Apple PARS: Iyi ndiyo nzira iyo Apple inoda kuverenga yako uropi chiitiko kubva munzeve yako.

Pfungwa yekuti mahedhifoni anogona Teerera kumimhanzi yako uye, panguva imwe chete, "teerera" kuuropi hwako Zvinoita senge chimwe chinhu chiri kunze kwesainzi ngano, asi Apple yave kutogadzira nzira nemusanganiswa une simba: nyowani-munzeve sensors uye mhando dzepamberi dze. njere yekufungaSeri kwezvose izvi kune nzira inonzi PARS (Pairwise Relative Shift), nzira yekuzvidzidzira inobvumira algorithm kuti inzwisise kushanda kwemagetsi ehuropi pasina kuvimba nenyanzvi dzekurekodha data.

Panzvimbo pekutarisa pane chaiyo gadget, tsvakiridzo yeApple inotarisisa kuti sei modhi ye IA anogona dzidza chimiro chenguva cheEEG zviratidzo (electroencephalography) wobva washandisa ruzivo irworwo kumabasa akadai sekuronga nhanho dzekurara kana kuona kusagadzikana kwetsinga. Uye, kunyangwe chidzidzo chacho chisingatauri zvakananga AirPods, inowedzera kumatendi uye prototypes inonongedza kune ramangwana umo mahedhifoni akareruka anogona kuve rudzi rwe "mini-rabhoritari" yekutarisisa uropi kubva munzeve.

Chii chinonzi Apple PARS (Pairwise Relative Shift) uye nei yakakosha?

Iyo nzira yePARS yakabva mukutsvagisa yakapihwa nechikwata cheApple nevabatsiri vezvidzidzo, mubepa rakagamuchirwa pachikoro. NeurIPS 2025 Foundation Models yeBrain neMuviri musanganoChidzidzo ichi, chakanzi "Kudzidza hukama hwemasaini eEEG uchishandisa pairwise hama shift pretraining," inopa nzira yakasiyana yekudzidzisa mamodeli ane electroencephalography masaini pasina kushandisa mazita evanhu.

Mukuita, PARS inyanzvi ye kudzidzira kuzvitarisira inoshandiswa kuEEG. Panzvimbo pekubvunza vezveuropiyo kuti varatidze nemaoko kuti ndeipi chikamu chechiratidzo chinoenderana nechikamu chega chega chekurara kana kutanga kwekubatwa nepfari, modhi inodzidziswa nedata risingazivikanwe uye inomanikidzwa kugadzirisa dambudziko rekugadzira asi rinobatsira zvakanyanya: kufanotaura kuti chinhambwe chenguvai chinoparadzanisa zvimedu zvechiratidzo zviviri.

Pfungwa huru ndeyokuti kana modhi ikadzidza kufungidzira kuti inguva yakadii pakati peEEG windows, inozopedzisira yanzwisisa chimiro chepasi rose uye kutsamira kwenguva refu yekushanda kwehuropi. Izvi zvinobva zvaita kuti iite zvirinani mumabasa ekiriniki chaiwo, sekuona maitiro ekurara, kuona pfari, kana kuziva masaini emota.

Vanyori vanosimbisa kuti, mukupesana neyekare EEG nzira dzekuzvidzidzira, dzinonyanya kutarisisa gadzirazve zvikamu zvakafukidzwa zvechiratidzo (seyakavharidzirwa autoencoders, MAE, ita), PARS inotarisa pane hukama hwenguva pfupi. Ndokunge, haingozadzi "magapu" emunharaunda, asi inotora kuti zvidimbu zvakaparadzana zvechiratidzo zvinokwana sei nekufamba kwenguva.

