Introduction to Machine Learning Typical ML task : Linear Regression Typical ML task : Classification Fundamentals Foundation of Machine Learning and Deep learning Anthony FAUSTINE [email protected] 05th OCTOBER 2017 Anthony FAUSTINE [email protected] pythontz 05th OCTOBER 2017 1 / 55 

Introduction to Machine Learning Typical ML task : Linear Regression Typical ML task : Classification Fundamentals Learning goal • Understand the basics of Machine learning and deep learning. • Explore opportunities and research direction in machine learning (ML) and AI. • Understand different python libararies for doing ML and AI research and development. Anthony FAUSTINE [email protected] pythontz 05th OCTOBER 2017 2 / 55 

Introduction to Machine Learning Typical ML task : Linear Regression Typical ML task : Classification Fundamentals Presenter Bio • PhD Scholar at Nelson Mandela African Institution of Science and Technology, • Research : Applied machine learning and signal processing for computational sustainability. • Probabilistic-deep learning algorithms (Hybrid HMM-DNN) for energy dis-aggregation problem. • Unsupervised deep learning. • Bayesian reasoning. • Co-founder Pythontz, Awesome-tech (First ML and AI startup in Tz). • Assistant Lecturer (UDOM), Research and Consultant (Machine Intelligence and Data Science). • Contact : sambaiga.github.io, [email protected], @sambaiga Anthony FAUSTINE [email protected] pythontz 05th OCTOBER 2017 3 / 55 

Introduction to Machine Learning Typical ML task : Linear Regression Typical ML task : Classification Fundamentals Pythontz Anthony FAUSTINE [email protected] pythontz 05th OCTOBER 2017 4 / 55 

Introduction to Machine Learning Typical ML task : Linear Regression Typical ML task : Classification Fundamentals Pythontz About Pythontz • A postive peer learning community for interested Python users in Tanzania. Vision • To create a vibrant and diverse python community in Tanzania. Mission • To foster the application of python programming across industries, learning centers, schools and community in Tanzania. • Focus : Web-development, data science, Machine learning and Artificial Intelligence. Anthony FAUSTINE [email protected] pythontz 05th OCTOBER 2017 5 / 55 

Introduction to Machine Learning Typical ML task : Linear Regression Typical ML task : Classification Fundamentals Awesome-Tech About Awesometech • We are all about continuous innovation, making ideas happen and crafting amazing products. Vision • We focus on the development and deployment of data driven application and systems using Machine Learning and Artificial Intelligence techniques. Mission • To unleash the power of machine learning and artificial intelligence in solving real-community problems. Anthony FAUSTINE [email protected] pythontz 05th OCTOBER 2017 6 / 55 

Introduction to Machine Learning Typical ML task : Linear Regression Typical ML task : Classification Fundamentals Outline 1 Introduction to Machine Learning 2 Typical ML task : Linear Regression 3 Typical ML task : Classification 4 Fundamentals of Deep Learning 5 Research direction and opportunities 6 Python libraries for ML and AI Anthony FAUSTINE [email protected] pythontz 05th OCTOBER 2017 7 / 55 

Introduction to Machine Learning Typical ML task : Linear Regression Typical ML task : Classification Fundamentals Introduction Data Science ? The future belongs to the companies and people that turn data into products. By Mike Loukides June 2, 2010 Data science : deals with analyzing and manipulating data to derive insights and build data products. • Applies machine learning to create data products Data product : any tool created with the help of data to make a more informed decision. • The end goal of DS ⇒ data product. • Data Science is the real-world application of machine learning, with the goal of creating data products. Anthony FAUSTINE [email protected] pythontz 05th OCTOBER 2017 8 / 55 

Introduction to Machine Learning Typical ML task : Linear Regression Typical ML task : Classification Fundamentals Introduction Data Science ? The future belongs to the companies and people that turn data into products. By Mike Loukides June 2, 2010 Data science : deals with analyzing and manipulating data to derive insights and build data products. • Applies machine learning to create data products Data product : any tool created with the help of data to make a more informed decision. • The end goal of DS ⇒ data product. • Data Science is the real-world application of machine learning, with the goal of creating data products. Anthony FAUSTINE [email protected] pythontz 05th OCTOBER 2017 8 / 55 

