﻿68:@0.095238:0.964164:0.121391:0.964164:0.121391:0.942953:0.095238:0.942953:0.000000:0.000000
ID 3/2024:@0.837225:0.958865:0.904081:0.958865:0.904081:0.945229:0.837225:0.945229:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000
L’ARCHITETTURA TRANSFORMER:@0.095238:0.160465:0.345206:0.160465:0.345206:0.144261:0.095238:0.144261:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000
Con l’introduzione dell’architettura :@0.095238:0.176234:0.369279:0.176234:0.369279:0.161083:0.095238:0.161083:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000
Transformer:@0.380517:0.176234:0.464880:0.176234:0.464880:0.161059:0.380517:0.161059:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000
 (modelli implementati per la :@0.464880:0.176234:0.718880:0.176234:0.718880:0.161083:0.464880:0.161083:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000
comprensione di sequenze di dati, testo o serie temporali), gli LLMs hanno segnato :@0.095238:0.192326:0.718831:0.192326:0.718831:0.177175:0.095238:0.177175:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000
un’evoluzione rapida, evidenziata da modelli rivoluzionari quali GPT e BERT, che con :@0.095238:0.208418:0.718855:0.208418:0.718855:0.193267:0.095238:0.193267:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000
miliardi  di  parametri,  possiedono  notevoli  capacità  di  comprensione,  traduzione  e :@0.095238:0.224509:0.718856:0.224509:0.718856:0.209358:0.095238:0.209358:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000
generazione testuale a fronte di un limitato addestramento.:@0.095238:0.240601:0.523100:0.240601:0.523100:0.225450:0.095238:0.225450:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000
Parallelamente, i LVMs hanno trasformato la visione computazionale, superando i limiti :@0.095238:0.256692:0.718913:0.256692:0.718913:0.241542:0.095238:0.241542:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000
delle prestazioni in vari compiti di riconoscimento visivo utilizzando i :@0.095238:0.272784:0.577935:0.272784:0.577935:0.257633:0.095238:0.257633:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000
Vision Transformers :@0.576712:0.272784:0.718855:0.272784:0.718855:0.257608:0.576712:0.257608:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000
(ViTs) e i :@0.095238:0.288876:0.160334:0.288876:0.160334:0.273725:0.095238:0.273725:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000
Convolutional Neural Networks:@0.160334:0.288876:0.383991:0.288876:0.383991:0.273700:0.160334:0.273700:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000
 (CNNs).:@0.383991:0.288876:0.445140:0.288876:0.445140:0.273725:0.383991:0.273725:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000
I LMMs rappresentano la frontiera :@0.095238:0.530250:0.342468:0.530250:0.342468:0.515099:0.095238:0.515099:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000
nell’integrazione di dati visivi :@0.095238:0.546342:0.342483:0.546342:0.342483:0.531191:0.095238:0.531191:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000
e linguistici, promuovendo una :@0.095238:0.562433:0.342483:0.562433:0.342483:0.547282:0.095238:0.547282:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000
comprensione:@0.095238:0.578525:0.194839:0.578525:0.194839:0.563374:0.095238:0.563374:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000
e:@0.218441:0.578525:0.227143:0.578525:0.227143:0.563374:0.218441:0.563374:0.000000
generazione:@0.250746:0.578525:0.337897:0.578525:0.337897:0.563374:0.250746:0.563374:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000
multimodale  che  apre  nuove :@0.095238:0.594616:0.342473:0.594616:0.342473:0.579465:0.095238:0.579465:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000
possibilità:@0.095238:0.610708:0.167977:0.610708:0.167977:0.595557:0.095238:0.595557:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000
