{"id":"https://openalex.org/W2908087012","doi":null,"title":"A Scalable Framework for Acceleration of CNN Training on Deeply-Pipelined FPGA Clusters with Weight and Workload Balancing.","display_name":"A Scalable Framework for Acceleration of CNN Training on Deeply-Pipelined FPGA Clusters with Weight and Workload Balancing.","publication_year":2019,"publication_date":"2019-01-04","ids":{"openalex":"https://openalex.org/W2908087012","mag":"2908087012"},"language":"en","primary_location":{"id":"mag:2908087012","doi":null,"is_oa":false,"landing_page_url":"http://dblp.uni-trier.de/db/journals/corr/corr1901.html#abs-1901-01007","pdf_url":null,"source":null,"license":null,"license_id":null,"version":"submittedVersion","is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":null},"volume":null,"type":"article","is_traditional":true,"indexed_in":[],"open_access":{"is_oa":false,"oa_status":"closed","oa_url":null,"any_repository_has_fulltext":false},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5078443672","display_name":"Tong Geng","orcid":"https://orcid.org/0000-0002-3644-2922","observed_orcids":["https://orcid.org/0000-0002-3644-2922"]},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Tong Geng","raw_affiliation_strings":[],"raw_orcid":"https://orcid.org/0000-0002-3644-2922","affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5100734913","display_name":"Tianqi Wang","orcid":"https://orcid.org/0000-0002-5921-6565","observed_orcids":["https://orcid.org/0000-0002-5921-6565"]},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Tianqi Wang","raw_affiliation_strings":[],"raw_orcid":"https://orcid.org/0000-0002-5921-6565","affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5100413657","display_name":"Ang Li","orcid":"https://orcid.org/0000-0003-3734-9137","observed_orcids":["https://orcid.org/0000-0003-3734-9137"]},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Ang Li","raw_affiliation_strings":[],"raw_orcid":"https://orcid.org/0000-0003-3734-9137","affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5046555671","display_name":"Xi Jin","orcid":"https://orcid.org/0009-0009-1019-4324","observed_orcids":["https://orcid.org/0009-0009-1019-4324"]},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Xi Jin","raw_affiliation_strings":[],"raw_orcid":"https://orcid.org/0000-0003-2087-0698","affiliations":[]},{"author_position":"last","author":{"id":"https://openalex.org/A5021051610","display_name":"Martin Herbordt","orcid":"https://orcid.org/0000-0002-3443-9113","observed_orcids":["https://orcid.org/0000-0002-3443-9113"]},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Martin C. Herbordt","raw_affiliation_strings":[],"raw_orcid":"https://orcid.org/0000-0002-3443-9113","affiliations":[]}],"institutions":[],"countries_distinct_count":0,"institutions_distinct_count":0,"corresponding_author_ids":[],"corresponding_institution_ids":[],"apc_list":null,"apc_paid":null,"fwci":0.9333,"has_fulltext":false,"cited_by_count":4,"citation_normalized_percentile":{"value":0.74730672,"is_in_top_1_percent":false,"is_in_top_10_percent":false},"cited_by_percentile_year":{"min":87,"max":95},"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/T10904","display_name":"Embedded Systems Design Techniques","score":0.9106000065803528,"subfield":{"id":"https://openalex.org/subfields/1708","display_name":"Hardware and Architecture"},"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/T10904","display_name":"Embedded Systems Design Techniques","score":0.9106000065803528,"subfield":{"id":"https://openalex.org/subfields/1708","display_name":"Hardware and Architecture"},"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/T10036","display_name":"Advanced Neural Network Applications","score":0.04450000077486038,"subfield":{"id":"https://openalex.org/subfields/1707","display_name":"Computer Vision and Pattern Recognition"},"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/T10054","display_name":"Parallel Computing and Optimization Techniques","score":0.017500000074505806,"subfield":{"id":"https://openalex.org/subfields/1708","display_name":"Hardware and