Upgrade to Pro — share decks privately, control downloads, hide ads and more …

Intro to AI - HandsOn

Intro to AI - HandsOn

Introduction course on Artificial Intelligence and Neural Networks with useful links to getting started programming on Google Colab or Kaggle

Aletheia

April 06, 2019
Tweet

More Decks by Aletheia

Other Decks in Technology

Transcript

  1. 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.
  2. Luca Bianchi Who am I? Chief Technology Officer @ Neosperience

    Stuff that makes me happy: • Discussing about Software Architectures • Spreading Serverless verb • Developing on Blockchain technologies • Implementing Neural Networks github.com/aletheia @bianchiluca https://it.linkedin.com/in/lucabianchipavia https://speakerdeck.com/aletheia
  3. 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
  4. 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
  5. Artificial Intelligence “the theory and development of computer systems able

    to perform tasks normally requiring human intelligence, such as visual perception, speech recognition, decision-making, and translation between languages.”
  6. Tribes of Artificial Intelligence for decades individual “tribes” of artificial

    intelligence researchers have vied one another for dominance. Is the time now for tribes to collaborate? They may be forced to, as collaboration and algorithm blending are the only ways to reach true AGI. What are the five Tribes? Symbolists Use symbols, rules, and logic to represent knowledge and draw logical inference Favored algorithm Rules and decision trees Bayesians Assess the likelihood of occurrence for probabilistic inference Favored algorithm Naive Bayes or Markov Connectionists Recognise and generalise patterns dynamically with matrices of probabilistic weighted neurons Favored algorithm Neural Networks Evolutionaries Generate variations and then assess the fitness of each for a given purpose Favored algorithm Genetic Programs Analogizers Optimize a function in light of constraints (“going as high as you can while staying on the road”) Favored algorithm Support vectors
  7. Consider computing flow An operative definition Deterministic computing Deterministic computing

    Computer algorithm data output Learner data output (e) algorithm
  8. Machine Learning — Taxonomy Input-based taxonomy • Supervised Learning •

    Unsupervised Learning • Reinforcement Learning Types of Machine Learning Machine learning tasks are typically classified into three broad categories, depending on the nature of the learning "signal" or "feedback" available to a learning system. Output-based taxonomy • Regression
 • Classification
 • Clustering
 • Density estimation
 • Dimensionality reduction
  9. Regression Regression analysis helps one understand how the typical value

    of the dependent variable (or 'criterion variable') changes when any one of the independent variables is varied, while the other independent variables are held fixed. Is a statistical method of data analysis. The most common algorithm least square method that provides an estimation of regression parameters. When dataset is not trivial estimation is achieved through is gradient descent.
  10. Regression — Use cases Common scenarios •Stock price value •Product

    Price Estimation •Age estimation •Customer satisfaction rate defining variables such as response-time, resolution-ration we can forecast satisfaction level or churn •Customer Conversion rate estimation (based on click data, origin, timestamp, ...) Statistical regression is used to make predictions about data, filling the gaps Regression, even in the most simple form of Linear Regression is a good tool to learn from data and make predictions based on data trend.
  11. Classification Classification is the problem of identifying to which of

    a set of categories (sub-populations) a new observation belongs, on the basis of a training set of data containing observations (or instances) whose category membership is known. Most used algorithms for classification are: • Logit Regression • Decision Trees • Random Forest
  12. Classification — Use cases Common scenarios •Credit scoring •Human Activity

    Recognition •Spam/Not Spam classification •Customer conversion prediction •Customer churn prediction •Customer personas classification Classification is used to detect the binary outcome of a variable Classification is often used to classify people into pre-defined clusters (good-payer/bad-payer, in/out target, etc.)
  13. Clustering is the task of grouping a set of objects

    in such a way that objects in the same group (called a cluster) are more similar (in some sense or another) to each other than to those in other groups (clusters). The difference between algorithms is due to the similarity function that is used: • Centroid based clusters • Density based cluster • Random Forest
  14. Clustering — Use cases Common scenarios •Similar interests recognition •Shape

    detection •Similarity analysis •Customer base segmentation Clustering is used to segment data Clustering labels each sample with a name representing its belonging cluster. Labelling can be exclusive or multiple. Clusters are dynamic structures: they adapt to new sample coming into the model as soon as thy label them.
  15. Convergence happened in 2012 Why is happening now? Back to