in:@0.192655:0.610708:0.205120:0.610708:0.205120:0.595557:0.192655:0.595557:0.000000:0.000000
ambiti:@0.229797:0.610708:0.274567:0.610708:0.274567:0.595557:0.229797:0.595557:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000
come:@0.299244:0.610708:0.337915:0.610708:0.337915:0.595557:0.299244:0.595557:0.000000:0.000000:0.000000:0.000000
la creazione di immagini da :@0.095238:0.626800:0.342468:0.626800:0.342468:0.611649:0.095238:0.611649:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000
testi. Sfruttando l’architettura :@0.095238:0.642891:0.342485:0.642891:0.342485:0.627740:0.095238:0.627740:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000
Transformer:@0.095238:0.658983:0.179602:0.658983:0.179602:0.643807:0.095238:0.643807:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000
,:@0.179602:0.658983:0.183281:0.658983:0.183281:0.643832:0.179602:0.643832:0.000000
questi:@0.212679:0.658983:0.256945:0.658983:0.256945:0.643832:0.212679:0.643832:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000
modelli:@0.286343:0.658983:0.337899:0.658983:0.337899:0.643832:0.286343:0.643832:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000
offrono rappresentazioni unificate :@0.095238:0.675074:0.342485:0.675074:0.342485:0.659924:0.095238:0.659924:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000
applicabili a una vasta gamma di :@0.095238:0.691166:0.342501:0.691166:0.342501:0.676015:0.095238:0.676015:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000
compiti.:@0.095238:0.707258:0.151884:0.707258:0.151884:0.692107:0.095238:0.692107:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000
Nel  panorama  attuale  delle :@0.095238:0.723349:0.342473:0.723349:0.342473:0.708198:0.095238:0.708198:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000
biotecnologie  e  della  salute :@0.095238:0.739441:0.342473:0.739441:0.342473:0.724290:0.095238:0.724290:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000
digitale, i modelli di intelligenza :@0.095238:0.755533:0.342501:0.755533:0.342501:0.740382:0.095238:0.740382:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000
artificiale  di  grandi  dimensioni :@0.095238:0.771624:0.342475:0.771624:0.342475:0.756473:0.095238:0.756473:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000
sono:@0.095238:0.787716:0.129457:0.787716:0.129457:0.772565:0.095238:0.772565:0.000000:0.000000:0.000000:0.000000
potenti:@0.150926:0.787716:0.203020:0.787716:0.203020:0.772565:0.150926:0.772565:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000
strumenti:@0.224489:0.787716:0.292961:0.787716:0.292961:0.772565:0.224489:0.772565:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000
per:@0.314430:0.787716:0.337915:0.787716:0.337915:0.772565:0.314430:0.772565:0.000000:0.000000:0.000000
affrontare  e  risolvere  complesse :@0.095238:0.803807:0.342473:0.803807:0.342473:0.788657:0.095238:0.788657:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000
sfide in vari settori critici, dalla :@0.095238:0.819899:0.342468:0.819899:0.342468:0.804748:0.095238:0.804748:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000
bioinformatica:@0.095238:0.835991:0.197846:0.835991:0.197846:0.820840:0.095238:0.820840:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000
alla:@0.225833:0.835991:0.250460:0.835991:0.250460:0.820840:0.225833:0.820840:0.000000:0.000000:0.000000:0.000000
robotica:@0.278447:0.835991:0.337915:0.835991:0.337915:0.820840:0.278447:0.820840:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000
medica:@0.095238:0.852082:0.146676:0.852082:0.146676:0.836931:0.095238:0.836931:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000
biomedica,:@0.169506:0.852082:0.246916:0.852082:0.246916:0.836931:0.169506:0.836931:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000