Introduction to Machine Learning Typical ML task : Linear Regression Typical ML task : Classification Fundamentals Machine learning Machine learning (ML) : a set of algorithms that automatically detect patterns in data and use the uncovered pattern to make inferences or predictions. ML is a subfield of AI ⇒ aims to enable computers to learn on their own. ML algorithms : 1 Identify patterns in observed data. 2 Build models that explain the world. 3 Predict or do inference. Anthony FAUSTINE [email protected] pythontz 05th OCTOBER 2017 9 / 55 

Introduction to Machine Learning Typical ML task : Linear Regression Typical ML task : Classification Fundamentals Machine learning Aplication • It is an exciting and fast-moving field of computer science with many recent applications. • Computer vision : Object Classification in Photograph, image captioning. • Speech recognition, Automatic Machine Translation, • Communication systems • Robots learning complex behaviors : • Recommendations services ML use cases Anthony FAUSTINE [email protected] pythontz 05th OCTOBER 2017 10 / 55 

Introduction to Machine Learning Typical ML task : Linear Regression Typical ML task : Classification Fundamentals Why Machine learning ? • Hard problems in high dimensions, like many modern CV or NLP problems require complex models ⇒ difcult to program the correct behavior by hand. • Machines can discover hidden, non-obvious patterns. • A system might need to adapt to a changing environment. • A learning algorithm might be able to perform better than its human programmers. Anthony FAUSTINE [email protected] pythontz 05th OCTOBER 2017 11 / 55 

Introduction to Machine Learning Typical ML task : Linear Regression Typical ML task : Classification Fundamentals Machine learning types Machine learning is usually divide into three major types : 1 Supervised Learning • Learn a model from a given set of input-output pairs, in order to predict the output of new inputs. • Further grouped into Regression and classification problems. 2 Unsupervised Learning • Discover patterns and learn the structure of unlabelled data. • Example Distribution modeling and Clustering. 3 Reiforcement Learning • Learn what actions to take in a given situation, based on rewards and penalties. More details on RL • Example consider teaching a dog a new trick : you cannot tell it what to do, but you can reward/punish it. Anthony FAUSTINE [email protected] pythontz 05th OCTOBER 2017 12 / 55 

Introduction to Machine Learning Typical ML task : Linear Regression Typical ML task : Classification Fundamentals Outline 1 Introduction to Machine Learning 2 Typical ML task : Linear Regression 3 Typical ML task : Classification 4 Fundamentals of Deep Learning 5 Research direction and opportunities 6 Python libraries for ML and AI Anthony FAUSTINE [email protected] pythontz 05th OCTOBER 2017 13 / 55 

Introduction to Machine Learning Typical ML task : Linear Regression Typical ML task : Classification Fundamentals Linear Regression In regression : predict a scalar-valued target, such as the price of stock. • The target is predicted as a linear function of the inputs. Example applications : 1 weather forecasting. 2 house pricing prediction. 3 student performance prediction. Anthony FAUSTINE [email protected] pythontz 05th OCTOBER 2017 14 / 55 

Introduction to Machine Learning Typical ML task : Linear Regression Typical ML task : Classification Fundamentals Linear Regression : Formulate a learning problem To formulate ML problem mathematically, you need to define two things : 1 Model (Hypothesis) : set of allowable functions that compute predictions from the inputs • In linear regression, the model consists of linear functions given by : ˆ y = f (y, x) = j wj xj + b where w is the weights, and b is the bias. 2 Loss function : defines how well the model fit the data • How far off the prediction ˆ y is from the target y given as : L(ˆ y, y) = 1 2 (ˆ y − y)2 Anthony FAUSTINE [email protected] pythontz 05th OCTOBER 2017 15 / 55 

Introduction to Machine Learning Typical ML task : Linear Regression Typical ML task : Classification Fundamentals Formulate a learning problem : Cost Function Cost Function Jθ : Is the loss, averaged over all the training examples given by. Jθ = 1 N N i=1 L(ˆ y(i), y(i)) = 1 2N N i=1 (ˆ y(i) − y(i))2 = 1 2N N i=1   j wj x(i) j + b − y(i)   In vectorized form : Jθ = 1 2N ˆ y − y 2= 1 2N (ˆ y − y)T (ˆ y − y) where ˆ y = wTx Anthony FAUSTINE [email protected] pythontz 05th OCTOBER 2017 16 / 55 

