["https://openalex.org/W7133184177","https://openalex.org/W6947179685","https://openalex.org/W6926325669","https://openalex.org/W6922517452","https://openalex.org/W7125506360","https://openalex.org/W4417123928","https://openalex.org/W3123178201","https://openalex.org/W4388444893","https://openalex.org/W7162449434","https://openalex.org/W4417446068"],"abstract_inverted_index":{"Despite":[0],"the":[1,69,84,89,112,124,159,164,173,176,200,210,221,231,245,260,330,365,380,383,386],"remarkable":[2],"success":[3],"of":[4,14,72,167,175,212,263,354,367,382],"artificial":[5],"intelligence":[6],"(AI),":[7],"we":[8,75,99,131,162,184,219,248,279,310,333,363],"still":[9],"lack":[10],"a":[11,43,101,151,188,195,315,341],"fundamental":[12,222],"understanding":[13,313],"why":[15,20],"deep":[16],"learning":[17,242],"works":[18],"--":[19],"certain":[21,294],"architectural":[22,301,336],"designs":[23],"are":[24],"preferred,":[25],"how":[26,250],"optimization":[27,165],"navigates":[28],"complex":[29,50],"landscapes,":[30],"and":[31,53,114,139,236,253,284,303,360,376,385],"what":[32],"data":[33,251,304,384],"characteristics":[34,252],"enable":[35],"learning.":[36],"This":[37],"dissertation":[38],"addresses":[39],"these":[40],"questions":[41],"through":[42],"phenomenological":[44],"approach,":[45],"abstracting":[46],"essential":[47,201],"features":[48],"from":[49,329],"real-world":[51,206],"experiments":[52],"constructing":[54],"minimal":[55,189,285],"setups":[56],"that":[57,116,133,186,293,344],"faithfully":[58],"capture":[59],"them.":[60],"The":[61,108],"investigation":[62],"unfolds":[63],"in":[64,79,205,216,357],"three":[65],"interconnected":[66],"parts.":[67],"In":[68,158,244],"first":[70],"part":[71],"this":[73,134,312],"dissertation,":[74],"study":[76],"signal":[77],"propagation":[78],"randomly":[80],"initialized":[81],"networks":[82,118,169,270],"under":[83,180],"infinite-width":[85],"limit.":[86],"By":[87,208],"examining":[88],"inter-layer":[90,348],"Jacobian":[91],"norm":[92],"as":[93],"an":[94,140],"observable":[95,109],"at":[96,123,240],"large":[97,241],"depth,":[98],"identify":[100],"phase":[102,125],"diagram":[103],"with":[104],"two":[105],"distinct":[106,368],"phases.":[107],"precisely":[110],"characterizes":[111],"transition":[113],"predicts":[115],"neural":[117,168,264],"best":[119],"maintain":[120],"signals":[121],"exactly":[122],"boundary.":[126],"Through":[127,351],"extensive":[128],"empirical":[129],"validation,":[130],"demonstrate":[132],"metric":[135],"provides":[136],"universal":[137],"guidance":[138],"efficient":[141],"algorithm":[142],"for":[143,288],"selecting":[144],"network":[145,192,254,286],"initializations":[146],"across":[147],"diverse":[148],"architectures,":[149],"offering":[150],"principled":[152],"alternative":[153],"to":[154,259,314],"heuristic":[155],"initialization":[156],"schemes.":[157],"second":[160],"part,":[161,247],"elucidate":[163],"landscape":[166],"by":[170,268,319,372],"carefully":[171],"analyzing":[172,209,269],"dynamics":[174,239],"loss":[177],"Hessian":[178],"(sharpness)":[179],"gradient":[181,237],"descent.":[182],"Remarkably,":[183],"show":[185],"even":[187],"two-layer":[190],"linear":[191],"trained":[193,271,355],"on":[194,272],"single":[196],"example":[197],"exhibits":[198],"all":[199],"sharpness":[202,225],"phenomenology":[203],"observed":[204],"scenarios.":[207],"structure":[211],"dynamical":[213],"fixed":[214],"points":[215],"function":[217],"space,":[218],"reveal":[220],"mechanisms":[223],"governing":[224],"evolution":[226],"during":[227],"training,":[228],"thereby":[229],"clarifying":[230],"interplay":[232],"between":[233,379],"initialization,":[234],"parameterization,":[235],"descent":[238],"rates.":[243],"third":[246],"investigate":[249],"architecture":[255],"choices":[256],"give":[257],"rise":[258],"emergent":[261],"capabilities":[262,296],"networks.":[265],"We":[266,339],"begin":[267],"modular":[273],"arithmetic":[274],"tasks.":[275,325],"Across":[276],"different":[277],"setups,":[278],"establish":[280],"necessary":[281],"data-diversity":[282],"thresholds":[283],"sizes":[287],"successful":[289],"task":[290],"performance,":[291],"demonstrating":[292],"computational":[295,389],"emerge":[297],"only":[298],"when":[299],"both":[300,358],"capacity":[302],"coverage":[305],"exceed":[306],"critical":[307],"values.":[308],"Furthermore,":[309],"extend":[311],"more":[316],"practical":[317],"setting":[318],"tackling":[320],"Pseudo-Random":[321],"Number":[322],"Generator":[323],"(PRNG)":[324],"Finally,":[326],"synthesizing":[327],"insights":[328],"preceding":[331],"parts,":[332],"challenge":[334],"conventional":[335],"design":[337],"paradigms.":[338],"propose":[340],"new":[342],"(proto-)architecture":[343],"enables":[345],"arbitrary,":[346],"learned":[347],"connectivity":[349,369],"patterns.":[350,390],"careful":[352],"analysis":[353],"models":[356],"vision":[359],"language":[361],"domains,":[362],"observe":[364],"emergence":[366],"structures,":[370],"accompanied":[371],"interpretable":[373],"layer":[374],"specialization":[375],"meaningful":[377],"correlations":[378],"nature":[381],"network\u2019s":[387],"internal":[388]},"counts_by_year":[],"updated_date":"2026-10-01T08:54:57.152042","created_date":"2026-01-29T00:00:00"}