SensiML Ƙara Hasashen Kulawa a cikin Na'urorin Ginin Waya

Ajanda

Pre-aiki: Masu amfani sun shigar da Sauƙi Studio da Kayan Aikin Binciken SensiML a gaba

• Gabatarwar Mai watsa shiri - 5 minutes
• Gabatar da ra'ayoyi da burin lab - 10 minutes
• "Real-time" aiwatar da mataki-mataki hanya don ƙirƙirar samfurin - 60 minutes
  • Flash SensiML mai jituwa firmware tattara bayanai zuwa Thunderboard Sense 2 (TBS2)
  • Saita kuma haɗa TBS2 zuwa SensiML Data Capture Lab
  • Ɗauki bayanan 'slide demo' tare da allo (masu amfani ba za su sami kayan aikin Fan ba)
  • Lakabi bayanai da adanawa da sample project (ba za mu yi amfani da ragowar karatun ba ko da yake)
  • Kira Studio Studio (a wannan lokacin, masu amfani za su yi aiki daga bayanan demo na TBS2 da aka riga aka tattara)
  • Yi aiki ta hanyar matakan ƙirar ƙirar gina ƙirar gano jihar fan
  • Ƙirƙiri Kunshin Ilimi
  • Na zaɓi: Samfurin Flash zuwa TBS2
• Smart Building Applications demo bidiyo - 5 minutes
• Q&A – 10 minutes

Gabatarwa SensiML

• SensiML kamfani ne na kayan aikin software na B2B don AI a gefen IoT
  • Yana ba masu haɓaka damar ƙirƙira ƙirar firikwensin ML mai ƙarfi ba tare da ƙwarewar kimiyyar bayanai ba
  • Samfura masu ƙanƙanta kamar 10KB!
  • Tsohuwar ƙungiyar kayan aikin software na Intel Curie/Quark MCU AI, an bar su don ƙirƙirar SensiML a cikin 2017
• Silicon Labs da SensiML Magani
  • Kawo ingantaccen ML mai ƙarfi ga dangin EFR32/EFM32 MCU
  • Samfurin aikace-aikacen IoT mai sauri tare da Thunderboard Sense 2
• SensiML yana da kwanciyar hankali da tallafi na duniya
  • An samu a cikin 2019 ta QuickLogic Corp; saitin kuma gudanar da shi azaman na gabaɗaya na software mai zaman kansa (wanda yake a Portland, OR)
  • Kafa abokan haɗin gwiwa (Avnet, Future Electronics, Mouser, Shinko Shoji)
  • Ofisoshin tallace-tallace / Tallafawa a Burtaniya, Amurka, Japan, Taiwan, China

Dama don TinyML a cikin Gine-ginen Waya

Kalubale tare da Haɓaka Aikace-aikacen Sensor Smart IoT

Cloud-Centric AI 

• High Network Traffic Load
• Babban Latitude
• Kadan Mai Haƙuri Laifi
• Haɗarin Tsaron Bayanai wanda ba a sani ba
• Damuwa na Keɓantawa

Zurfin Ilmantarwa

• Babban buƙatun bayanan horo
• Large memory footprint
• Babban aikin sarrafawa
• Babban amfani da wutar lantarki
• Rayuwar baturi mara kyau

Wuraren Ƙarshen Ƙarshen Hannu 

• Sannu a hankali kuma mai tsananin aiki
• Girman lambar da ba a sani ba a gaba
• Ƙwarewar kimiyyar bayanai kaɗan
• Complex AI/ML code dakunan karatu
• Ba mai daidaitawa / gasa ba

TinyML = IoT Edge ML + AutoML

• IoT Edge ML: Madaidaitan wuraren ƙarewa
  • Fitar hanyar sadarwa maras muhimmanci da tsawon rayuwar batir mara waya
  • Babu sarrafa girgije ko dogaro da hanyar sadarwa
  • Mai da martani na ainihi
• AutoML: Inganta Ba tare da Kwarewar AI ba
  • Auto-optimizer yana zaɓar mafi kyawun samfuri don bayanan da aka bayar
  • Classic inji koyo (ML) ta hanyar zurfin koyo
  • SensiML TinyML yana samar da ƙira waɗanda ƙanana kamar 10KB!
• Ba a buƙatar yin coding da hannu
  • Lambar ƙirar ƙira ta atomatik daga bayanan horo na ML
  • Yana adana watanni na ƙoƙarin haɓakawa, da ƙwarewar kimiyyar bayanai
  • Mai haɓakawa na iya canza kowane bangare na lambar AutoML kamar yadda ake so

