LE NUOVE FRONTIERE DALL AI ALL AR E L IMPATTO SULLA QUOTIDIANITÀ Deloitte Analytics & Information Management Torino, 26/03/2018 1
AUGMENTED REALITY FUNDAMENTALS EXAMPLES OF DEEP LEARNING ARTIFICIAL INTELLIGENCE It is man's attempt to design and build an artificial computer capable of thinking and acting like a human being. MACHINE LEARNING It represents a set of methods developed over the last few decades, which share the ability to learn from a set of data and make predictions. DEEP LEARNING It is a particular field of research, belonging to Machine Learning, which fully exploits the concept of Artificial Neural Networks in order to process information. 1950s 1960s 1970s 1980s 1990s 2000s 2010s Deep Learning Focus Theoretical results suggest that in order to replicate complicated functions that can represent high-level abstractions (e.g. in vision, language, and other AI-level tasks), deep architectures become mandatory. These are composed of multiple levels of non-linear operations, such as in neural nets with many hidden layers or in complicated propositional formulae that leverage a big set of sub-formulae. Searching the parameter space of deep architectures is a difficult task, but learning algorithms such as those for Deep Belief Networks have recently been proposed to tackle this problem with notable success. Natural Language Processing It is the process of automatic processing of information written or spoken in a natural language. Computer Vision It is the set of processes that allow a machine to observe and then process the information included in an image or video, giving it a definite "recognition" capability. Speech Recognition It is the process through a machine recognizes and then processes human oral language. 2
WHAT IS AUGMENTED REALITY? AUGMENTED REALITY It is the overlay of computer generated digital contents to real-world environment, altering it. The overlaid information can be constructive, enriching the natural environment, or destructive, masking one or more objects. It s different from Virtual Reality, which completely replaces the real world with a virtual one It s real-time Deloitte predicts that over a billion smartphones users will create augmented reality (AR) content at least in 2018 Pokèmon Go 3
DEVICE PREFERENCE FOR VARIOUS ACTIVITIES Total Male Female 18-24 25-34 35-44 45-54 55-64 65+ Make online purchases Online searches Watch short videos Check bank balances Video calls Check social networks Read the news Play games Voice calls (VoIP) Take photos Record videos Stream films and/or TV series Watch live TV Mobile phone Tablet Desktop Console Laptop Television Weighted base: smartphone owners in 16 developed markets (22.929 respondents). The figure is the average of 16 countries in our study, namely Australia, Belgium, Canada, Denmark, Finland, Germany, Ireland, Italy, Japan, Luxembourg, Netherlands, Norway, Spain, Sweden, the UK and the USA. Source: Deloitte s Global Mobile Consumer Survey, developed countries, May-July 2017 4
AI AND SMARTPHONES Awareness and usage of application featuring ML (developed markets) Predictive text Route suggestion Voice assistant Voice search Automated news or information updates Translation apps Voice-to-text Email classification Automated calendar entries Location based app suggestions Automated photo classification App suggestions Travel pop-ups Any of these 5% 4% 6% 12% 9% 13% 11% 7% 12% 10% 9% 15% 13% 24% 27% 25% 25% 23% 21% 20% 20% 31% 35% 37% 39% 50% 65% 79% 0% 10% 20% 30% 40% 50% 60% 70% 80% 90% Awareness Weighted base: smartphone owners in 16 developed markets (24.563 respondents). The figure is the average of 16 countries in our study, namely Australia, Belgium, Canada, Denmark, Finland, Germany, Ireland, Italy, Japan, Luxembourg, the Nederland's, Norway, Spain, Sweden, the UK and the USA. Source: Deloitte s Global Mobile Consumer Survey, developed countries, May-July 2017 Usage 5
SMARTPHONE DEVELOPMENT WILL STRENGTHEN AUGMENTED REALITY ADOPTION 1 2 3 4 5 Over the past three years, AR has become an increasing popular smartphone application, often for entertainment applications Mobile phones will be increasingly equipped with dedicated OS AR framework, Visual Inertial Odometer systems (VIOs) Dedicated frameworks within standard OS, lowers the cost of developing AR apps and this should increase the supply of apps embedding this kind of feature HW manufacturer have improved the precision of their latest chips, allowing camera and inertial measurement unit (IMU) to work closely together Algorithms are also critical to creating and displaying compelling AR content 6