Mune bvunzo dzakaitwa, iyo PARS-yakavakirwa modhi inoratidza kuti vanokwanisa kuenzanisa kana kudarika mazano ekare mune akati wandei EEG mabhenji, kunyanya kana mashoma mavara aripo (chiitiko chinowanzoitika mukurapa). Izvi zvinoita kuti PARS ive sarudzo inoyevedza kune chero sisitimu inoda kutora mukana wemavhoriyamu makuru emasaini ehuropi pasina kuvimba neanopedza zvirevo.

Mashandiro anoita PARS: kubva patokenization kuenda kuhama yekutama fungidziro

Kushandisa nzira yePARS kuEEG, vaongorori vanogadzira dhizaini yakavakirwa pa Transformers ine akati wandei matanho akakoshaIzvo zvese zvinotanga nepreprocessing chiratidzo uye kuchishandura kuita chinomiririra icho modhi inogona kubata nyore.

Kutanga, chiratidzo cheEEG chakakamurwa kuita mahwindo enguva pfupi kana "tokens"Iyi tokenization process inobvumira chidimbu chega chega chechiratidzo kuti chimiririrwe seyuniti iyo transformer inogona kushanda, senge mashandisirwo ezvinyorwa kana mifananidzo. Aya ma tokeni anobva awedzerwa... positional embeddings, kunyangwe neimwe nzira, nekuti PARS inotamba chaizvo nemasiki uye nekugadzirisa kweizvi zvinzvimbo zvenguva.

Chimwe chezvikamu zvakasiyana zveiyo nzira ndeye kushandiswa kwe masked positional embeddingPanzvimbo yekupa iyo modhi iyo chaiyo chinzvimbo ruzivo zvakananga mukati el tiempo Kune imwe neimwe tokeni, imwe positional data yakavanzwa kana kuchinjwa. Izvi zvinomanikidza encoder kuti tarisisa chimiro chenguva kubva pane zviri mukatipasina kuvimba chete nendekisi kana kuti chitambi chenguva chakajeka.

Mwoyo wePARS pre-workout ibasa re pairwise hama shift estimationIyo modhi inogamuchira mahwindo maviri eEEG akatorwa zvisina tsarukano kubva pakurekodha kumwechete uye anofanira kufanotaura kureba kwechinguvana pakati pavo. Haisi yekungofungidzira kuti vari padhuze here kana kure kure, asi nezvekudzidza inoenderera kana yakasarudzika mepu inoratidza hukama hwenguva.

Nokuda kweizvi, a decoder ine muchinjika-kutarisisa michinaIchi chikamu chinonongedza ruzivo rwakavharirwa kubva kumahwindo ese uye chinodzidza kuenderana nehunhu hwavo hwemukati kuti vatore nguva yekuvaparadzanisa. Nekuda kwemaitiro aya, shandurudzo inoguma yave kuenzanisa kutsamira kwenguva refu uye mapatani ehuropi chiitiko shanduko inoenda kure kupfuura nharaunda yenzvimbo yemamilliseconds mashoma.

Muzvikamu zvinozotevera, modhi inochinjirwa kumabasa akasiyana kuburikidza Multi-channel zvakanaka-tuning uye chaiyo ongororoIzvi zvinoreva kuti, kana yakambodzidziswa nePARS pane zvakasiyana-siyana zveEEG rekodhi (kusanganisira multi-electrode setups), inogadziridzwa zvakanaka kune mamwe mabasa akadai se. kupatsanurwa kwekurara, kuona EEG isina kujairika kana pfari pfari.

Chinyorwa chehunyanzvi zvakare chinodonongodza zvinhu zvinoshanda senge datasets anoshandiswa, iwo chaiwo mavakirwo eencoder, mhando yedecoder yakasarudzwa, mask uye zvigamba sampling zvirongwa., pamwe neiyo komputa zviwanikwa zvakashandiswaUye zvakare, akasiyana ezvivakwa akasiyana anofananidzwa neablation zvidzidzo kuti uone kuti ndedzipi sarudzo dzekugadzira dzinopa kuita kwakanyanya.