rendendo:@0.269745:0.852082:0.337899:0.852082:0.337899:0.836931:0.269745:0.836931:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000
agevoli, semplici e diffuse attività :@0.095238:0.868174:0.342493:0.868174:0.342493:0.853023:0.095238:0.853023:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000
che fino a pochi anni fa erano di :@0.095238:0.884265:0.342510:0.884265:0.342510:0.869115:0.095238:0.869115:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000
dominio di pochi e selezionati :@0.095238:0.900357:0.342501:0.900357:0.342501:0.885206:0.095238:0.885206:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000
professionisti.:@0.095238:0.916449:0.194201:0.916449:0.194201:0.901298:0.095238:0.901298:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000
Ferdinando SPAGNOLO:@0.095238:0.081672:0.248846:0.081672:0.248846:0.068036:0.095238:0.068036:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000
Schema di un :@0.761904:0.411064:0.854285:0.411064:0.854285:0.398220:0.761904:0.398220:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000
Convolutional Neural :@0.761904:0.422947:0.903229:0.422947:0.903229:0.410103:0.761904:0.410103:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000
Networks:@0.761904:0.434830:0.823980:0.434830:0.823980:0.421986:0.761904:0.421986:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000
their societal impact, including the deskilling of professionals like doctors [ ]. Therefore,:@0.102984:0.006419:0.659352:0.006419:0.659352:-0.006282:0.102984:-0.006282:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000
41:@0.569155:0.006419:0.583215:0.006419:0.583215:-0.006282:0.569155:-0.006282:0.000000:0.000000
while the application of machine learning techniques in healthcare is inevitable, establish-:@0.102331:0.019472:0.659985:0.019472:0.659985:0.006771:0.102331:0.006771:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000
ing standardized criteria for interpretable ML in this field is urgently needed to enhance:@0.102984:0.032525:0.657415:0.032525:0.657415:0.019824:0.102984:0.019824:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000
transparency, fairness, and safety.:@0.102984:0.045578:0.310821:0.045578:0.310821:0.032877:0.102984:0.032877:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000
3. Primaries on Vision Transformer:@0.102984:0.069073:0.327608:0.069073:0.327608:0.056269:0.102984:0.056269:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000
Vision Transformer (ViT) [ ] is a deep learning model that has gained significant:@0.132985:0.085048:0.657421:0.085048:0.657421:0.072347:0.132985:0.072347:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000
17:@0.301119:0.085048:0.315178:0.085048:0.315178:0.072347:0.301119:0.072347:0.000000:0.000000
attention in computer vision. In contrast to traditional convolutional neural networks:@0.102990:0.098101:0.657421:0.098101:0.657421:0.085400:0.102990:0.085400:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000
(CNNs), which have been the dominant architecture for image recognition tasks, ViT:@0.102479:0.111154:0.657904:0.111154:0.657904:0.098453:0.102479:0.098453:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000
adopts a transformer-based architecture inspired by its success in natural language process-:@0.102984:0.124207:0.659992:0.124207:0.659992:0.111506:0.102984:0.111506:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000
ing (NLP) [:@0.102984:0.137260:0.172928:0.137260:0.172928:0.124559:0.102984:0.124559:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000
42:@0.172928:0.137260:0.186988:0.137260:0.186988:0.124559:0.172928:0.124559:0.000000:0.000000
]. ViT breaks down an image into fixed-size patches, which are then linearly:@0.186988:0.137260:0.658027:0.137260:0.658027:0.124559:0.186988:0.124559:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000
