Google DeepMind Princeton
Type Laboratori ea lipatlisiso
US 2019 Batho ba 11-50 
Themes
Google DeepMind Princeton ke laboratori ya dipatlisiso e Princeton, New Jersey, e thehilweng ka 2019 e le karolo ya Google DeepMind. E tsepamisa maikutlo dipatlisisong tsa motheo tsa machine learning, ho kenyeletswa di-algorithm tse sebetsang hantle tse nang le ditiisetso tsa theory, spectral transformers, mekgwa ya adaptive gradient, le ditshebediso mafapheng a bongaka a kang taolo ya ventilation. Laboratori ena e sebetsa mmoho le dihlopha tse ding tsa dipatlisiso tsa Google ho phatlalla le U.S., India, Israel, le Europe ho ntshetsa pele mosebetsi wa AI wa theory le wa tshebediso.
E sebetsa e le yuniti e nyane e ikgethileng, Google DeepMind Princeton e shebane le bafuputsi le baenjiniere ba nang le thahasello mekgweng e tiileng, e thehilweng dipalong, ya machine learning. Mosebetsi wa yona o hokahanya theoretical computer science le ntshetsopele ya AI e sebetsang, mme o kenya letsoho mafapheng a kang non-stochastic control theory le bokgoni ba di-algorithm. Sebaka sa laboratori ena Princeton se e hokahanya le setjhaba se seholo sa thuto le sa mahlale tikolohong eo.
Motheo ea Lipatlisiso tsa Ho Ithuta ha Mechini
Google DeepMind Princeton e sebetsa e le laboratori e khethehileng ea lipatlisiso e tsepamisitseng maikutlo metheong ea lipalo ea ho ithuta ha mechini. Sehlopha se hlahisa li-algorithm tse lekanyang tshebetso e sebetsang le ditiisetso tsa thuto, ho netefatsa botsitso ditshebedisong tsa nnete tsa lefatshe.
Dibaka tsa bohlokwa tsa lipatlisiso di kenyelletsa spectral transformers, adaptive gradient methods, le non-stochastic control theory. Mekgwa ena e reretswe ho ntlafatsa bokgoni ba ho kwetlisa mehlala ha ho ntse ho bolokwa tlhaloseho le botshepehi.
Mosebetsi wa laboratori o boetse o atolosetswa ditshebedisong tsa bongaka, tse kang ho ntlafatsa ditsamaiso tsa taolo ya ho kenya moya. Sena se bontsha kamoo tsoelopele ya thuto e ka fetolelang ntlafatsong e bonahalang ya theknoloji ya tlhokomelo ya bophelo.
Dikamano tsa Tshebedisano le tsa Thuto
Laboratori ya Princeton e boloka dikamano tse haufi le Princeton University, e hodisang tikoloho eo ho yona botsitso ba thuto bo kopanang le boqapi ba indasteri. Sebaka sa yona se haufi le khamphase se nolofalletsa dipuisano tse atisang ho ba teng le barupelli le baithuti, ho matlafatsa lipatlisiso le thuto ka bobedi.
Tshebedisano e akaretsa dihlopha tse ngata tsa dipatlisiso tsa Google ho phatlalla le U.S., India, Israel, le Europe. Marangrang ana a lefatshe lohle a dumella laboratori ho kopanya maikutlo a fapaneng le ho potlakisa tswelopele dipatlisisong tsa motheo tsa AI.
Boholo ba sehlopha bo dula bo le bonyenyane ka boomo, ho etelletsa botebo pele ho feta boholo. Sebopeho sena se tshehetsa ho hlahloba ka tsepamiso mathata a rarahaneng ha ho ntse ho bolokwa bokgoni ba ho ikamahanya le ditsela tsa dipatlisiso.
Merero ya Bohlokwa le Boqapi
Merero ya laboratori e bontsha motswako wa dipatlisiso tsa thuto le tse sebediswang. Ka mohlala, spectral transformers e hlahloba meaho e metjha bakeng sa ho etsa mohlala wa tatelano, athe adaptive gradient methods e ntlafatsa mekgwa ya optimization bakeng sa ho ithuta ha mechini ka tekanyo e kgolo.
Ho reinforcement learning, sehlopha se hlahisitse mekgwa e ka arolohang le non-stochastic control theory. Tsoelopele ena e sebetsana le diphephetso dibakeng tse fetohang moo mekgwa ya setso e sa fihlelleng katleho.
Ditshebediso tsa bongaka di ntse di le tsepamiso ya bohlokwa. Morero wa AI for Better Medical Ventilators o sebedisa ho ithuta ha mechini ho ntlafatsa diphetho tsa bakudi dibakeng tsa tlhokomelo e mahlonoko. Maiteko a mang, a kang SAMUEL le GGT, a shebane le adaptive regularization e sebedisang memori ka bokgoni bakeng sa kwetliso e ka hodiswang.
Sehlopha le Tshusumetso ya Dipatlisiso
Laboratori e etellwa pele ke bafuputsi ba nang le semelo sa computer science, optimization, le control theory. Ditho tsa jwale di kenyelletsa Elad Hazan, Wenhan Xia, le Ani Majumdar, hara ba bang, ka boitseanape bo akaretsang moralo wa li-algorithm, reinforcement learning, le AI ya bongaka.
Ditho tsa mehleng di se di fetetse mesebetsing ya thuto le indasteri, di kenya letsoho tsoelopeleng e pharalletseng ya ho ithuta ha mechini. Mosebetsi wa tsona hangata o hlaha dikopanong tsa maemo a hodimo, tse kang NeurIPS le ICML, ho bontsha tshusumetso ya laboratori lefapheng lena.
Dikgatiso tse tswang laboratori di akaretsa dihlooho tse fapaneng, ho tloha ho thuto e akaretsang ya nonlinear dynamics ho isa ho benchmarking robotics policies. Mosebetsi wa moraorao o kenyelletsa ho lekola large language models dibakeng tse etsisitsweng le ho hlahisa meralo ya polokeho bakeng sa ditsamaiso tsa liroboto.
Dipatlisiso tsa laboratori di boetse di atolosetswa ho li-algorithm tsa kwetliso, ka merero e kang VeLO e hodisang learned optimizers bakeng sa tshebediso e pharalletseng. Maiteko ana a totobatsa boitlamo ba sehlopha ba ho hokahanya thuto le tshebetso ntshetsopeleng ya AI.