Kuenzanisa nedzimwe nzira: MAE, MP3, DropPos nevamwe

Chidzidzo chacho hachingotsananguri PARS, asi chinochisiyanisa nacho Reference nzira mukuzvidzidzira wega kweEEGPakati pematanho akaenzaniswa ndeaya masked autoencoders (MAE), MP3 uye DropPos, imwe neimwe iine huzivi hwakasiyana kana zvasvika pakudzidza kubva kune isina kunyorwa data.

MAEs anotarisa pa gadzirazve zvikamu zvakafukidzwa zvechiratidzoPanguva yekudzidziswa kusati kwaitwa, zvikamu zvekuisa zvakavanzwa, uye muenzaniso unoedza kuvadzosa kubva pane zviri mukati. Izvi zvinomanikidza mukodhuri kudzidza zvinomiririra zvine musoro, asi akanangana nemaitiro emunharaunda-kureva, kuzadza "magapu" ari pedyo pane kunzwisisa hukama hwenguva refu.

MP3 nedzimwe nzira dzakafanana dzinoongororawo nzira dzekufungidzira dzekutora ruzivo rwezvimiroNekudaro, zvinoenderana nemhedzisiro yebepa, ivo vachiri kushoma pane PARS pakuenzanisira nguva dzenguva pakati pezvikamu zviri kure zvechiratidzo.

DropPos, kune chikamu chayo, inogadzirisa kana kubvisa pachena ruzivo rwechinzvimbo mu transformer nepfungwa ye ita kuti muenzaniso uwedzere kusimba munzvimbo chaiyoKunyangwe iyi mhando yehunyanzvi ichibatsira network kuti isatsamira zvakanyanya pakumisikidzwa kwenzvimbo, zvakaratidzwa kuti, pachezvayo, hazvina kukwana kushandisa zvakanyanya chimiro chenguva chezviratidzo zveEEG.

Miedzo yekuedza inoratidza kuti mamodheru zvakafanodzidziswa nePARS zvakaenzana kana kudarika dzimwe nzira idzi mumatatu eEEG mabhenji akashandiswa. Kwainopenya zvakanyanya mune zvimiro zve label kunyatsoshandaNdiko, kana chikamu chiduku chezvakarekodhwa zviripo. Izvi zvakakosha muchirongwa chekiriniki, uko kunyora nemazvo maminetsi ese eEEG kunotora nguva uye kunoda nyanzvi dzakasarudzika.

Apendikisi yebasa inotsanangura zvakadzama the kumisikidzwa kweimwe neimwe yekutanga, iyo yakaedzwa hyperparameter, mhedzisiro yeakasiyana masiki mazinga kana akasiyana decoder architecture. uye yekupedzisira quantitative mhedzisiro. Mharidzo yakajeka ndeyekuti, yeEEG, kunyatsodzidza hukama hwenguva pakati pezvimedu zvechiratidzo kunopa mukana unoshanda pane kungovaka patsva kana masking.

Datasets inoshandiswa: kubva munzeve-EEG kurara kusvika pakuonekwa kwepfari

Kusimbisa PARS, Apple nevabatsiri vayo vakashandisa mana anozivikanwa EEG datasetsiyo inovhara akasiyana ekushandisa mamiriro: kurara, pathologies, mota chiitiko uye kunyange electrode masisitimu munzeve.

Yekutanga dataset ndeye Wearable Sleep Staging (EESM17), yakanangana nekutarisa kurara nemidziyo inopfeka. Inosanganisira marekodha ehusiku ezvidzidzo zve9 nehurongwa hwekutarisa kurara 12-channel ear-EEG uye 6-channel scalp EEGIyi dataset inonyanya kunakidza nekuti inoratidza kuti maelectrodes akaiswa munzeve anogona kutora chikamu chakakosha chebasa rehuropi zvinoenderana nekusiyanisa matanho ekurara.