Architecture"},"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/fpga-clusters","display_name":"FPGA clusters","score":0.7570000290870667},{"id":"https://openalex.org/keywords/cnn-training","display_name":"CNN training","score":0.6010000109672546},{"id":"https://openalex.org/keywords/distributed-training","display_name":"distributed training","score":0.5899999737739563},{"id":"https://openalex.org/keywords/model-parallelism","display_name":"model parallelism","score":0.7549999952316284},{"id":"https://openalex.org/keywords/load-balancing","display_name":"load balancing","score":0.9890000224113464},{"id":"https://openalex.org/keywords/weight-balancing","display_name":"weight balancing","score":0.5350000262260437},{"id":"https://openalex.org/keywords/pipeline-parallelism","display_name":"pipeline parallelism","score":0.5559999942779541},{"id":"https://openalex.org/keywords/reconfigurable-computing","display_name":"reconfigurable computing","score":0.6919999718666077},{"id":"https://openalex.org/keywords/energy-efficiency","display_name":"energy efficiency","score":0.5550000071525574},{"id":"https://openalex.org/keywords/on-chip-memory","display_name":"on-chip memory","score":0.6620000004768372}],"concepts":[{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.8382445573806763},{"id":"https://openalex.org/C48044578","wikidata":"https://www.wikidata.org/wiki/Q727490","display_name":"Scalability","level":2,"score":0.7185180187225342},{"id":"https://openalex.org/C173608175","wikidata":"https://www.wikidata.org/wiki/Q232661","display_name":"Parallel computing","level":1,"score":0.6122163534164429},{"id":"https://openalex.org/C42935608","wikidata":"https://www.wikidata.org/wiki/Q190411","display_name":"Field-programmable gate array","level":2,"score":0.5752643942832947},{"id":"https://openalex.org/C115537543","wikidata":"https://www.wikidata.org/wiki/Q165596","display_name":"Cache","level":2,"score":0.48663362860679626},{"id":"https://openalex.org/C2742236","wikidata":"https://www.wikidata.org/wiki/Q924713","display_name":"Efficient energy use","level":2,"score":0.44667595624923706},{"id":"https://openalex.org/C2778476105","wikidata":"https://www.wikidata.org/wiki/Q628539","display_name":"Workload","level":2,"score":0.4410242736339569},{"id":"https://openalex.org/C120314980","wikidata":"https://www.wikidata.org/wiki/Q180634","display_name":"Distributed computing","level":1,"score":0.42524927854537964},{"id":"https://openalex.org/C149635348","wikidata":"https://www.wikidata.org/wiki/Q193040","display_name":"Embedded system","level":1,"score":0.35300618410110474},{"id":"https://openalex.org/C111919701","wikidata":"https://www.wikidata.org/wiki/Q9135","display_name":"Operating system","level":1,"score":0.1512964963912964},{"id":"https://openalex.org/C127413603","wikidata":"https://www.wikidata.org/wiki/Q11023","display_name":"Engineering","level":0,"score":0.0},{"id":"https://openalex.org/C119599485","wikidata":"https://www.wikidata.org/wiki/Q43035","display_name":"Electrical engineering","level":1,"score":0.0}],"mesh":[],"locations_count":1,"locations":[{"id":"mag:2908087012","doi":null,"is_oa":false,"landing_page_url":"http://dblp.uni-trier.de/db/journals/corr/corr1901.html#abs-1901-01007","pdf_url":null,"source":null,"license":null,"license_id":null,"version":"submittedVersion","is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":null}],"best_oa_location":null,"sustainable_development_goals":[],"study_designs":[],"awards":[],"funders":[],"has_content":{"pdf":false,"grobid_xml":false},"content_urls":null,"referenced_works_count":36,"referenced_works":["https://openalex.org/W2083842231","https://openalex.org/W2094756095","https://openalex.org/W2186615578","https://openalex.org/W2271840356","https://openalex.org/W2272300165","https://openalex.org/W2336650964","https://openalex.org/W2523060838","https://openalex.org/W2525778437","https://openalex.org/W2542189141","https://openalex.org/W2557355796","https://openalex.org/W2559305746","https://openalex.org/W2591253321","https://openalex.org/W2597692291","https://openalex.org/W2617411258","https://openalex.org/W2622263826","https://openalex.org