Trifork Group AG · OMXCOP:TRIFOR

Trifork — Technical Presentation (September 2020)

31 slides from Trifork's technical presentation, Sep 29, 2020

  1. Slide 1: Trifork Tech Update: Making AI more accessible with AutoML

    Trifork Tech Update Making Al more accessible with AutoML The Tech Update will begin shortly... While you wait, please answer this*questioOn,and submit your answer in the chat: Do you predict that Machine Learning will be important for your industry? TRIFORK. .. think software we?

  2. Slide 2: A few practical notes

    A few practical notes Recording We will be recording the Tech Update and make it available for you after the event Questions & Answers We will be using the chat function to engage with you during the Tech Update. We look forward to your questions Survey We will ask for your feedback after the Tech U

  3. Slide 3: Making AI more accessible with AutoML

    TRIFORK. Dan Rose Co-founder & Al enthusiast Making Al more accessible with | | AutoML . - Nicholai Stalung September 29, 2020 Z Data Scientist & Machine Learning Expert °® Nursel Yildirim Head of Digital Design & Moderator TRIFORK. think software

  4. Slide 4: AutoML

    AutoML AI and deep learning without coding

  5. Slide 5: About me

    About me • Dan Rose • Co-founder at Paperflow.com – An Applied AI Company • Blog about AI at danrose.ai

  6. Slide 6: What is AutoML and why is it relevant?

    What is AutoML and why is it relevant? • Not a new thing • Automated Machine learning = No code • Mostly provided by the big vendors – But keep an eye out for the small players • When to use it and when not to

  7. Slide 7: How AutoML works

    How AutoML works • AutoML find the right algorithm for your data through experiments • You only provide data • Transfer learning

  8. Slide 8: Flip the food pyramid

    Flip the food pyramid Data Data Problem Problem Algoritms Algoritms

  9. Slide 9: Kims example

    Kims example • 40 USD and 3 hours of work • +90% correct • Only 50 examples per type

  10. Slide 10: It can be used for..

    It can be used for.. • NLP and language models • Document classification • Vision(Image and video) • Tabular data (Like predictive maintaince) • Basicly everything…

  11. Slide 11: Before you make the jump

    Before you make the jump There’s some pros and cons

  12. Slide 12: It's fast and good for prototyping

    It’s fast and good for prototyping

  13. Slide 13: Doesn't grow old

    Doesn’t grow old

  14. Slide 14: Less infrastructure – Skip 95%

    Less infrastructure – Skip 95%

  15. Slide 15: Fewer experts needed

    6 afab eo vw Fewer experts needed ee \nusnan CO )Basic Civevits Basic Civewl’s = equals Flaw ands o\rage ~ isthe fovce. ond itis labled measuved in omy as wats CeESISTAYUE - isthe apposition and js measured im Ones (22) ot Theory _ ee . ct a Poo sh oe > cane

  16. Slide 16: Better quality at first

    Better quality at first

  17. Slide 17: Less insights

    Less insights

  18. Slide 18: Less flexible

    Less flexible

  19. Slide 19: Costs doesn't scale

    = Costs doesn’t scale

  20. Slide 20: Will never be the best

    a : nal a pew ta 4 Will never be the best

  21. Slide 21: The future

    The future • Data is the competive advantage • Everybody can do AI now

  22. Slide 22: AutoML at Nicholai Stalung (Trifork)

    $WVQ/. -µT*QFV 'TKXG9GNN 0KEJQNCK5V¢NWPI .($''$6$5&Ͳ(06Ͳ5664Ͳ)14- PTU#VTKHQTMEQO

  23. Slide 23: AutoML in the development cycle

    $WVQ/.KPVJGFGXGNQROGPVE[ENG 8CNKFCVKQP2TQQHKP2TQFWEVKQP 4GNGCUGE[ENG 'CVC /QFGN 5QHVYCTG /QFGN 2NCPPKPI 2NCPPKPI 4GVTCKPKPI RTGRTQEGUUKPI VTCKPKPI +CTFYCTG OCVWTCVKQP

