Serverless Machine Learning

Discover how Neosperience Cloud implements Deep Learning inference through serverless

Transcript

1. www.neosperience.com | blog.neosperience.com | info@neosperience.com Serverless Machine Learning un'architettura per l'implementazione di modelli su architetture ibride March 28th, 2019
2. Safe Harbor Statement Certain information set forth in this presentation contains “forward-looking information”, including “future oriented financial information” and “financial outlook”, under applicable securities laws (collectively referred to herein as forward-looking statements). Except for statements of historical fact, information contained herein constitutes forward-looking statements and includes, but is not limited to, the (i) projected financial performance of the Company; (ii) completion of, and the use of proceeds from, the sale of the shares being offered hereunder; (iii) the expected development of the Company’s business, projects and joint ventures; (iv) execution of the Company’s vision and growth strategy, including with respect to future M&A activity and global growth; (v) sources and availability of third-party financing for the Company’s projects; (vi) completion of the Company’s projects that are currently underway, in development or otherwise under consideration; (vi) renewal of the Company’s current customer, supplier and other material agreements; and (vii) future liquidity, working capital, and capital requirements. Forward-looking statements are provided to allow potential investors the opportunity to understand management’s beliefs and opinions in respect of the future so that they may use such beliefs and opinions as one factor in evaluating an investment. These statements are not guarantees of future performance and undue reliance should not be placed on them. Such forward-looking statements necessarily involve known and unknown risks and uncertainties, which may cause actual performance and financial results in future periods to differ materially from any projections of future performance or result expressed or implied by such forward-looking statements. Although forward-looking statements contained in this presentation are based upon what management of the Company believes are reasonable assumptions, there can be no assurance that forward-looking statements will prove to be accurate, as actual results and future events could differ materially from those anticipated in such statements. The Company undertakes no obligation to update forward-looking statements if circumstances or management’s estimates or opinions should change except as required by applicable securities laws. The reader is cautioned not to place undue reliance on forward-looking statements.
3. Luca Bianchi Who am I? Chief Technology Officer @ Neosperience github.com/aletheia @bianchiluca https://it.linkedin.com/in/lucabianchipavia Stuff that makes me happy: • Discussing about Software Architectures • Spreading Serverless verb • Developing on Blockchain technologies • Implementing Neural Networks
4. Join the (r)evolution! Interested in serverless and living in Italy? 844 members Monthly meetups Serverless OnStage and OnTheRoad Join the group, attend meetups, 
 tell your story June, 21st 2019 Grand Visconti Palace @ Milano CFP deadline April, 21st http://milan.serverlessdays.io http://bit.ly/sls-meetup
5. Neosperience Cloud allows to create personalized, relevant experiences that strengthen 
 the relationship with the customer across touchpoints: web, app, platforms, point of sale How Neosperience Cloud delivers digital experience innovation The first digital experience platform to establish empathic relationships with customers that takes into account their uniqueness. A set of application modules condensing multi-disciplinary skills: data scientists, designers, software architects, cognitive, behavioral and social psychologists, to unleash your brand’s potential. Increase customer engagement • Tailor storytelling and call-to-action • Grow the value of the customer • Suggest the most suitable products and services • Accelerate on-boarding and increase conversions • Generate recurring revenues, evolving loyalty into membership • Send personalized notifications • Delight the customer with gamification • Make digital experiences come alive in extended reality • Nudge advocacy 01 Understand Listen to customers
 across channels 02 Engage Deliver relevant
 experiences at scale 03 Grow Transform prospects
 into customers for life
6. Neosperience Cloud Cloud Customer Experience Customer Value Customer Profile Business User Architects and Developers Creatives and Designers Engage Grow Nudging, Gamification, Loyalty Personalized
 Content Personalized
 Advertising App Memorability Web Conversational
 Interface Neosperience Cloud Console API and Integration 3rd Parties Service Connectors Layer Image Conversation Relation Interaction Understand Behavior Proximity Marketing CRM &
 Customer Service Video Personalized
 Commerce Extended
 Reality E-Commerce
7. Image Memorability Use machine learning to predict memorability score of an image and detect which features are relevant.
8. Neosperience Cloud Cloud Customer Experience Customer Value Customer Profile Business User Architects and Developers Creatives and Designers Engage Grow Nudging, Gamification, Loyalty Personalized
 Content Personalized
 Advertising App Memorability Web Conversational
 Interface Neosperience Cloud Console API and Integration 3rd Parties Service Connectors Layer Image Conversation Relation Interaction Understand Behavior Proximity Marketing CRM &
 Customer Service Video Personalized
 Commerce Extended
 Reality E-Commerce
9. Why choosing memorable images? Image Memorability — A business perspective • Brands spend a huge amount of money in marketing campaigns to increase conversion • Most of today campaigns are visual messages • Gaining people attention is vital to campaign success (the so-called “battle for attention”) • Good images make customers remember a brand • Memorable brands emerge from the crowd
10. What is memorability score? Image Memorability — A business perspective • Memorability is a measure of how much an image sticks into the memory of an average customer respect to average baseline images • A memorability score is a number representing memorability of an image, compared to the average capability of a human to remember an image which is 0.72 • Images with a score higher than 0.72 have high memorability and are suitable for campaigns • Images with a score lower than 0.72 underperform and should be avoided because are not remembered memorability score: 0.8 memorability score: 0.79
11. A memorable image is a good image? Image Memorability — A business perspective • High memorability score is a good starting point, but using it to select an image could be too naive • More relevant than memorability itself is understanding which feature makes an image memorable • Assigning a score to each pixel of the image regarding its contribution to the resulting score • In this case memorability analysis outperforms humans because it is able not only to tell the score • but also to understand what makes this score
12. How to detect scores and heat maps? Image Memorability — A technical perspective • Build an experiment to measure memorability (ground truth) • Deep Learning comes into help with CNNs • A CNN learns from experiment dataset how to estimate a memorability score • From a given inference, finding layer activations (through back propagation) • Convolutions and back propagation are compute intensive tasks that require GPUs even with inference • GPU inference is achieved through DeepLearning AMIs and on-premise instances • We needed an architecture to support inference through GPU in production in a scalable and cost effective way
14. Lessons learned Conclusions • Lambda functions can be effective also when dealing with servers to improve scalability • Tasks are balanced through instances using DynamoDB Tables as a registry and Lambda to manage dispatching • OnPrem instances open support to edge deployment of Memorability model • Elastic GPUs support and Lambda Layers could improve architecture • We’re just scratching the surface of Deep Learning implementations
15. Thank you. http://bit.ly/nsp-sls-ml we’re hiring future@neosperience.com
16. www.neosperience.com | blog.neosperience.com | info@neosperience.com