    2012 we had enough data to be processed representing dataset for learning, fast enough GPUs and G. Hinton solved a tedious issue with Deep Neural Networks making them suitable with many hidden layers Geoffrey Hinton in 2006 proposed a solution to the vanishing gradient and adoption of ReLU activation functions
  16. Practical applications Neural Networks - Use Cases • Voice Search/Recognition:

    Automotive, Messaging, IoT • Sentiment Analysis: CRM • Text Recognition/Translation, Language Detection, Text Summarization : Messaging, Personalizaiton, Profile analysis • Image Recognition, Classification, Motion Detection: Computer Vision, Image Classification, Face Detection, Face Tracking • Time Series: behavioural analysis, Pattern matching, Trading • Video detection/Object Tracking: Vehicle Tracking, Path detection, Object Classification, Face Recognition
  17. Data pre-processing is the key Training a Machine Learning Model

    Data must be prepared before being feeded into the model for training. To an abstract perspective, data must be reduced to a single value for each sample (in each dimension), made available to NN processing. Data pre-processing (filtering) is a key moment in every Machine/Deep Learning application since can have an impact on performance and (if not done correctly) could introduce some bias. (i.e. filtering out relevant dimensions thus making the model less accurate)
 Data filtering is a critical step which complexity can tackle the result of the whole process.
  18. Neural Networks are a type of Deep Learning. Neural Networks

    - Classification Neural Networks can be used either for supervised, unsupervised or reinforcement learning, they are the most known expression of Deep Learning (even if not the only one). Neural Networks can be classified on signal propagation method: • Convolutionary Neural Networks (CNN): are the most promising networks, with a wide range of application from Computer Vision to NLP • Recurrent Neural Networks (RNN): used for language translation, content based image processing, word prediction, Language Detection, Text Summarization • Feed-Forward Neural Network (FFNN): used for standard prediction (a powerful alternative to regression/ classification)
  19. Convolution 19 CONVOLUTION 0 0 0 0 0 0 0

    0 1 1 1 0 0 0 0 1 2 2 1 1 1 0 1 2 2 2 1 1 0 1 2 2 2 1 1 0 0 1 1 1 1 1 0 0 0 0 0 0 0 4 0 0 0 0 0 0 0 -4 1 0 -8 Source Pixel Convolution kernel New pixel value (destination pixel) Centre element of the kernel is placed over the source pixel. The source pixel is then replaced with a weighted sum of itself and nearby pixels. http://www.biomachina
  20. Convolutions are great matematical operations to exploit pixel correlations Convolutional

    Neural Network (CNN) 20 Image “Volvo XC90” Image source: “Unsupervised Learning of Hierarchical Representations with Convolutional Deep Belief Networks” ICML 2009 & Comm. ACM 2011. Honglak Lee, Roger Grosse, Rajesh Ranganath, and Andrew Ng. CONVOLUTIONAL NEURAL NETWORKS
  21. Recurring Neural Network (RNN) 24 RNN Miro Enev, NVIDIA Recurring

    Networks are suitable to identify sequences
  22. Allows formvery long networks, to analyze temporal data series Long