USE CASE OVERVIEW WHAT WHERE WHY WHEN The app enables people to take photos with mobile devices, recognize the garment, identify similar items available and related product information (characteristics) Eligible for Fashion & Luxury Market, since it s difficult to describe a style just with words Modern shoppers get inspired on Instagram, Pinterest and other social networks, want to shop specific styles, and require personalized recommendations and services, without wasting time Now. We live in digital revolution in which our communication is based on images. Search is going to be about pictures instead of keywords (65% of the population consists of visual learners) It is a trusted app to choose the best outfit with custom searches based on product, price, retailer, category and more 7
MANAGED DATA STOCK KEEPING UNIT (SKU) STYLE & OUTFITS INFLUENCERS & TESTIMONIALS TRENDS ON SOCIAL MEDIA Detailed characteristics of the product such as model, collection, color, material, etc. Suggestions of items belonging to the same category or products frequently bought together Related advertisement material and video from fashion shows, testimonials and influencers Trends and reviews of the product on social media such as Instagram, Facebook, Pinterest, etc. STORES CUSTOMER DATA ACTIVE PROMOTIONS INVENTORY Price and stores where the product is sold (both online and offline) Demographic data and purchasing behavior of the customer (e.g. average spending, preferred product categories, etc.) Information on ongoing promotions and special offers related to a specific product Information on the product availability of physical and online shops 8
HOW IT ENRICHES THE OVERALL CUSTOMER EXPERIENCE PRODUCT DISCOVERY PHOTO IMAGE RECOGNITION The app recognizes the product shown in the picture Gain information on product characteristics (model, material, color, ) STYLE & OUTFITS Explore related products and recommended outfits based on your previous purchasing behavior SOCIAL TRENDS Take a look at the reviews from testimonials, influencers and explore current trends on social media PRICING Take advantage of active promotions and special offers to upsell or buy related products SHOP Shop at the closest shop or order your favorite products online STORE LOCATOR Locate in real time the nearest physical shop or go directly to the e-commerce platform 9
EXAMPLE OF LESSON LEARNED INCORRECT CLASSIFICATION Image classified as BOOTS Image classified as SHOES Image classified as SANDALS The noise introduced by the presence of the background and the additional details of the environment reduce the classifier performances. 10
EXAMPLE OF LESSON LEARNED APPROACH TECHNICAL SOLUTION Train model Increase the training set manipulating the original images to create different variations Image Manipulation Sub-Module This facility is part of the Engine module included in the Web Service of our system. It is in charge of preprocessing all the images received through the Rest infrastructure before sending them to the Classifier. Image Manipulation Steps: Original image Remove Background noise Improve Color and Contrast Resize image to standard dimension Tech: Python, the Pillow package Reduce noise Apply filters to reduce noise and improve the overall quality of the image Dataset enrichment This task has been included to improve the quality of the training set for the neural network. It manipulates the original images and creates alternative views of them to be included in the training set. Training Set enrichment steps: Original image Image.rotate(rotation degrees) Image.convert(mode=mode) Image.thumbnail(dimension) Image.filter(ImageFilter.mode) Tech: Python, the Pillow package The pre-processing performed by the Image Manipulation Sub-Module and the Dataset enrichment allows for an increase of the confidence level also in the case of images in which the footwear is not the only subject. 11
Deloitte refers to one or more of Deloitte Touche Tohmatsu Limited, a UK private company limited by guarantee, and its network of member firms, each of which is a legally separate and independent entity. Please see www.deloitte.com/about for a detailed description of the legal structure of Deloitte Touche Tohmatsu Limited and its member firms. Deloitte provides audit, tax, consulting and financial advisory services to public and private clients spanning multiple industries. With a globally connected network of member firms in more than 150 countries, Deloitte brings world-class capabilities and high-quality service to clients, delivering the insights they need to address their most complex business challenges. The more than 200 000 professionals of Deloitte are committed to becoming the standard of excellence. This communication contains general information only, and none of Deloitte Touche Tohmatsu Limited, its member firms, or their related entities (collectively, the Deloitte Network ) is, by means of this communication, rendering professional advice or services. No entity in the Deloitte Network shall be responsible for any loss whatsoever sustained by any person who relies on this communication. 12