Inda za a je: abubuwan da suka faru kyauta masu zuwa don ƙwararrun IT a Moscow (Janairu 14-18)

Abubuwan da suka faru tare da buɗe rajista:

• AI & Wayar hannu
• NeurIPS Sabuwar Shekara Bayan party
• Taron R Moscow #5
• Ganawa na gaba a Moscow (Gastromarket Balchug)
• Koyarwar koyon inji a cikin Avito 2.0

AI & Wayar hannu

Janairu 14, 19: 00-22: 00, Talata

Muna gayyatar ku zuwa taro game da hankali na wucin gadi, aikace-aikacen sa akan na'urorin hannu da mafi mahimmancin fasaha da yanayin kasuwanci na sabbin shekaru goma. Shirin ya ƙunshi rahotanni masu ban sha'awa, tattaunawa, pizza da yanayi mai kyau.

Daya daga cikin masu magana shine majagaba wajen gabatar da sabbin fasahohi a Hollywood, Fadar White House; An ambaci littafinsa mai suna "Augmented: Life in the Smart Lane" a matsayin daya daga cikin littattafan da shugaban kasar Sin ya fi so a cikin jawabinsa na sabuwar shekara.

NeurIPS Sabuwar Shekara Bayan party

Janairu 15, farawa daga 18:00, Laraba

• 18:00 Rajista
• 19:00 Yana buɗewa - Mikhail Bilenko, Yandex
• 19:05 Ƙarfafa koyo a NeurIPS 2019: yadda ya kasance - Sergey Kolesnikov, TinkoffKowace shekara batun ƙarfafa koyo (RL) yana ƙara zafi kuma yana ƙara haɓakawa. Kuma kowace shekara, DeepMind da OpenAI suna ƙara mai a cikin wuta ta hanyar fitar da sabon bot. Shin akwai wani abu da gaske a bayan wannan? Kuma menene sabbin abubuwan da ke faruwa a cikin duk bambancin RL? Bari mu gano!
• 19:25 Binciken aikin NLP a NeurIPS 2019 - Mikhail Burtsev, MIPTA yau, abubuwan da suka fi samun ci gaba a fagen sarrafa harshe na halitta suna da alaƙa da gina gine-ginen gine-gine bisa tsarin harshe da zane-zane na ilimi. Rahoton zai ba da bayyani kan ayyukan da ake amfani da waɗannan hanyoyin don gina tsarin tattaunawa don aiwatar da ayyuka daban-daban. Misali, don sadarwa a kan batutuwa na gaba ɗaya, haɓaka tausayawa da gudanar da tattaunawa mai ma'ana.
• 19: 45 Hanyoyi don fahimtar nau'in saman aikin hasara - Dmitry Vetrov, Faculty of Computer Science, National Research University Higher School of EconomicsZan tattauna takardu da yawa waɗanda ke bincika abubuwan da ba a saba gani ba a cikin zurfin koyo. Wadannan tasirin suna ba da haske game da bayyanar aikin hasara a cikin sararin samaniya kuma suna ba mu damar gabatar da ra'ayoyin da dama. Idan an tabbatar, za'a iya ƙara daidaita girman matakin cikin hanyoyin ingantawa. Wannan kuma zai ba da damar yin hasashen ƙimar da za a iya samu na aikin asara akan samfurin gwajin tun kafin ƙarshen horo.
• 20:05 Binciken ayyuka akan hangen nesa na kwamfuta a NeurIPS 2019 - Sergey Ovcharenko, Konstantin Lakhman, YandexZa mu dubi manyan wuraren bincike da aiki a hangen nesa na kwamfuta. Mu yi kokarin fahimtar ko an riga an warware dukkan matsalolin ta fuskar makarantar, ko ana ci gaba da gudanar da tattakin nasara na kungiyar GAN a dukkanin fagage, wane ne ke adawa da shi, da kuma lokacin da za a yi juyin juya hali na rashin kulawa.
• 20:25 Karfe kofi
• 20:40 Tsarin ƙirar ƙira tare da tsari mara iyaka - Dmitry Emelianenko, Yandex.Muna ba da shawarar ƙirar da za ta iya saka kalmomi zuwa wurare na sabani a cikin jumlar da aka ƙirƙira. Samfurin a fakaice yana koyon tsari mai dacewa da yanke hukunci dangane da bayanai. Ana samun mafi kyawun inganci akan ɗakunan bayanai da yawa: don fassarar injin, amfani da LaTeX da bayanin hoto. An sadaukar da rahoton ga labarin wanda a cikinsa muke nuna cewa tsarin yanke hukunci a zahiri yana da ma'ana kuma ya keɓance ga matsalar da ake warwarewa.
• 20:55 Koyarwar KL-Bambancin Horarwar hanyoyin sadarwa na farko: Ingantacciyar rashin tabbas da Ƙarfin Ƙarfafawa - Andrey Malinin, Yandex.Kwanan nan an yi amfani da hanyoyin haɗin kai don ƙimar rashin tabbas ga ayyukan gano ɓarna, gano shigar da ba-rarrabuwa da gano harin gaba. An gabatar da hanyoyin sadarwa na baya a matsayin hanya mai inganci don yin koyi da tarin samfura don rarrabuwa ta hanyar daidaita Dirichlet kafin rarrabawa akan rarraba kayan fitarwa. An nuna waɗannan samfuran sun fi sauran hanyoyin haɗin kai, kamar Monte-Carlo Dropout, akan aikin gano shigarwar ba-rarrabuwa. Koyaya, ƙaddamar da Cibiyoyin sadarwa na gaba zuwa rikitattun bayanai tare da azuzuwan da yawa yana da wahala ta amfani da ƙa'idodin horon da aka gabatar. Wannan takarda tana ba da gudummawa biyu. Na farko, mun nuna cewa ma'aunin horon da ya dace don Cibiyoyin sadarwa na Farko shine juyewar KL tsakanin rarrabawar Dirichlet. Wannan al'amurra suna magance yanayin rarraba bayanan horo, yana ba da damar cibiyoyin sadarwa na farko don samun nasarar horar da su akan ayyukan rarrabuwa tare da azuzuwan da yawa ba bisa ka'ida ba, tare da haɓaka aikin ganowa ba tare da rarrabawa ba. Na biyu, yin amfani da wannan sabon ma'aunin horo, wannan takarda ta yi bincike ta amfani da hanyoyin sadarwa na farko don gano hare-haren abokan gaba da kuma ba da shawarar wani nau'i na horo na gaba ɗaya. An nuna cewa gina nasarar nasarar da aka samu ta akwatin farin akwatin, wanda ke shafar tsinkaya da gano ganowa, a kan hanyoyin sadarwa na farko da aka horar da su akan CIFAR-10 da CIFAR-100 ta amfani da tsarin da aka tsara yana buƙatar babban adadin ƙoƙarin lissafi fiye da cibiyoyin sadarwar da aka kare ta amfani da daidaitaccen abokin gaba. horo ko MC-dropout.
• 21:10 Tattaunawar kwamiti: "NeurlPS, wanda ya yi girma da yawa: wanene ke da laifi kuma me za a yi?" - Alexander Krainov, Yandex
• 21:40 Bayan party