Introduction to Machine Learning Typical ML task : Linear Regression Typical ML task : Classification Fundamentals Formulate a learning problem : Optimization Problem Combining the model and loss function, you get an optimization problem : Objective : minimize a cost function Jθ with respect to the model parameters θ(i.e. the w and b) • A popular minimization technique is gradient descent θt+1 = θt − α ∂Jθ ∂θ where α is the learning rate. Common issues to look out for : • Convergence to a local, non-global minimum. Anthony FAUSTINE [email protected] pythontz 05th OCTOBER 2017 17 / 55 

Introduction to Machine Learning Typical ML task : Linear Regression Typical ML task : Classification Fundamentals Linear Regression Use gradient descent to solve the minimum cost function Jθ θt+1 = θt − α ∂Jθ ∂θ For parameter w and b : wt+1 = wt − α ∂Jθ ∂w bt+1 = bt − α ∂Jθ ∂b where : ∂Jθ ∂w = 1 N xT(ˆ y − y) ∂Jθ ∂b = 1 N (ˆ y − y) Anthony FAUSTINE [email protected] pythontz 05th OCTOBER 2017 18 / 55 

Introduction to Machine Learning Typical ML task : Linear Regression Typical ML task : Classification Fundamentals Outline 1 Introduction to Machine Learning 2 Typical ML task : Linear Regression 3 Typical ML task : Classification 4 Fundamentals of Deep Learning 5 Research direction and opportunities 6 Python libraries for ML and AI Anthony FAUSTINE [email protected] pythontz 05th OCTOBER 2017 19 / 55 

Introduction to Machine Learning Typical ML task : Linear Regression Typical ML task : Classification Fundamentals Classification Goal is to learn a mapping from inputs x to target y such that y ∈ {1 . . . k} where k is the number of classes. • If k = 2, this is called binary classification (in which case we often assume y ∈ {0, 1} • If k > 2, this is called multiclass classification. Anthony FAUSTINE [email protected] pythontz 05th OCTOBER 2017 20 / 55 

Introduction to Machine Learning Typical ML task : Linear Regression Typical ML task : Classification Fundamentals Classification : Logistic regression Goal is to predict the binary target class y ∈ {0, 1}. Model is given by : ˆ y = σ(z) = 1 1 + e−z where z = wTx + b This function squashes the predictions to be between 0 and 1 such that : p(y = 1 | x, θ) = σ(z) and p(y = 0 | x, θ) = 1 − σ(z) Anthony FAUSTINE [email protected] pythontz 05th OCTOBER 2017 21 / 55 

Introduction to Machine Learning Typical ML task : Linear Regression Typical ML task : Classification Fundamentals Classification : Logistic regression Loss function : it is called crossentropy and defined as : LCE (ˆ y, y) = − log ˆ y if y = 1 − log(1 − ˆ y) if y = 0 • The crossentropy can be written in other form as : LCE (ˆ y, y) = −y log ˆ y − (1 − y) log(1 − ˆ y) • The cost function Jθ with respect to the model parameters θ is thus : Jθ = 1 N N i=1 LCE (ˆ y, y) = 1 N N i=1 −y(i) log ˆ y(i) − (1 − y(i)) log(1 − ˆ y(i)) Anthony FAUSTINE [email protected] pythontz 05th OCTOBER 2017 22 / 55 

Introduction to Machine Learning Typical ML task : Linear Regression Typical ML task : Classification Fundamentals Classification : Logistic regression Use gradient descent to solve the minimum cost function Jθ θt+1 = θt − α ∂Jθ ∂θ For parameter w and b : wt+1 = wt − α ∂Jθ ∂w bt+1 = bt − α ∂Jθ ∂b where : ∂Jθ ∂w = 1 N xT(ˆ y − y) ∂Jθ ∂b = 1 N (ˆ y − y) Anthony FAUSTINE [email protected] pythontz 05th OCTOBER 2017 23 / 55 