embedded and processed using a transformer encoder to capture global contextual infor-:@0.102984:0.150313:0.659985:0.150313:0.659985:0.137612:0.102984:0.137612:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000
mation. This approach allows ViT to handle local and global image features effectively,:@0.102984:0.163365:0.659360:0.163365:0.659360:0.150664:0.102984:0.150664:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000
leading to remarkable performance in various computer vision tasks, including image:@0.102984:0.176418:0.657415:0.176418:0.657415:0.163717:0.102984:0.163717:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000
classification and object detection.:@0.102984:0.189471:0.314251:0.189471:0.314251:0.176770:0.102984:0.176770:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000
Generally, a Vision Transformer consists of a patch embedding layer and several:@0.132985:0.202524:0.657412:0.202524:0.657412:0.189823:0.132985:0.189823:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000
consecutively connected encoders, as depicted in Figure . The self-attention layer is:@0.102980:0.215577:0.657412:0.215577:0.657412:0.202876:0.102980:0.202876:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000
1:@0.471877:0.215577:0.478977:0.215577:0.478977:0.202876:0.471877:0.202876:0.000000
the key component that enables ViT to achieve many state-of-the-art vision recognition:@0.102980:0.228630:0.657413:0.228630:0.657413:0.215929:0.102980:0.215929:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000
performances. The self-attention layer first transforms the input image into three different:@0.102520:0.241683:0.657421:0.241683:0.657421:0.228982:0.102520:0.228982:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000
vectors—the query vector, the key vector, and the value vector. Subsequently, the attention:@0.102548:0.254736:0.657414:0.254736:0.657414:0.242035:0.102548:0.242035:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000
layer then computes the scores between each pair of vectors and determines the degree of:@0.102984:0.267789:0.657408:0.267789:0.657408:0.255088:0.102984:0.255088:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000
attention when given other tokens.:@0.102984:0.280842:0.319481:0.280842:0.319481:0.268141:0.102984:0.268141:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000
Figure 1.:@0.102984:0.522815:0.152902:0.522815:0.152902:0.511291:0.102984:0.511291:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000
The basic framework of Vision Transformer (ViT) [:@0.156824:0.522647:0.438516:0.522647:0.438516:0.511216:0.156824:0.511216:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000
17:@0.438516:0.522647:0.451169:0.522647:0.451169:0.511216:0.438516:0.511216:0.000000:0.000000
] and its encoder architecture.:@0.451169:0.522647:0.616689:0.522647:0.616689:0.511216:0.451169:0.511216:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000
Formally, given an image:@0.132985:0.546208:0.294678:0.546208:0.294678:0.533508:0.132985:0.533508:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000
x:@0.299165:0.546436:0.306265:0.546436:0.306265:0.533528:0.299165:0.533528:0.000000
∈:@0.312146:0.550644:0.321903:0.550644:0.321903:0.538455:0.312146:0.538455:0.000000
R:@0.327778:0.547810:0.338355:0.547810:0.338355:0.533346:0.327778:0.533346:0.000000
H W:@0.338904:0.541918:0.366567:0.541918:0.366567:0.532108:0.338904:0.532108:0.000000:0.000000:0.000000
× ×:@0.347772:0.545115:0.375977:0.545115:0.375977:0.535852:0.347772:0.535852:0.000000:0.000000:0.000000
3:@0.376113:0.541745:0.381456:0.541745:0.381456:0.532092:0.376113:0.532092:0.000000
, the patch embedding layer first splits and:@0.382159:0.546208:0.657422:0.546208:0.657422:0.533508:0.382159:0.533508:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000