Chechipiri ndechekuti TUAB (Temple University Abnormal EEG Corpus), korasi yakagadzirirwa iyo kuonekwa kweEEG isina kujairikaInounganidza marekodhi akanyorwa seakajairwa kana pathological, anobatsira pakudzidzisa mamodheru anoona general neurological alterations, kupfuura mamwe mamiriro.

Wechitatu, TUSZ (Temple University Seizure Corpus), yakanangana ne kuonekwa kwepfariInosanganisira zvirevo zvinomaka kutanga uye kupera kwekubatwa, pamwe nezvikamu zve interictal. Ndiyo imwe yemareferensi dataset mupfari yekuongorora AI algorithms.

Pakupedzisira, dataset yechina ndeye PhysioNet-MI, yakanangidzirwa pa mota fungidziro mabasaMuchiitiko ichi, vatori vechikamu vanofungidzira mafambiro (somuenzaniso, kufambisa ruoko) nepo EEG ichirekodhwa, ichibvumira kudzidziswa kwemhando dzinoziva mapatani anoenderana nechinangwa chekufamba, chimwe chinhu chakakosha muuropi-machine interfaces.

PARS pre-training inoitwa pane aya uye mamwe ma dataset anotsanangurwa mune tekinoroji appendix, nepo kunyatso-tuning kunorongedzerwa kune chaiwo mabasa mukati memumwe nemumwe wavo. Kusarudzwa kwemabhenji akasiyana-siyana anoratidza maitiro Izvo hazvina kuganhurirwa kune imwechete yekushandisa kesi uye izvo zvinogona kushanda sehwaro hwese hwekuzviongorora wega EEG modhi.

Ear-EEG uye AirPods: Tsvagiridzo yePARS inobatana sei nemahedhifoni eApple

Chimwe chinhu chinonyanya kushamisa cheiyi nyaya yese kushandiswa kwe nzeve-EEG, ndiko kuti, kutorwa kwezviratidzo zveuropi kubva munzeveIyo EESM17 dataset yatoshandisa masisitimu anoisa ma electrode munzeve canal uye nzeve, panzvimbo yemusoro wemusoro, iyo inoderedza zvakanyanya kutaridzika kwekuona uye inovandudza kunyaradza.

Zvichakadaro, magwaro eruzhinji uye Apple patents anoratidza kuti kambani yanga ichiongorora kwenguva yakati mahedhifoni anokwanisa kuyera biosignals kubva munzeve. Mune 2023 patent application, kambani inotsanangura "chigadzirwa chemagetsi chinopfekeka" chakagadzirirwa kurekodha chiitiko chehuropi uchishandisa. ma electrodes ari mukati kana kumativi nzeve, seimwe nzira isingaonekwe kune classic scalp EEG systems.

Iyo patent pachayo inobvuma kuti yakajairwa nzeve-EEG mhinduro dzinowanzoda zvishandiso zvakagadzirirwa kumushandisi wega wega (yakarongeka-yakagadzirirwa kuhukuru uye chimiro chenzeve yako, nzeve yenzeve, nezvimwewo), iyo inodhura uye isingaiti. Uyezve, kunyange chigadziro-chakagadzirwa chinogona kurasikirwa neganda nekufamba kwenguva, kuderera kwechiratidzo chemhando.

Kugadzirisa matambudziko aya, Apple inopa mhinduro mugwaro iro zvichienderana nekuisa mamwe ma electrode kupfuura anonyatsodikanwa, akagoverwa munzeve yeseuye rega AI modhi ione kuti ndedzipi dzinopa kuverenga kwakanyanya chero nguva. Kuti uite izvi, metrics akadai se impedance, ruzha nhanho, ganda rekubata mhando, kana chinhambwe pakati peanoshanda uye mareferenzi emagetsi.

