adapt transfer learn discrimin learn method wide natur languag process method work train test data drawn distribut nlp task confront domain label data scarc exist case seek adapt exist model resourcerich sourc domain resourc poor target domain introduc structur correspond learn automat induc correspond featur domain test techniqu part speech tag show perform gain vari amount sourc target train data improv target domain pars accuraci improv tagger cross ddomain sentiment classif spectral featur align survei transfer learn major assumpt machin learn data mine algorithm train futur data featur space distribut real world applic assumpt hold classif task domain interest suffici train data domain interest data featur space follow data distribut case knowledg transfer successfulli greatli improv perform learn avoid expens data label effort recent year transfer learn emerg learn framework address problem survei focus categor review current progress transfer learn classif regress cluster problem survei discuss relationship transfer learn relat machin learn techniqu domain adapt multitask learn sampl select bia variat shift explor potenti futur issu transfer learn research knowledg transfer heterogen inform network knowledg transfer heterogen inform network transfer local model space machin learn approach indoor wifi local involv offlin phase onlin phase offlin phase data collect environ build local model appli data collect onlin phase locat estim collect label data entir build time consum paper present approach transfer learn model train data area build learn map function signal space locat space solv optim problem base manifold learn techniqu low dimension manifold share data collect area environ bridg propag knowledg environ transfer knowledg significantli reduc amount label data requir build local model test effect propos solut real indoor wifi environ transfer multi ddevic local model latent multi dtask learn paper propos latent multi task learn algorithm solv multi devic indoor local problem tradit indoor local system assum collect signal data distribut fix local model learn devic devic adapt empir studi signal variat devic found assumpt invalid practic solv problem treat multipl devic multipl learn task propos multi task learn algorithm algorithm assum hypothes learn origin data space relat task similar requir hypothes learn latent featur space similar establish algorithm emploi altern optim approach iter learn featur map multi task regress model devic appli latent multi task learn algorithm real world indoor local data demonstr effect transfer learn dimension reduct transfer learn address problem util plenti label data sourc domain solv relat problem target domain train test problem distribut featur paper transfer learn dimension reduct solv problem learn low dimension latent featur space distribut sourc domain data target domain data close latent featur space project data relat domain appli standard learn algorithm train classif regress model latent featur space treat bridg transfer knowledg sourc domain target domain main contribut work propos dimension reduct method find latent space minim distanc distribut data domain latent space effect approach transfer learn verifi experi real world applic indoor wifi local binari text classif digit wall power deffici solut locat dbase data share estim locat wi dfi real world activ recognit multipl goal recogn understand activ peopl sensor read import task ubiquit comput activ recognit difficult task inher uncertainti complex data collect sensor research tackl problem overli simplist set assum user carri singl activ time multipl activ consecut formal explor degre human perform concurr interleav activ studi detect multipl goal real world scenario articl fundament question user carri multipl concurr interleav activ singl activ daili life complex behavior detect accur sensor defin class complex level goal taxonomi describ granular activ relat recognit accuraci complex level granular present theoret framework recogn multipl concurr interleav activ evalu framework real world ubiquit comput environ estim locat wi dfi onlin dlocal indoor wireless network dimens reduct adapt local dynam wifi environ multi dview learn accur locat user wireless environ import task pervas comput ai applic activ recognit wifi environ mobil devic local signal receiv transmitt access point ap local approach build map signal space physic locat space offlin phase receiv signal strength rss map estim locat onlin phase map outdat signal strength valu chang time due environment dynam infeas expens repeat data calibr reconstruct rss map case import adapt model learnt time period time period recalibr paper present locat estim approach base manifold regular machin learn techniqu build map function data describ lemancor system adapt map function signal space physic locat space time period base manifold regular show lemancor effect transfer knowledg time period requir calibr effort illustr leman cor effect real wifi environ domain dconstrain semi dsupervis mine track model sensor network mine compet case base case dbase reason preprocess time seri data classif applic crm 