flattens the sample:@0.102991:0.561033:0.221318:0.561033:0.221318:0.548332:0.102991:0.548332:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000
x:@0.225038:0.561261:0.232138:0.561261:0.232138:0.548353:0.225038:0.548353:0.000000
into sequential patches:@0.235858:0.561033:0.380114:0.561033:0.380114:0.548332:0.235858:0.548332:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000
x:@0.383834:0.561261:0.390934:0.561261:0.390934:0.548353:0.383834:0.548353:0.000000
p:@0.391632:0.562053:0.396975:0.562053:0.396975:0.552243:0.391632:0.552243:0.000000
∈:@0.402263:0.565465:0.412020:0.565465:0.412020:0.553277:0.402263:0.553277:0.000000
R:@0.416473:0.562632:0.427050:0.562632:0.427050:0.548167:0.416473:0.548167:0.000000
N:@0.427558:0.556739:0.435871:0.556739:0.435871:0.546929:0.427558:0.546929:0.000000
×:@0.436470:0.559936:0.445121:0.559936:0.445121:0.550673:0.436470:0.550673:0.000000
(:@0.445389:0.556388:0.449708:0.556388:0.449708:0.548184:0.445389:0.548184:0.000000
P d:@0.450188:0.556739:0.467248:0.556739:0.467248:0.546929:0.450188:0.546929:0.000000:0.000000:0.000000
2:@0.456851:0.553239:0.461068:0.553239:0.461068:0.545618:0.456851:0.545618:0.000000
):@0.467515:0.556388:0.471835:0.556388:0.471835:0.548184:0.467515:0.548184:0.000000
, where:@0.472669:0.561030:0.519045:0.561030:0.519045:0.548329:0.472669:0.548329:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000
(:@0.522939:0.560797:0.528623:0.560797:0.528623:0.550002:0.522939:0.550002:0.000000
H W:@0.528795:0.561258:0.559177:0.561258:0.559177:0.548350:0.528795:0.548350:0.000000:0.000000:0.000000
,:@0.540043:0.561030:0.543557:0.561030:0.543557:0.548329:0.540043:0.548329:0.000000
):@0.559853:0.560797:0.565537:0.560797:0.565537:0.550002:0.559853:0.550002:0.000000
represents the:@0.569424:0.561030:0.657420:0.561030:0.657420:0.548329:0.569424:0.548329:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000
height:@0.102984:0.574311:0.138026:0.574311:0.138026:0.561403:0.102984:0.561403:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000
and:@0.141531:0.574083:0.165261:0.574083:0.165261:0.561382:0.141531:0.561382:0.000000:0.000000:0.000000
width:@0.168766:0.574311:0.201467:0.574311:0.201467:0.561403:0.168766:0.561403:0.000000:0.000000:0.000000:0.000000:0.000000
of the input image,:@0.204972:0.574083:0.322392:0.574083:0.322392:0.561382:0.204972:0.561382:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000
(:@0.326066:0.573850:0.331750:0.573850:0.331750:0.563056:0.326066:0.563056:0.000000
P P:@0.332379:0.574311:0.356039:0.574311:0.356039:0.561403:0.332379:0.561403:0.000000:0.000000:0.000000
,:@0.341145:0.574083:0.344649:0.574083:0.344649:0.561382:0.341145:0.561382:0.000000
):@0.356390:0.573850:0.362074:0.573850:0.362074:0.563056:0.356390:0.563056:0.000000
is the resolution of each image patch, denotes:@0.365748:0.574083:0.657414:0.574083:0.657414:0.561382:0.365748:0.561382:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000
d:@0.598585:0.574311:0.605593:0.574311:0.605593:0.561403:0.598585:0.561403:0.000000
the output channel, and:@0.102984:0.587136:0.255518:0.587136:0.255518:0.574435:0.102984:0.574435:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000
N:@0.260601:0.587364:0.271539:0.587364:0.271539:0.574456:0.260601:0.574456:0.000000
=:@0.277889:0.586903:0.289272:0.586903:0.289272:0.576109:0.277889:0.576109:0.000000
HW P:@0.295726:0.587364:0.339126:0.587364:0.339126:0.574456:0.295726:0.574456:0.000000:0.000000:0.000000:0.000000
/:@0.321384:0.587136:0.329904:0.587136:0.329904:0.574435:0.321384:0.574435:0.000000
2:@0.339303:0.582671:0.344646:0.582671:0.344646:0.573019:0.339303:0.573019:0.000000
is the number of image tokens. The list of patch:@0.349759:0.587136:0.657412:0.587136:0.657412:0.574435:0.349759:0.574435:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000
tokens is further fed into Transformer encoders for attention calculation.:@0.102984:0.600189:0.551107:0.600189:0.551107:0.587489:0.102984:0.587489:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000
