Iziganeko ezinobhaliso oluvulekileyo:
NgoJanuwari 14, 19:00-22:00, ngoLwesibini
Sikumema kwintlanganiso malunga nobukrelekrele bokwenziwa, ukusetyenziswa kwayo kwizixhobo eziphathwayo kunye nezona ndlela zibalulekileyo zetekhnoloji kunye neshishini leminyaka elishumi entsha. Inkqubo ibandakanya iingxelo ezinomdla, iingxoxo, ipizza kunye nomoya omhle.
Esinye sezithethi singuvulindlela ekuziseni itekhnoloji yamva nje eHollywood, i-White House; incwadi yakhe ethi "Augmented: Life in the Smart Lane" yakhankanywa njengenye yeencwadi zakhe ezizithandayo zereferensi nguMongameli waseTshayina kwintetho yakhe yoNyaka oMtsha.
NgoJanuwari 15, ukuqala ngo-18:00, ngoLwesithathu
- 18:00 Ubhaliso
- 19:00 Ukuvulwa - Mikhail Bilenko, Yandex
- 19:05 Ukomeleza ukufunda kwi-NeurIPS 2019: bekunjani-uSergey Kolesnikov, TinkoffMinyaka le isihloko se-reinforcement learning (RL) siya sisiba shushu kwaye sishukunyiswa ngakumbi. Kwaye minyaka le, i-DeepMind kunye ne-OpenAI yongeza amafutha emlilweni ngokukhupha i-bot entsha yokusebenza kwamandla angaphezu kwawomntu. Ngaba ikho into exabiseke ngokwenene emva koku? Kwaye zeziphi iindlela zamva nje kuzo zonke iintlobo ngeentlobo zeRL? Masifumanise!
- 19:25 Uphononongo lomsebenzi we-NLP kwi-NeurIPS 2019-Mikhail Burtsev, MIPTNamhlanje, ezona ndlela zihamba phambili kwintsimi yokulungiswa kolwimi lwendalo zihambelana nokwakhiwa kwezakhiwo ezisekelwe kwiimodeli zolwimi kunye neegrafu zolwazi. Ingxelo iya kubonelela ngamagqabantshintshi emisebenzi apho ezi ndlela zisetyenziselwa ukwakha iinkqubo zengxoxo ukuphumeza imisebenzi eyahlukeneyo. Umzekelo, ukunxibelelana ngezihloko ngokubanzi, ukwandisa uvelwano kunye nokuqhuba incoko ejolise kwiinjongo.
- 19:45 Iindlela zokuqonda uhlobo lomphezulu womsebenzi wokulahleka - uDmitry Vetrov, iFaculty of Computer Science, iNational Research University School of EconomicsNdiza kuxubusha amaphepha amaninzi aphonononga iziphumo ezingaqhelekanga ekufundeni nzulu. Ezi ziphumo zikhanyisela ukubonakala komphezulu womsebenzi wokulahleka kwisithuba sobunzima kwaye sivumele ukuba sibeke phambili inani leengcamango. Ukuba kuqinisekisiwe, kuya kwenzeka ukulawula ngokufanelekileyo ubungakanani benyathelo kwiindlela zokuphucula. Oku kuya kwenza ukuba kunokwenzeka ukuqikelela ixabiso elifikelelekayo lomsebenzi wokulahlekelwa kwisampuli yokuvavanya ixesha elide ngaphambi kokuphela koqeqesho.
- 20:05 Uphononongo lwemisebenzi kumbono wekhompyuter kwi-NeurIPS 2019-uSergey Ovcharenko, uKonstantin Lakhman, YandexSiza kujonga iinkalo eziphambili zophando kunye nokusebenza kumbono wekhompyutheni. Makhe sizame ukuqonda ukuba zonke iingxaki sele zixazululwe ukusuka kwindawo yokujonga i-academy, nokuba uhambo oluphumelelayo lwe-GAN luyaqhubeka kuzo zonke iindawo, ngubani ochasayo, kwaye xa ukuguqulwa okungalawulwayo kuya kwenzeka.
- 20:25 Ikhefu ikofu
- 20:40 Imodeli yolandelelwano kunye nomyalelo ongenamkhawulo wesizukulwana-Dmitry Emelianenko, YandexSicebisa imodeli enokufaka amagama kwiindawo ezingafanelekanga kwisivakalisi esivelisiweyo. Imodeli ifunda ngokungafihlisiyo ulandelelwano olululo lwe-decoding olusekwe kwidatha. Owona mgangatho ugqwesileyo uphunyezwa kwiiseti zedatha ezininzi: ukuguqulelwa komatshini, sebenzisa kwiLaTeX kunye nenkcazo yomfanekiso. Ingxelo inikezelwe kwinqaku apho sibonisa ukuba i-decoding order efundiweyo iyavakala kwaye ingqalileyo kwingxaki esonjululwayo.