Samfurin Gina Aikin Gina

Ɗaukar Bayanai 

• Lokaci: Sa'o'i zuwa Makonni* (Ya danganta da wahalar tattara bayanan aikace-aikacen)
• Ƙwarewa: Ƙwararrun Ƙwararru (Kamar yadda ake buƙata don tattarawa da lakabi abubuwan sha'awa)

Lura: Za mu yi amfani da wasu bayanan da aka tattara a baya don haɓaka wannan matakin don taron bitar

Gina Misali 

• Lokaci: Minti zuwa Sa'o'i (Ya danganta da matakin sarrafa samfurin)
• Ƙwarewa: Babu (Cikakken AutoML)
  • Mahimman Ka'idodin ML (Babban kunna UI)
  • Python Programming (Cikakken sarrafa bututun mai)

Na'urar Gwaji 

• Lokaci: Mintuna zuwa Makonni (Ya danganta da buƙatun haɗa lambar app)
• Ƙwarewa: Babu (firmware na binary tare da lambar rufe I/O ta atomatik)
  Haɗin Shirye-shiryen (Haɗin ɗakin ɗakin karatu na SensiML ko tushen C tare da lambar mai amfani)

Burin Bita

• Gabatar da kayan aikin SensiML na TinyML da tsarin ginin ƙirar akan Silicon Labs Thunderboard Sense 2
• Ƙwarewa tare da ci gaban algorithm na firikwensin ML mai kulawa da bayanai
• Koyi tsarin aiki daga tarin bayanai ta hanyar inganci da gwajin kan na'ura don gina ƙirar IoT
• Gina samfurin tsinkayar HVAC mai aiki fara-zuwa-ƙarewa
• Amsar tambayoyin da za ku iya yi game da tsarin ƙirƙirar TinyML

Aikace-aikacen Kula da Hasashen HVAC Aiki

• Don dalilai na hannun-kan mu, za mu gina na'urar kula da fan
• Magoya bayan sun yi amfani da ko'ina wajen gina tsarin HVAC: Blowers, sanyaya kayan aiki, masu sarrafa iska, iskar iska.
• Rashin gazawa ko lalacewa na iya haifar da asarar inganci, ƙara yawan kuzari, gazawar HVAC
• Za mu gina na'urar sa ido mai sauƙi wacce za ta iya gano jihohi na al'ada da mara kyau:
  • Kashe / kunna fan
  • Matsakaicin sako-sako
  • Magoya bayan gadi
  • Sashi ko cikakken katange kwararar iska
  • Tsagewar ruwa
  • Yawan girgiza

Mu Fara Tsarin

Tsarin bita na "Real-time" mataki-mataki don ƙirƙirar samfurin - minti 60

• Flash SensiML mai jituwa firmware tattara bayanai zuwa Thunderboard Sense 2 (TBS2)
• Saita kuma haɗa TBS2 zuwa SensiML Data Capture Lab
• Ɗauki bayanan 'slide demo' tare da allo (masu amfani ba za su sami kayan aikin Fan ba)
• Lakabi bayanai da adanawa da sample project (ba za mu yi amfani da ragowar karatun ba ko da yake)
• Kira Studio Studio (a wannan lokacin, masu amfani za su yi aiki daga bayanan demo na TBS2 da aka riga aka tattara)
• Yi aiki ta hanyar matakan ƙirar ƙirar gina ƙirar gano jihar fan
• Ƙirƙiri Kunshin Ilimi
• Samfuran Flash zuwa TBS2

Demo Bidiyo

Haƙƙin mallaka © 2021 SensiML Corporation. An kiyaye duk haƙƙoƙi.

Takardu / Albarkatu

SensiML Ƙara Hasashen Kulawa a cikin Na'urorin Ginin Waya [pdf] Umarni Ƙara Hasashen Kulawa a cikin Na'urorin Ginin Waya, Kulawa a cikin Na'urorin Gina Mai Waya, Na'urar Gina Mai Wayo, Na'urorin Gina

Magana

• Manual mai amfani