Taron R Moscow #5

Janairu 16, 18: 30-21: 30, Alhamis

• 19: 00-19: 30 "Maganin matsalolin aiki ta amfani da R don dummies" - Konstantin Firsov (Netris JSC, Babban Injiniyan Aiwatarwa).
• 19: 30-20: 00 "Haɓaka kayayyaki a cikin tallace-tallace" - Genrikh Ananyev (PJSC Beluga Group, Shugaban bayar da rahoto ta atomatik).
• 20: 00-20: 30 "BMS a cikin X5: yadda ake yin kasuwanci-tsarin hakar ma'adinai a kan rajistan ayyukan POS marasa tsari ta amfani da R" - Evgeniy Roldugin (X5 Retail Group, Head of Service Quality Control Tools Department), Ilya Shutov (Media Tel, Head Masanin kimiyyar Sashen bayanai).

Ganawa na gaba a Moscow (Gastromarket Balchug)

Janairu 18, 12: 00-18: 00, Asabar

• "Yaushe ya cancanci sake rubuta aikace-aikacen daga karce, da kuma yadda za a shawo kan wannan kasuwancin?" - Alexey Pyzhyanov, mai haɓakawa, SiburGaskiyar labarin yadda muka magance bashin fasaha a cikin mafi m hanya. Zan gaya muku game da shi:
  1. Me yasa aikace-aikace mai kyau ya zama mummunan gado.
  2. Yadda muka yanke shawara mai wahala don sake rubuta komai.
  3. Yadda muka sayar da wannan ra'ayin ga mai samfurin.
  4. Abin da ya fito daga wannan ra'ayin a ƙarshe, da kuma dalilin da ya sa ba ma yin nadama game da shawarar da muka yanke.

• "Vuejs API ba'a" - Vladislav Prusov, Frontend developer, AGIMA

Koyarwar koyon inji a cikin Avito 2.0

Janairu 18, 12: 00-15: 00, Asabar

• 12:00 "Zindi Sendy Logistics Challenge (rus)" - Roman Pyankov
• 12:30 "Data Souls Wildfire AI (rus)" - Ilya Plotnikov
• 13:00 Karfe kofi
• 13:20 "Topcoder SpaceNet 5 Kalubale & Sa hannu kan Kalubalen Tauraron Dan Adam na 3 na Tellus" - Ilya Kibardin
• 14:00 Karfe kofi
• 14:10 "Codalab Automated Time Series Regression (eng)" - Denis Vorotyntsev

source: www.habr.com