- 20:55 Ukuguqula i-KL-Divergence yoQeqesho lweeNethiwekhi zangaphambili: Ukungaqiniseki okuPhuculweyo kunye nokuQiniseka kwe-Adversarial - Andrey Malinin, YandexIindlela zokuhlanganisana zoqikelelo lokungaqiniseki zisanda kusetyenziswa kwimisebenzi yokuchongwa kokuhlelwa ngendlela engeyiyo, ukubhaqwa kwegalelo elingaphandle kokusasazwa kunye nokufunyanwa kohlaselo lotshaba. IiNethiwekhi zangaphambili ziye zacetywa njengendlela yokulinganisa ngokufanelekileyo ukuhlanganiswa kweemodeli zokuhlelwa ngokumisela i-Dirichlet ngaphambi kokusasazwa kokusasazwa kwemveliso. Le mifuziselo ibonakaliswe ukuba igqwesile ezinye iindlela zokudityaniswa, ezifana ne-Monte-Carlo Dropout, kumsebenzi wokukhangela igalelo elingaphandle kokusasazwa. Nangona kunjalo, ukulinganisa iiNethiwekhi zaPhambili kwiiseti zedatha ezintsonkothileyo ezineeklasi ezininzi kunzima usebenzisa iindlela zoqeqesho ezazicetywa ekuqaleni. Eli phepha lenza imirhumo emibini. Okokuqala, sibonisa ukuba ikhrayitheriya yoqeqesho efanelekileyo yeeNethiwekhi eziPhambili yi-reverse KL-divergence phakathi kokunikezelwa kweDirichlet. Le miba ijongana nobume bonikezelo lwethagethi yedatha yoqeqesho, okwenza ukuba uthungelwano lwangaphambili luqeqeshwe ngempumelelo kwimisebenzi yokuhlelwa kunye neeklasi ezininzi ngokungenasizathu, kunye nokuphucula ukusebenza kokubona ngaphandle kokusasazwa. Okwesibini, ukuthatha ithuba lomgaqo omtsha woqeqesho, eli phepha liphanda ngokusebenzisa iiNethiwekhi zangaphambili ukufumana uhlaselo oluchasayo kwaye liphakamisa uhlobo oluqhelekileyo loqeqesho lotshaba. Kuboniswa ukuba ukwakhiwa kohlaselo oluphumeleleyo lwebhokisi emhlophe, echaphazela ukuqikelelwa kunye nokuphepha ukubhaqwa, ngokuchasene neeNethiwekhi zangaphambili eziqeqeshwe kwi-CIFAR-10 kunye ne-CIFAR-100 kusetyenziswa indlela ecetywayo ifuna inani elikhulu lemizamo yokubala kunokuchasana nothungelwano olukhuselweyo usebenzisa i-adversarial eqhelekileyo. uqeqesho okanye MC-dropout.
- 21:10 Ingxoxo yephaneli: βNeurlPS, eye yanda kakhulu: ngubani na obekekβ ityala, makenze ntoni na?β - Alexander Krainov, Yandex
- 21:40 Emva kwepati
NgoJanuwari 16, 18:30-21:30, ngoLwesine
- 19: 00-19: 30 "Ukuxazulula iingxaki zokusebenza usebenzisa i-R yeedummies" - uKonstantin Firsov (uNetris JSC, iNjineli eyiNtloko yokuPhunyezwa).
- 19: 30-20: 00 "Ukulungiswa kwe-inventri kwintengiso" - uGenrikh Ananyev (iQela le-PJSC Beluga, iNtloko ye-automation yengxelo).
- 20:00-20:30 "I-BMS kwi-X5: indlela yokwenza imigodi ye-business-process kwi-logs ye-POS engacwangciswanga usebenzisa i-R" - u-Evgeniy Roldugin (iQela le-X5 yokuthengisa, iNtloko yeSebe leZixhobo zoLawulo lweZixhobo zeNkonzo), u-Ilya Shutov (I-Media Tel, iNtloko yeNzululwazi yedatha yeSebe).
NgoJanuwari 18, 12:00-18:00, ngoMgqibelo
- "Kufanelekile nini ukubhala kwakhona isicelo ukusuka ekuqaleni, kunye nendlela yokuqinisekisa ishishini ngale nto" - uAlexey Pyzhyanov, umphuhlisi, uSiburIbali lokwenyani lendlela esijongene ngayo namatyala obugcisa ngeyona ndlela inzima. Ndiza kukuxelela ngayo:
- Kutheni isicelo esilungileyo sijike saba lilifa elibi.
- Indlela esenze ngayo isigqibo esinzima sokuphinda sibhale yonke into.
- Siwuthengise njani lo mbono kumnini wemveliso.
- Yintoni eyaphuma kule ngcamango ekugqibeleni, kwaye kutheni singazisoli ngesigqibo esasenzayo.
- "I-API ye-Vuejs ihlekisa" - Vladislav Prusov, umphuhlisi we-Frontend, i-AGIMA
NgoJanuwari 18, 12:00-15:00, ngoMgqibelo
- 12:00 "Zindi Sendy Logistics Challenge (rus)" - Roman Pyankov
- 12:30 "Data Souls Wildfire AI (rus)" - Ilya Plotnikov
- 13:00 Ikhefu ikofu
- 13:20 βI-Topcoder SpaceNet 5 Umngeni kunye nokuSayina i-3rd Tellus Satellite Challenge (eng)β - Ilya Kibardin
- 14:00 Ikhefu ikofu
- 14:10 "Codalab Automated Time Series Regression (eng)" - uDenis Vorotyntsev
umthombo: www.habr.com