/W2625954420","https://openalex.org/W2727238169","https://openalex.org/W2729080111","https://openalex.org/W2764337919","https://openalex.org/W2768797272","https://openalex.org/W2769918628","https://openalex.org/W2786066748","https://openalex.org/W2798956872","https://openalex.org/W2799069271","https://openalex.org/W2812009592","https://openalex.org/W2883929540","https://openalex.org/W2890068895","https://openalex.org/W2901299405","https://openalex.org/W2903298089","https://openalex.org/W2904445802","https://openalex.org/W2949744956","https://openalex.org/W2963228337","https://openalex.org/W2969388332","https://openalex.org/W2972087877","https://openalex.org/W2983971024","https://openalex.org/W7098642242"],"related_works":["https://openalex.org/W3035430964","https://openalex.org/W2904445802","https://openalex.org/W2890068895","https://openalex.org/W4401212495","https://openalex.org/W2949469598","https://openalex.org/W4200617448","https://openalex.org/W3116045160","https://openalex.org/W4292951128","https://openalex.org/W7203734226","https://openalex.org/W2955052486"],"abstract_inverted_index":{"Deep":[0],"Neural":[1],"Networks":[2],"(DNNs)":[3],"have":[4],"revolutionized":[5],"numerous":[6,103],"applications,":[7],"but":[8],"the":[9,40,49,56,70,106,131,148,167,176,183,204,208],"demand":[10,163],"for":[11,157,175],"ever":[12],"more":[13],"performance":[14],"remains":[15],"unabated.":[16],"Scaling":[17],"DNN":[18],"computations":[19],"to":[20,47,59,92,97,122,140,152,166,197,220,229],"larger":[21],"clusters":[22,96],"is":[23,45,134,164,173,225],"generally":[24],"done":[25],"by":[26],"distributing":[27],"tasks":[28],"in":[29,69],"batch":[30,112],"mode":[31],"using":[32],"methods":[33],"such":[34],"as":[35],"distributed":[36,50,58,94],"synchronous":[37],"SGD.":[38],"Among":[39],"issues":[41],"with":[42,53,182,203,227],"this":[43,75],"approach":[44,101],"that":[46,192],"make":[48],"cluster":[51],"work":[52],"high":[54,141],"utilization,":[55],"workload":[57,117],"each":[60],"node":[61],"must":[62],"be":[63,153],"large,":[64],"which":[65,83],"implies":[66],"nontrivial":[67],"growth":[68],"SGD":[71],"mini-batch":[72],"size.":[73],"\r\nIn":[74],"paper,":[76],"we":[77],"propose":[78],"a":[79,85,135,160,198],"framework":[80],"called":[81],"FPDeep,":[82],"uses":[84],"hybrid":[86],"of":[87,125,201],"model":[88],"and":[89,118,143,145,186],"layer":[90],"parallelism":[91,142],"configure":[93],"reconfigurable":[95],"train":[98],"DNNs.":[99],"This":[100,138],"has":[102,194],"benefits.":[104],"First,":[105],"design":[107],"does":[108],"not":[109],"suffer":[110],"from":[111],"size":[113],"growth.":[114],"Second,":[115],"novel":[116],"weight":[119],"partitioning":[120],"leads":[121,139],"balanced":[123],"loads":[124],"both":[126],"among":[127],"nodes.":[128],"And":[129],"third,":[130],"entire":[132],"system":[133],"fine-grained":[136],"pipeline.":[137],"utilization":[144],"also":[146],"minimizes":[147],"time":[149],"features":[150],"need":[151],"cached":[154],"while":[155],"waiting":[156],"back-propagation.":[158],"As":[159],"result,":[161],"storage":[162],"reduced":[165],"point":[168],"where":[169],"only":[170],"on-chip":[171],"memory":[172],"used":[174],"convolution":[177],"layers.":[178],"We":[179],"evaluate":[180],"FPDeep":[181,193,216,231],"Alexnet,":[184],"VGG-16,":[185],"VGG-19":[187],"benchmarks.":[188],"Experimental":[189],"results":[190],"show":[191],"good":[195],"scalability":[196],"large":[199],"number":[200],"FPGAs,":[202],"limiting":[205],"factor":[206],"being":[207],"FPGA-to-FPGA":[209],"bandwidth.":[210],"With":[211],"6":[212],"transceivers":[213],"per":[214],"FPGA,":[215],"shows":[217],"linearity":[218],"up":[219],"83":[221],"FPGAs.":[222],"Energy":[223],"efficiency":[224,238],"evaluated":[226],"respect":[228],"GOPs/J.":[230],"provides,":[232],"on":[233],"average,":[234],"6.36x":[235],"higher":[236],"energy":[237],"than":[239],"comparable":[240],"GPU":[241],"servers.":[242]},"counts_by_year":[{"year":2021,"cited_by_count":1},{"year":2020,"cited_by_count":2},{"year":2019,"cited_by_count":1}],"updated_date":"2026-11-04T10:54:05.112534","created_date":"2019-01-11T00:00:00"}