Kana ma metrics aya akaverengerwa, sisitimu inogovera zviyero zvakasiyana zve electrode imwe neimwe uye inosanganisa masaini ayo kuita imwechete yakagadziridzwa waveform. Iyo patent inotosanganisira masisitimu akareruka, sekudzvanya kana kudzvanya nzeve, kutanga kana kumisa kuyerwa, pamwe neakasiyana dhizaini uye kusanganisa kusiyanisa kunoita kuti Hardware.

Kubva padzidziso kuenda kuchigadzirwa: masensa muAirPods uye yemazuva ese yehuropi yekutarisa

Iko kusanganiswa kweiyi mutsara wekutsvagisa nePARS kunoita kuti zvive nyore kufungidzira AirPods ane masensor anokwanisa kuyera EEG kubva mugero renzeveMuchokwadi, pakatove nekufambira mberi munzira iyoyo: iyo AirPods Pro 3 yakabatanidza photoplethysmograph (PPG) sensor kuyera kurova kwemoyo, uye Apple yanga ichiwedzera hutano kune ayo anopfeka mumakore mashoma apfuura.

Kana tikawedzera kune iyi maelectrodes enzeve-EEG uye yekuzviongorora modhi sePARS inokwanisa kududzira zviratidzo pasina chikuru dhatabhesi notedMhedzisiro yacho ingave mudziyo unogona kuona nhanho dzekurara, shanduko mukutarisisa, kana zviratidzo zvekutanga zveimwe tsinga pathologies, zvese pachena kumushandisi.

Mune zviedzo zvinotsanangurwa mubepa rekutsvagisa, iyo PARS algorithm inotora zvikamu zvisina tsarukano zvechiratidzo cheuropi uye dzidza kufanotaura chinhambwe chenguva pakati pazvoZvichienderana nekugona uku, modhi iyi inovandudza kunzwisisa kwekushanduka kwehuropi nekufamba kwenguva, zvichikonzera mhedzisiro iri nani pakuronga matanho ekurara, kutsvaga zviitiko zvepfari, kana kusiyanisa zvakajairwa kubva kune zvisina kujairika EEGs.

Chikumbiro chikuru cheiyi nzira ndechekuti inogona kushanda mumamiriro ezvinhu apo mavara ari mashoma. Muchigadzirwa chekutengesa, izvo zvinogona kureva kuti AirPods ane tekinoroji iyi anogona enderana nemushandisi wega wega ane ruzivo rudiki rwakatariswa, kutora mukana wehuwandu hukuru hwechiratidzo chakabatwa panguva yekushandiswa kwezuva nezuva.

Zvinoenderana nemhedzisiro yekudzidza, modhi dzakafanodzidziswa nePARS dzinosvika kuenzanisa kana kudarika kurongeka kwenzira dzakapfuura mumabasa akasiyana siyana, uye izvi zvinovhura musuwo wezvishandiso zvine mukurumbira zvevatengi, zvakaita semahedhifoni, kuti vatange kupa zviyero izvo zvaimbochengeterwa midziyo yechipatara yakakura uye yakakosha.

Ehe, zvese izvi zvinouya nemibvunzo inonzwisisika nezve data kuvanzika uye chengetedzo nemiganhu yetsika. Iyo pfungwa yekuti AI yaigona kuziva kwete chete kurova kwemoyo wako kana nhanho dzako dzezuva nezuva, asiwo maitiro ako ehuropi maitiro, inoburitsa imwe huwandu hwekunetseka. Parizvino, zvese zviri zviviri bepa uye patents zvinoramba zviri munzvimbo yekutsvagisa uye conceptual dhizaini, pasina zuva chairo rechigadzirwa chekutengesa.

Zvinogona kushandisa: hutano, hutano, kutyaira, uye kugona kwekuziva

Kana PARS uye nzeve-EEG-yakavakirwa tekinoroji ikapfeka mune ramangwana AirPods kana zvimwe zvinopfekwa, huwandu hwemashandisirwo hunogona kunge hwakapamhama. Chekutanga, yaizove yakanakira tarisa kurara nguva dzose uye zvakanaka, kurongedza otomatiki zvikamu zveREM neNREM (NREM 1, NREM 2 neNREM 3) uye nekupa ruzivo rwakadzama nezvemhando yekuzorora.