  24. Slide 24: Driving score case

    -µT*QFV 'TKXG9GNN Ѵ 'TKXKPI5EQTKPICRRD[0GO)QTUKMTKPI Ѵ 2TQXKFGUCUEQTGQPGCEJVTKR Ѵ 2C[CNQYGTRTGOKWOQPECTKPUWTCPEG Ѵ 4GEGKXGHGGFDCEMQP[QWTFTKXKPI Ѵ 8KUKQPVQOKPKOK\GCEEKFGPVU

  25. Slide 25: What was the situation?

    9JCVYCUVJG UKVWCVKQP" 4GNGCUGE[ENG Ѵ %WUKPGUUQWVEQOGYCUFGȤPGFCPFXCNKFCVGF Ѵ 5QHVYCTGFGXGNQROGPVJCFDGIWP Ѵ /CVWTGVJGWPFGTN[KPI/CEJKPG.GCTPKPIOQFGN 2NCPPKPI 5QHVYCTG /QFGN +CTFYCTG OCVWTCVKQP Ѵ 2GTHQTOCPEGKPUKIJVUHQTRTQFWEVKQP Ѵ 2KRGNKPGTGRNKECVKQPCPFOQFGNFGRNQ[OGPV

  26. Slide 26: Why did we choose AutoML?

    9J[FKFYGEJQQUG$WVQ/." Ѵ $WVQ/.JCUDGEQOGOCVWTG 2SHQ6RXUFH &RPPHUFLDO Ѵ 5WDUVCPVKCNN[TGFWEG FGXGNQROGPVVKOG Ѵ 4QDWUVCICKPUVǺJWOCPGTTQTUǻ &ORXG$XWR0/ Ѵ (ZRNQTGOQTGOQFGNQRVKQPU Ѵ 7PKȤGFCRRTQCEJVQ FGXGNQROGPVCPFFGRNQ[OGPV

  27. Slide 27: What was the outcome of using AutoML?

    9JCVYCUVJGQWVEQOGQHWUKPI$WVQ/." ͲPJCNHVJGPQTOCNVKOGYG YGTGCDNGVQ Ѵ 1PDQCTFC'CVC5EKGPVKUV Ѵ 4GOQFGNVJGKPRWVRKRGNKPG Ѵ 6GUVCPFTGFWEGDKCU Ѵ ͲORTQXGVJGǺCEEWTCE[ǻ

  28. Slide 28: How can you leverage AutoML?

    +QYECP[QWNGXGTCIG$WVQ/." $WVQ/.KUPQVKPFWUVT[URGEKȤE VQKPUWTCPEG 0QEWTTGPV/.KPKVKCVKXGU Ѵ 7UG$WVQ/.VQXCNKFCVG[QWT DWUKPGUUECUG &WTTGPV/.KPKVKCVKXGU Ѵ &JCNNGPIG[QWTUVTCVGI[VQFQ KVǺD[JCPFǻ JVVRUENQWFIQQINGEQODNQIRTQFWEVUCKOCEJKPGNGCTPKPIJQYMCIINGUQNXGFCURCORTQDNGOWUKPICWVQON

  29. Slide 29: How can you leverage AutoML?

    +QYECP[QW NGXGTCIG $WVQ/." JVVRUYYYOEMKPUG[EQODWUKPGUUHWPEVKQPUOEMKPUG[CPCN[VKEUQWTKPUKIJVUTGVJKPMKPICKVCNGPVUVTCVGI[CUCWVQOCVGFOCEJKPGNGCTPKPIEQOGUQHCIG

  30. Slide 30: Key takeaways

    -G[6CMGCYC[U $WVQ/.TGFWEGUVKOGVQOCTMGV Ѵ /WEJQHVJGǺJGCX[ǻEQFKPIKUCDUVTCEVGFCYC[ $WVQ/.YKNNOCMG/.OQFGNUDGVVGTQPCXGTCIG Ѵ 3WKEMN[VGUVQVJGTEQODKPCVKQPUCPFCNIQTKVJOU )QEWUOQTGQP[QWTDWUKPGUUEJCNNGPIGUVJCPGPIKPGGTKPI Ѵ 'CVCKPUKIJVUPQVGPIKPGGTKPICTGVJGMG[VQUWEEGGF 4GVJKPM[QWTVCNGPVUVTCVGI[ Ѵ %WUKPGUUFQOCKP

  31. Slide 31: Thank You

    6JCPM;QW

Trifork Group — Technical Presentation (September 2020) on 6ix