    Short Term Memory Network (LSTM) 25 Long short-term memory (LSTM) Hochreiter (1991) analysed vanishing gradient “LSTM falls out of this almost naturally” Gates control importance of the corresponding activations Training via backprop unfolded in time LSTM: input gate output gate Long time dependencies are preserved until input gate is closed (-) and forget gate is open (O) forget gate Fig from Vinyals et al, Google April 2015 NIC Generator Fig from Graves, Schmidhuber et al, Supervised Sequence Labelling with RNNs
  23. Forcing a network across a bottleneck is great for dimensonal

    reduction Autoencoder Network 22 EXAMPLE: AUTOENCODERS Feature Learning Sparse Representation Training Data Reconstruction Encoder Decoder Minimize Reconstruction Error - Input Layer Hidden Layer Hidden Layer Hidden Layer Bottleneck Layer Hidden Layer Hidden Layer Hidden Layer Output Layer
  24. Fields of application Push Messaging A large hotel chain wanted

    to solve this problem: 90k travelers are stranded every day in America across 5,145 airports. How can the hotel ensure that they show up at the right moment for all these people? The solution was to leverage real-time signals like bad weather, flight delays at 5,145 airports, and other such data, combine that with ML powered algorithms to automate ads and messaging in the proximity of local airports. All sans human-control. Result? 60% increase in bookings in targeted areas. ML + Automation = Profit. Marketing budget forecast leverage ML to do the above campaign across Search, Display and Video. AND (are you sitting down?) rather than guessing how much credit to give each marketing strategy from hundreds of thousands of customer touch points, ML powered attribution can magically analyze every individual’s journey and automatically recommend shifts to your budget.
  25. Fields of application Customer behaviour patterns Leverage pattern recognition to

    understand common interests between customer belonging to different clusters and push personalized messages. Generate smart content Smart Agents such as Persado generate personalized wording depending on the profile of customer landing on their site, starting from a few words. Customer tracking in store Customers can be tracked down extracting their movements in store. This leads to exploiting their interests, identifying returning customers (through face recognition) and sentiment (through face analysis). i.e. tracking person of interests with AWS Rekognition
  26. Max-Q operating mode (< 7.5 watts) delivers up to 2x

    energy efficiency vs. Jetson TX1 maximum performance Max- P operating mode (< 15 watts) delivers up to 2x performance vs. Jetson TX1 maximum performance NVIDIA JETSON TX2 EMBEDDED AI SUPERCOMPUTER Advanced AI at the edge JetPack SDK < 7.5 watts full module Up to 2X performance or 2X energy efficiency
  27. AI Next Talking about AI is not differentiating anymore A

    lot of companies are moving into this space with specific focus on markets
  28. Where are we going Looking for AI Graal is an

    algorithm capable of solving a narrow problem with no need to be developed specifically for a given context. The Master Algorithm The master algorithm is the ultimate learning algorithm. It's an algorithm that can learn anything from data and it's the holy grail of machine learning. It's what we're working towards and the book is about both how we're going to get there and the implications that it's going to have for life. Also, the impact that machine learning is already having.
  29. Learning Machine Learning Where to go from here? An Introduction

    to Statistical Learning 
 http://www-bcf.usc.edu/~gareth/ISL/ Artificial Intelligence for Humans
 http://www.heatonresearch.com/aifh/ Course on Fast.ai
 http://course.fast.ai Google Colab
 http://colab.research.google.com Google Seedbank
 https://research.google.com/seedbank/ Kaggle
 http://course.fast.ai
  30. Predict who survived the Titanic disaster Titanic: Machine Learning from

    Disaster The sinking of the RMS Titanic is one of the most infamous shipwrecks in history. On April 15, 1912, during her maiden voyage, the Titanic sank after colliding with an iceberg, killing 1502 out of 2224 passengers and crew. This sensational tragedy shocked the international community and led to better safety regulations for ships. One of the reasons that the shipwreck led to such loss of life was that there were not enough lifeboats for the passengers and crew. Although there was some element of luck involved in surviving the sinking, some groups of people were more likely to survive than others, such as women, children, and the upper-class. In this challenge, we ask you to complete the analysis of what sorts of people were likely to survive. In particular, we ask you to apply the tools of machine learning to predict which passengers survived the tragedy. https://www.kaggle.com/c/titanic
  31. What to expect from AI Additional Resources Immortality or Extinction


    https://waitbutwhy.com/2015/01/artificial-intelligence-revolution-1.html The Hyperion Cycle Dan Simmons