Introduction to Machine Learning Typical ML task : Linear Regression Typical ML task : Classification Fundamentals Multi-class Classification What about classification tasks with more than two categories ? • Targets form a discrete set {1, ..., K}. • It’s often more convenient to represent them as indicator vectors, or a one-of-K encoding : Anthony FAUSTINE [email protected] pythontz 05th OCTOBER 2017 24 / 55 

Introduction to Machine Learning Typical ML task : Linear Regression Typical ML task : Classification Fundamentals Multi-class Classification What about classification tasks with more than two categories ? • Targets form a discrete set {1, ..., K}. • It’s often more convenient to represent them as indicator vectors, or a one-of-K encoding : Anthony FAUSTINE [email protected] pythontz 05th OCTOBER 2017 24 / 55 

Introduction to Machine Learning Typical ML task : Linear Regression Typical ML task : Classification Fundamentals Multi-class Classification Model : softmax function ˆ yk = softmax(z1 . . . zk) = ezk k ezk where zk = j wkj xj + b Loss Function : cross-entropy for multiple-output case LCE (ˆ y, y) = − K k=1 yk log ˆ yk = −yT log ˆ y Anthony FAUSTINE [email protected] pythontz 05th OCTOBER 2017 25 / 55 

Introduction to Machine Learning Typical ML task : Linear Regression Typical ML task : Classification Fundamentals Multi-class Classification Cost funcion Jθ = 1 N N i=1 LCE (ˆ y, y) = −1 N N i=1 K k=1 yk log ˆ yk The gradient descent algorithm will be : wt+1 = wt − α ∂Jθ ∂w where ∂Jθ ∂w = 1 N xT(ˆ y − y) bt+1 = bt − α ∂Jθ ∂b where ∂Jθ ∂b = 1 N (ˆ y − y) Anthony FAUSTINE [email protected] pythontz 05th OCTOBER 2017 26 / 55 

Introduction to Machine Learning Typical ML task : Linear Regression Typical ML task : Classification Fundamentals Outline 1 Introduction to Machine Learning 2 Typical ML task : Linear Regression 3 Typical ML task : Classification 4 Fundamentals of Deep Learning 5 Research direction and opportunities 6 Python libraries for ML and AI Anthony FAUSTINE [email protected] pythontz 05th OCTOBER 2017 27 / 55 

Introduction to Machine Learning Typical ML task : Linear Regression Typical ML task : Classification Fundamentals What is Deep Learning Deep Learning a subclass of machine learning algorithms that : • use a cascade of multiple layers of nonlinear processing units for feature extraction and transformation. • flexible models with any input/output type and size These algorithms may be supervised or unsupervised. Anthony FAUSTINE [email protected] pythontz 05th OCTOBER 2017 28 / 55 

Introduction to Machine Learning Typical ML task : Linear Regression Typical ML task : Classification Fundamentals Deep Learning Success Speech Recognition Figure – https ://svail.github.io/mandarin/ Anthony FAUSTINE [email protected] pythontz 05th OCTOBER 2017 29 / 55 

Introduction to Machine Learning Typical ML task : Linear Regression Typical ML task : Classification Fundamentals Deep Learning Success Image caption Anthony FAUSTINE [email protected] pythontz 05th OCTOBER 2017 30 / 55 

Introduction to Machine Learning Typical ML task : Linear Regression Typical ML task : Classification Fundamentals Deep Learning Success Image painting Anthony FAUSTINE [email protected] pythontz 05th OCTOBER 2017 31 / 55 

Introduction to Machine Learning Typical ML task : Linear Regression Typical ML task : Classification Fundamentals Deep Learning Success Image classification Anthony FAUSTINE [email protected] pythontz 05th OCTOBER 2017 32 / 55 

Introduction to Machine Learning Typical ML task : Linear Regression Typical ML task : Classification Fundamentals Deep Learning Success Game Anthony FAUSTINE [email protected] pythontz 05th OCTOBER 2017 33 / 55 

