InoteveraGaidhi rinotevera
Empirical Risk Minimization
Basics
Nhungamiro yehunyanzvi
Sharpness-Aware Minimization
Sharpness-Aware Minimization (SAM) inzira yekusimudzira isingatsvage kurasikirwa kwakaderera asi kurasikirwa kwakaderera munzvimbo yese yehuremu - hushoma hwakadzika.
Teerera kugwaro iriAI-yakarondedzerwa · ~2 min
Pa peji ino2 min verenga
Pfupiso
Flatter minima tend to generalize better, so SAM often improves test accuracy and robustness without changing the model architecture.
Kudzika Kwakadzika
Kudzidziswa kwakajairwa kunoderedza kurasikirwa panzvimbo imwe chete munzvimbo yehuremu, asi mhinduro mbiri dzine kurasikirwa kwekudzidziswa kwakafanana dzinogona kuita zvakasiyana: 'yakapinza' hushoma hunogara mumupata wakamanikana umo huturu hwehuremu hunowedzera kurasikirwa, nepo 'flat' shoma inobvumira kukanganisa uye kazhinji inogadzirisa zvirinani kune data risingaonekwe. SAM, yakaunzwa neGoogle vatsvakurudzi muna 2020, inoita kuti izvi zvive pachena. Panhanho imwe neimwe inotanga yawana huremu hwepedyo (mukati meradius rho) iyo inowedzera kurasikirwa - muvakidzani wakaipisisa - yobva yagadzirisa huremu hwepakutanga kuderedza kurasikirwa panzvimbo iyoyo. Iyi min-max chinangwa inosundira optimization kumatunhu akadzikira zvakafanana, achipa zvinooneka zvirinani kurongwa kwemifananidzo uye kupfuura.
Kuitwa muNyika yeChokwadi
Kuwedzera Vision Transformer uye ResNet kurongeka paImageNet nekudzidziswa neSAM panzvimbo yeSGD yakajeka.
Kuvandudza kusimba kwekuisa ruzha, sezvo flat minima isingawanzo kubata nemusoro mavara akaora.
Kunyatsogadzirisa mamodheru emitauro yakadzidziswa neSAM kuti uwane kujekeswa kuri nani pamaseti madiki ezasi.
Kushandisa ESAM kana LookSAM akasiyana kana iyo yakapetwa kaviri compute mutengo wevanilla SAM uchidhura zvakanyanya.
Ramangwana reKupinza-Kuziva Kuderedza
SAM yakabereka mhuri yekutevera yakanangana nekushaya simba kwayo kukuru, iyo yakapetwa kaviri komputa: yakanyatso siyana seESAM, LookSAM, uye nzira dzinokanganisa chete subset yehuremu kana kushandisa SAM matanho mashoma ega ega. Adaptive SAM (ASAM) inogadzirisa radius kuti ive chikero-chisingachinji. Vatsvaguri vanoramba vachikakavadzana chaizvo kuti sei kupfava kunobatsira uye nekuiyera sei, uye kupinza-kuziva mazano ari kupararira kune kunyatso-tuta mikuru mamodheru emitauro uye nekuvandudza kusimba kune yekugovera shanduko.
Kuzvishandisa
Nei zvichikosha
- Mutengo uye bhajeti. Zvisarudzo zvezvivakwa zvinotyaira kuita uye mutengo wekushandisa kwemakore.
- Sarudzo dzakajeka. Dzidzo yehunyanzvi inobatsira zvikwata kusarudza murwi wakakodzera, kwete iwo mutsva chete.
- Kudzora kwemhando yepamusoro. Sarudzo dzeinjiniya dziri nani dzinoderedza zviitiko zvekuvimbika mukugadzira.
Chenjera
- Kugadzirisa imwe bhenji kunogona kuvanza yakafara system kushaya simba.
- Infrastructure uye mari yekugadzirisa inowanzotarisirwa pasi.
- Chengetedzo uye kucherechedzwa mapundu anogona kukura sezvo masisitimu anowedzera kuoma.
Matanho anotevera
- Tsanangura latency, mhando, uye mutengo zvinangwa usati waitwa.
- Benchmark pasi pechokwadi mutoro uye data mamiriro.
- Chishandiso chekutarisa zvikanganiso, kudonha, uye mushandisi maitiro.
- Gadzirira nzira dzekudzosera kumashure uye dzezviitiko usati wawedzera.
Ramba Uchiongorora
Tsamba yemahara
Tora iyo yemazuva ese AI pfupiso
Nhatu dzakasimbiswa dzeAI nyaya mangwanani ega ega pakati pevhiki, dzakanyorwa neChirungu chakajeka. Yemahara zvachose, hapana ads.
Imwe email zuva rega rega revhiki. Bvisa kunyoresa kamwe chete. Hatimbotengese kana kugovera kero yako.
Zviedze iwe pachako
Take the Sharpness-Aware Minimization quiz
Mhinduro yepakarepo pamhinduro yega yega, uye chitupa chinogovaniswa chine ID inobatika kana wapedza kosi.
Tsigira dzidzo yemahara yeAI. AI Understanding ndeye 501(c)(3) isingabatsiri - hapana ads, hapana paywall, zvachose. Ita chipo
Mibvunzo inowanzo bvunzwa
What is Sharpness-Aware Minimization?
Sharpness-Aware Minimization (SAM) inzira yekusimudzira isingatsvage kurasikirwa kwakaderera asi kurasikirwa kwakaderera munzvimbo yese yehuremu - hushoma hwakadzika. Flatter minima inowanzoita generalize zvirinani, saka SAM inowanzovandudza bvunzo kurongeka uye kusimba pasina kushandura modhi yekuvaka.
Ndeupi rudzi rwehushoma hunoyedzwa neSAM kuwana?
SAM inonangidzira flat minima nekuti kurasikirwa kunoramba kwakadzikira pasi pehuremu hudiki huremu hunowanzo wedzera zvirinani kune data risingaonekwe.
Mangani ekumberi-kumashure anopfuura anodiwa nhanho yeSAM yakajairwa?
SAM inoverengera gradient kuti iwane yakaipisisa-kesi iri padyo poindi, yozoita imwe gradient ipapo yekuvandudza - inopeta kaviri mutengo wakajairika.
Chii chinonzi radius hyperparameter rho control muSAM?
Rho inoisa kuti nhanho 'yekukwira' inosvika papi kuti uwane muvakidzani akashata, achitsanangura kuti yakakura sei nzvimbo yakati sandara inotsvakwa neSAM.
Ndeipi yekutanga nhanho mbiri dzeSAM pakudzokorora kwega kwega?
SAM inotanga kuvhiringa huremu mukati meradius rho munzira iyo inowedzera kurasikirwa, yobva yadzika kubva panzvimbo yekutanga ichishandisa gradient yakaverengerwa ipapo.
Sei SAM ichiwanzovandudza kusimba kuisa ruzha?
Kubata nemusoro mavara ane ruzha kazhinji kunoda akapinza, nhete minima; nekufarira nzvimbo dzakati sandara, SAM inoramba kuwandisa uku.
Ramba uchidzidza
Mazano anoenderana
Mamwe madhairekitori akasarudzirwa nyaya iyi