Big Data Week: Forget Big Data - Think Machine Learning!

Download Big Data Week: Forget Big Data - Think Machine Learning!

Post on 21-Jan-2017



Data & Analytics

2 download

Embed Size (px)


<ul><li><p>Forget Big Data - Think Machine Learning! </p><p>Jane Zavalishina, CEO of Yandex Data Factory</p></li><li><p>Big Data is a new oil</p></li><li><p>Finextra, 09.01.2015 on research from Aite Group</p><p> Big data is a particular sore point, invoking dissatisfaction among 76% of North American bankers</p><p>Why the bankers are so unhappy? </p></li><li><p>Volume More data than ever</p><p>Velocity High speed of change</p><p>Variety Different types, a lot of unstructured data</p></li><li><p>problemBig data = Big </p></li><li><p>problemBig data = Big IT</p></li><li><p>projectsBig data = Big IT</p></li><li><p>budgetsBig data = Big IT</p></li><li><p>What about value? </p></li><li><p>Heres how most people imagine use of Big Data</p></li><li><p>Heres how it will actually look like</p></li><li><p>Getting insights to help you make decisions</p><p>Brings only a fraction of value</p></li><li><p>The true economic value of Big Data</p><p>Using machine learning to automate decision making</p></li><li><p>Its true that change is coming (and data are generated) soquickly that human-in-the-loopinvolvement in all decision making is rapidly becoming impractical. Looking three to five years out, we expect to see far higher levels of artificial intelligence... </p><p>McKinsey&amp;Company, Anexecutivesguidetomachinelearning, June 2015</p></li><li><p>Robotsarecausinga new IndustrialRevolution.Similartowhathappenedtofarming,70percent(ormore)ofcurrentjobswillbereplaced bymachines. Replacementbyrobotsinmostjobsisjusta matteroftime. </p><p>American Thinker, TheNextPhaseoftheIndustrial Revolution, June2015</p></li><li><p>The robots might be coming for your job, even if you think it seems safe.</p><p>Business Insider, Jobs replaced by robots, June 2015</p></li><li><p>Deep Blue </p></li><li><p>Self driving car</p><p>Author: smoothgroover22 by CC BY 4.0</p><p></p></li><li><p>21</p></li><li><p>8</p><p>54,548</p><p>53,966</p><p>51,108</p><p>47,815</p><p>28,152</p><p>25,900</p><p>25,185</p><p>22,940</p><p>19,043</p><p>16,751</p><p>Top 10 Newspapers by Digital Traffic</p><p></p><p>Total number of unique visitors for January 2015 (in thousands)</p><p></p><p></p><p></p><p></p><p></p><p></p><p></p><p></p><p></p><p>1</p><p>2</p><p>3</p><p>4</p><p>5</p><p>6</p><p>7</p><p>9</p><p>10</p></li><li><p>8</p><p>54,548</p><p>53,966</p><p>51,108</p><p>47,815</p><p>28,152</p><p>25,900</p><p>25,185</p><p>22,940</p><p>19,043</p><p>Top 10 Newspapers by Digital Traffic</p><p></p><p>Total number of unique visitors for January 2015 (in thousands)</p><p></p><p></p><p></p><p></p><p></p><p></p><p></p><p></p><p>Yandex.News</p><p>1</p><p>2</p><p>3</p><p>4</p><p>5</p><p>6</p><p>7</p><p>9</p><p>10</p><p>23,164</p></li><li><p>23 000 000 monthly readers</p></li><li><p>0 editorial team</p></li><li><p>Online Gaming: Loyalty Management &amp; Personalisation </p><p>Predicting &amp; Preventing User Churn</p><p>Gamer data (number of victories, battles, purchase logs etc.) External data (e.g. weather)</p><p>Indication of potential churner Recommendation of the best retention offer </p></li><li><p>Retail Business: Next Best Offer</p><p>Upsell recommendations for a Bank </p><p>Customer profiles Historical data on communications and responses </p><p>13% increase in NPV</p></li><li><p>Anomaly detection for CERN (LHCb)</p><p>Up to 30 mln collisions per second terabytes of data per second several thousand of parameters to check</p><p>Industry and infrastructure: Predictive maintenance</p></li><li><p> IBM</p><p>Its estimated that 90 percent of the data in the world today has been created in the last two years alone. </p></li><li><p>Data quickly becomes commodity </p></li><li><p>Heres our prediction</p><p>In all the business processes, where: </p><p> we know exactly what we want to improve and can measure it </p><p> we have enough data we can experiment we can take automated action </p><p>In 10 years, well have algorithms doing the work.</p></li><li><p>Whats your plan?</p><p>Not big data strategy, but ML strategyBig data by itself means costs, not value </p><p>Value first: start with a few short projects todayTheres no value without implementation </p><p>Experiment and measurement is the key Continuous experimenting is the only way to stay on top</p><p>32</p></li><li><p>McKinsey&amp;Company, Anexecutivesguide tomachinelearning, June 2015</p><p>...Becausemachinelearningsemergence asamainstream managementtoolisrelativelyrecent,itoftenraisesquestions</p></li><li><p>Jane Zavalishina </p><p>CEO Yandex Data Factory</p><p>Happy to answer your questions!</p><p></p><p></p></li><li><p>Yandex Data Factory</p><p>Created in 2014 </p><p>Apply Yandexs machine learning expertise to other industries </p><p>Computational infrastructure </p><p>Proprietary machine learning tools </p><p>Data scientists</p></li><li><p>Title: Open Sans 100 px </p><p> Subtitle:OpenSans48px</p></li></ul>