Introduction to Machine Learning Typical ML task : Linear Regression Typical ML task : Classification Fundamentals Deep Learning Success Fashion Figure – Deep style Anthony FAUSTINE [email protected] pythontz 05th OCTOBER 2017 34 / 55 

Introduction to Machine Learning Typical ML task : Linear Regression Typical ML task : Classification Fundamentals Why Deep Learning and why now ? Why deep learning : Hand-Engineered Features vs. Learned features Anthony FAUSTINE [email protected] pythontz 05th OCTOBER 2017 35 / 55 

Introduction to Machine Learning Typical ML task : Linear Regression Typical ML task : Classification Fundamentals Why Deep Learning and why now ? Why now : • Large data-sets • GPU Hardware Advances + Price Decreases • Improved Techniques (Better algorithms & understanding) • Open source tools and models (Theano,Tensorflow, pytorch etc) Anthony FAUSTINE [email protected] pythontz 05th OCTOBER 2017 36 / 55 

Introduction to Machine Learning Typical ML task : Linear Regression Typical ML task : Classification Fundamentals The Perceptron A perceptron classifier is a simple model of a neuron • Invented in 1954 by Frank Rosenblatt • Inspired by neurobiology The output : y = f (x) = g(z(x)) • pre-activation : z(x) = wx + b • activation function : g(.) • x, y input, output. • w, b weight and bias parameter θ Anthony FAUSTINE [email protected] pythontz 05th OCTOBER 2017 37 / 55 

Introduction to Machine Learning Typical ML task : Linear Regression Typical ML task : Classification Fundamentals The Perceptron : Activation Function Importance of Activation Functions • Activation functions add non-linearity to our network’s function. • Most real-world problems + data are non-linear. Anthony FAUSTINE [email protected] pythontz 05th OCTOBER 2017 38 / 55 

Introduction to Machine Learning Typical ML task : Linear Regression Typical ML task : Classification Fundamentals Multilayer Perceptrons (MLP) We can connect lots perceptron units together into a directed acyclic graph. Anthony FAUSTINE [email protected] pythontz 05th OCTOBER 2017 39 / 55 

Introduction to Machine Learning Typical ML task : Linear Regression Typical ML task : Classification Fundamentals Multilayer Perceptrons (MLP) : Single hidden layer • Hidden layer pre-activation : z(x) = w(1)x + b(1) (z(x)i = j w(1) i,j xj + b(1) i ) • Hidden layer activation h(1)(x) = g(z(x)) • output layer activation f (x) = O w(2)h(1)(x) + b(2) where O(.) is the output activation usually softmax for classification problem. Anthony FAUSTINE [email protected] pythontz 05th OCTOBER 2017 40 / 55 

Introduction to Machine Learning Typical ML task : Linear Regression Typical ML task : Classification Fundamentals Multilayer Perceptrons (MLP) : L hidden layers • layer pre-activation for k > 0 : z(k)(x) = w(k)h(k−1)x + b(k) • Hidden layer activation from 1 . . . L h(k)(x) = g(z(k)(x)) • output layer activation f (x) = h(L+1)(x) = O z(x)(x) Anthony FAUSTINE [email protected] pythontz 05th OCTOBER 2017 41 / 55 

Introduction to Machine Learning Typical ML task : Linear Regression Typical ML task : Classification Fundamentals Training Multilayer Perceptrons (MLP) Objective : Find parameters θ : w and b that minimize the cost function : arg max θ 1 N i L(f (x(i) : θ), y(i)) To train a neural net, we need • Loss function : L(f (x(i) : θ), y(i)) • A procedure to compute gradients : ∂Jθ ∂θ Anthony FAUSTINE [email protected] pythontz 05th OCTOBER 2017 42 / 55 

Introduction to Machine Learning Typical ML task : Linear Regression Typical ML task : Classification Fundamentals Training Multilayer Perceptrons (MLP) : Stochastic Gradient Descent (SGD Initialize θ randomly. For N epochs perform : • Randomly select a small batch of samples • Compute gradients : ∂Jθ ∂θ • Update parameters θ with update rule : θ(t+1) := θ(t) − α ∂Jθ ∂θ Stop when reaching criterion Anthony FAUSTINE [email protected] pythontz 05th OCTOBER 2017 43 / 55 

