Accelerating Machine Learning Research in Fusion Through a Pydantic‑Centred Human‑in‑the‑Loop Architecture

{"Training":[0],"high-performing":[1],"ML":[2,44,137],"models":[3,102,138],"requires":[4],"large":[5],"volumes":[6],"of":[7,29,35,133],"high":[8],"quality,":[9],"labelled":[10],"data.":[11],"While":[12],"nuclear":[13],"fusion":[14],"experiments":[15],"generate":[16],"extensive":[17],"datasets":[18],"–":[19,33],"JET":[20,210],"produced":[21],"105,929":[22],"pulses,":[23],"each":[24],"with":[25,107,130,145,220],"more":[26],"than":[27,173],"10GB":[28],"raw":[30],"data":[31,37,87,101,124,189,215],"[1]":[32],"much":[34],"this":[36,68],"lacks":[38],"the":[39,58,143,152,161,199],"annotations":[40,109],"necessary":[41],"for":[42,83,116,163],"effective":[43],"training.":[45],"Retrospective":[46],"annotation":[47,81],"is":[48],"both":[49],"challenging":[50],"and":[51,71,95,147,154,184,192,217,227],"time":[52],"consuming,":[53],"as":[54],"domain":[55,72,185],"experts":[56,73],"lack":[57],"appropriate":[59],"tools":[60],"to":[61,76,223],"effectively":[62],"label":[63],"historic":[64],"pulses.":[65],"To":[66],"address":[67],"challenge,":[69],"RSEs":[70,183],"have":[74],"collaborated":[75],"develop":[77],"an":[78],"open-source":[79],"human-in-the-loop":[80],"platform":[82],"labelling":[84,174],"tokamak":[85],"diagnostic":[86],"[2].":[88],"This":[89,177],"combines":[90],"a":[91,96,112,195],"web-based":[92],"user":[93],"interface":[94],"REST":[97],"API":[98],"built":[99],"around":[100],"defined":[103,141],"using":[104,156],"Pydantic":[105],"[3],":[106],"validated":[108],"stored":[110],"in":[111,198,211],"MongoDB":[113,240],"[4]":[114],"database":[115],"later":[117],"reuse.":[118],"The":[119,239],"modular":[120],"design":[121],"supports":[122],"multiple":[123],"access":[125],"layers":[126],"from":[127,175],"different":[128],"machines,":[129],"simple":[131],"integration":[132],"new":[134],"sources.":[135],"Custom":[136],"can":[139,166],"be":[140],"by":[142],"user,":[144],"training":[146],"prediction":[148],"tasks":[149],"scheduled":[150],"via":[151],"UI":[153],"executed":[155],"Ray":[157],"[5].":[158],"These":[159],"reduce":[160],"workload":[162],"scientists,":[164],"who":[165],"focus":[167],"on":[168,188],"refining":[169],"model":[170],"predictions":[171],"rather":[172],"scratch.":[176],"demonstrates":[178],"how":[179],"infrastructure":[180],"co-developed":[181],"between":[182],"experts,":[186],"focussed":[187],"validation,":[190],"extensibility":[191],"reusability,":[193],"plays":[194],"key":[196],"role":[197],"research":[200],"journey.":[201],"[1]:":[202],"Vega,":[203],"J.,":[204],"et":[205],"al.":[206],"\\"New":[207],"developments":[208],"at":[209],"diagnostics,":[212],"real-time":[213],"control,":[214],"acquisition":[216],"information":[218],"retrieval":[219],"potential":[221],"application":[222],"ITER.\\"":[224],"Fusion":[225],"Engineering":[226],"Design":[228],"84.12":[229],"(2009):":[230],"2136-2144.":[231],"[2]:":[232],"TokTagger.":[233],"https://github.com/ukaea/toktagger":[234],"[3]:":[235],"Pydantic.":[236],"https://github.com/pydantic/pydantic":[237],"[4]:":[238],"Database.":[241],"https://github.com/mongodb/mongo":[242],"[5]:":[243],"Ray.":[244],"https://github.com/ray-project/ray":[245]}