Uyezve, kuonekwa kwe mazinga ekutarisisa, zvikamu zvekushushikana, kana mamiriro ekuchenjerera Izvi zvingabatsira zvakanyanya mumamiriro ezvinhu akadai sekutyaira, basa rinonetsa mupfungwa, kana kudzidza. Mudziyo unoratidza kudonha kwakaerekana kwekutarisa unogona kunyevera mushandisi paanenge ari panjodzi yekubatwa nehope ari pavhiri kana kuita zvikanganiso zvakakomba nekuda kwekuneta.

Munzvimbo yekiriniki, kungwara uye kuenderera mberi kwehuropi kunogona kufambisa kukurumidza kuona kusagadzikana kwakadai sepfari, matambudziko ekurara, kana neurodegenerative zvirwereHaisi yekutsiva chiremba wetsinga kana chipatara, asi nezvekupa data rakakosha rinogona kushanda seyambiro yekukurumidza kana kuzadzisa kuongororwa.

Imwe mutsara unonakidza ndeye biofeedback yakanangana nekugara zvakanaka kwepfungwaKana mudziyo uchikwanisa kuwiriranisa mamwe maitiro eEEG kumatunhu ekuzorora, kudzika midzi, kana kushushikana mupfungwa, unogona kutungamira mushandisi mukudzidzira kufema, kufungisisa, kana kudzidziswa kwekuziva, zvichipa zviratidzo zvenguva-chaiyo zvekuti maitiro aya ari kuita zvaanoda.

Zvaisazomira pauropi chete. Mapepa eruzhinji eApple anotaura nezvekuwedzera kwe masensa ekuyera huwandu hweropa, kuita kwemhasuru dzechiso, uye kufamba kwemaziso kubva paearphone pachayo. Yakasanganiswa neiyo EEG chiratidzo uye yakagadziriswa mu a iPhone kana imwe mudziyo, iyi data inogona kudyisa AI modhi inokwanisa kupa mufananidzo wakazara wemushandisi physiological uye manzwiro mamiriro.

Zvose izvi zvinoshandiswa zvinofanirwa kuperekedzwa ne kudzora kwakasimba pamusoro pemvumo, manejimendi data uye kuwana nevechitatu mapatoMumienzaniso yakaratidzwa, zvinofungidzirwa kuti ruzivo runogona kugovaniswa nevashandi vehutano chete kana mushandisi akarubvumidza, uye kuti kwakawanda kwekugadziriswa kunoitwa munharaunda kuderedza njodzi.

Mukuita, izvo zvese zviri zviviri bepa rePARS uye patents uye prototypes zvinoratidzira kuchinjika kwakajeka: Apple iri kuongorora, kune rumwe rutivi, nzira yakanakisa yekuunganidza zviratidzo zvehuropi kubva munzeve uye, kune imwe, nzira yakanakisa yekuvadudzira neAI pasina kuvimba neanonongedzerwa nevanhuKana zvese zvidimbu zvikabatana, mahedhifoni anogona kumira kuve akareruka ekuridza odhiyo uye ave epamusoro hutano uye maturusi ekuita, chero ekuti tsika nekuvanzika zvinhu zvinyatsogadziriswa.

Zvese zvinonongedza kune chokwadi chekuti isu tiri pakutanga kwechizvarwa chitsva chekupfeka umo nzira dzakadai seApple PARS (Parwise Relative Shift) uye discrete sensors senge nzeve-EEG Vanogona kushandura manzwisisiro atinoita kurara kwedu, kutarisisa kwedu, uye hutano hwedu hwetsinga, kutanga nechimwe chinhu semazuva ese seAirPods.