Each Transformer encoder mainly consists of two types of sub-layers—a multi-head:@0.132985:0.613241:0.657415:0.613241:0.657415:0.600541:0.132985:0.600541:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000
self-attention layer (MSA) and an MLP layer. In MSA, the tokens are linearly projected and:@0.102984:0.626294:0.657417:0.626294:0.657417:0.613593:0.102984:0.613593:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000
further re-formulated into three vectors, namely:@0.102984:0.639347:0.397557:0.639347:0.397557:0.626646:0.102984:0.626646:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000
Q K:@0.401173:0.639523:0.430223:0.639523:0.430223:0.626730:0.401173:0.626730:0.000000:0.000000:0.000000
,:@0.413061:0.639347:0.416541:0.639347:0.416541:0.626647:0.413061:0.626647:0.000000
and:@0.434033:0.639347:0.457597:0.639347:0.457597:0.626647:0.434033:0.626647:0.000000:0.000000:0.000000
V:@0.460931:0.639523:0.470308:0.639523:0.470308:0.626730:0.460931:0.626730:0.000000
. The self-attention calculation:@0.471721:0.639347:0.657422:0.639347:0.657422:0.626647:0.471721:0.626647:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000
is performed on:@0.102984:0.652400:0.202312:0.652400:0.202312:0.639700:0.102984:0.639700:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000
Q K:@0.206136:0.652576:0.234212:0.652576:0.234212:0.639782:0.206136:0.639782:0.000000:0.000000:0.000000
,:@0.218016:0.652400:0.221531:0.652400:0.221531:0.639700:0.218016:0.639700:0.000000
and:@0.238233:0.652400:0.262035:0.652400:0.262035:0.639700:0.238233:0.639700:0.000000:0.000000:0.000000
V:@0.265578:0.652576:0.274956:0.652576:0.274956:0.639782:0.265578:0.639782:0.000000
by:@0.279891:0.652400:0.295482:0.652400:0.295482:0.639700:0.279891:0.639700:0.000000:0.000000
x:@0.229207:0.683934:0.236237:0.683934:0.236237:0.671026:0.229207:0.671026:0.000000
′:@0.236706:0.682091:0.239768:0.682091:0.239768:0.672827:0.236706:0.672827:0.000000
ℓ:@0.236562:0.685934:0.241193:0.685934:0.241193:0.677731:0.236562:0.677731:0.000000
=:@0.246111:0.683468:0.257493:0.683468:0.257493:0.672674:0.246111:0.672674:0.000000
Attention:@0.261579:0.683701:0.321133:0.683701:0.321133:0.671001:0.261579:0.671001:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000
(:@0.321309:0.683468:0.326993:0.683468:0.326993:0.672674:0.321309:0.672674:0.000000
Q K V:@0.327480:0.683877:0.371359:0.683877:0.371359:0.671084:0.327480:0.671084:0.000000:0.000000:0.000000:0.000000:0.000000
, ,:@0.339375:0.683701:0.359606:0.683701:0.359606:0.671001:0.339375:0.671001:0.000000:0.000000:0.000000
)=:@0.372939:0.683468:0.394255:0.683468:0.394255:0.672674:0.372939:0.672674:0.000000:0.000000
Softmax:@0.398345:0.683701:0.449366:0.683701:0.449366:0.671001:0.398345:0.671001:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000:0.000000
(:@0.449541:0.683468:0.455225:0.683468:0.455225:0.672674:0.449541:0.672674:0.000000
Q K:@0.457400:0.676865:0.489510:0.676865:0.489510:0.664071:0.457400:0.664071:0.000000:0.000000:0.000000
·:@0.472196:0.681124:0.476254:0.681124:0.476254:0.668935:0.472196:0.668935:0.000000
⊤:@0.490156:0.675595:0.498807:0.675595:0.498807:0.666331:0.490156:0.666331:0.000000
√:@0.468574:0.687846:0.480777:0.687846:0.480777:0.675658:0.468574:0.675658:0.000000
d:@0.481129:0.692800:0.488159:0.692800:0.488159:0.679892:0.481129:0.679892:0.000000
):@0.501511:0.683468:0.507195:0.683468:0.507195:0.672674:0.501511:0.672674:0.000000
·:@0.510272:0.688137:0.514330:0.688137:0.514330:0.675948:0.510272:0.675948:0.000000
V:@0.517284:0.683877:0.526662:0.683877:0.526662:0.671084:0.517284:0.671084:0.000000
,:@0.528067:0.683701:0.531582:0.683701:0.531582:0.671001:0.528067:0.671001:0.000000
(1):@0.642566:0.683701:0.658959:0.683701:0.658959:0.671001:0.642566:0.671001:0.000000:0.000000:0.000000