Introduction to Machine Learning Typical ML task : Linear Regression Typical ML task : Classification Fundamentals Training Multilayer Perceptrons (MLP) : Computing Gradients Backpropagation : an efficient way to compute partial derivatives of MLP. Anthony FAUSTINE [email protected] pythontz 05th OCTOBER 2017 44 / 55 

Introduction to Machine Learning Typical ML task : Linear Regression Typical ML task : Classification Fundamentals Deep learning Architecture : Convolutional Neural Network • Enhances the capabilities of MLP by inserting convolution layers. • Composed of many “filters”, which convolve, or slide across the data, and produce an activation at every slide position • Suitable for spatial data, object recognition and image analysis. • The common usage of CNN : self driving cars, drones, computer vision, text analytics Anthony FAUSTINE [email protected] pythontz 05th OCTOBER 2017 45 / 55 

Introduction to Machine Learning Typical ML task : Linear Regression Typical ML task : Classification Fundamentals Deep learning Architecture : Recurrent Neural Networks (RNN) • Have recurrent memory loops which take the input from the previous and/or same layers or states. • Have unique capability to model along the time dimension and arbitrary sequence of events and inputs. • Suitable for sequenced data analysis such as time-series, sentiment analysis, NLP, language translation, speech recognition, image captioning, and script recognition. • Common type : LSTM and GRUs. Anthony FAUSTINE [email protected] pythontz 05th OCTOBER 2017 46 / 55 

Introduction to Machine Learning Typical ML task : Linear Regression Typical ML task : Classification Fundamentals Deep learning Architecture : Deep Generative models Idea :learn to understand data through generation → replicate the data distribution that you give it. Two types : Variational Autoencoders (VAE), and Generative Adversarial Networks(GAN). • Can be used to generate Musics, Speach, Langauge, Image, Handwriting, Language • Suitable for unsupervised learning as they need lesser labelled data to train. Anthony FAUSTINE [email protected] pythontz 05th OCTOBER 2017 47 / 55 

Introduction to Machine Learning Typical ML task : Linear Regression Typical ML task : Classification Fundamentals Outline 1 Introduction to Machine Learning 2 Typical ML task : Linear Regression 3 Typical ML task : Classification 4 Fundamentals of Deep Learning 5 Research direction and opportunities 6 Python libraries for ML and AI Anthony FAUSTINE [email protected] pythontz 05th OCTOBER 2017 48 / 55 

Introduction to Machine Learning Typical ML task : Linear Regression Typical ML task : Classification Fundamentals Research direction • Unsupervised deep learning. • Add more reasoning (uncertatinity) abilities in deep learning models (deep-probabilistic models) • Many applications which are under-explored especially in developing countries. • Deep reiforcement learning. • Computational efficiency. Anthony FAUSTINE [email protected] pythontz 05th OCTOBER 2017 49 / 55 

Introduction to Machine Learning Typical ML task : Linear Regression Typical ML task : Classification Fundamentals Outline 1 Introduction to Machine Learning 2 Typical ML task : Linear Regression 3 Typical ML task : Classification 4 Fundamentals of Deep Learning 5 Research direction and opportunities 6 Python libraries for ML and AI Anthony FAUSTINE [email protected] pythontz 05th OCTOBER 2017 50 / 55 

Introduction to Machine Learning Typical ML task : Linear Regression Typical ML task : Classification Fundamentals Jupyter Jupyter : Open-source web application for interactive and exploratory computing. • Allows to create and share documents that contain live code, equations, visualizations and explanatory text. • It is a platform for Data Science at scale. • Covers all the life-cycle of scientific ideas :ideas to publications. • Demo Anthony FAUSTINE [email protected] pythontz 05th OCTOBER 2017 51 / 55 

Introduction to Machine Learning Typical ML task : Linear Regression Typical ML task : Classification Fundamentals Scikit-Learn for ML Scikit-Learn (sklearn) is Python’s premier general-purpose machine learning library. Anthony FAUSTINE [email protected] pythontz 05th OCTOBER 2017 52 / 55 

Introduction to Machine Learning Typical ML task : Linear Regression Typical ML task : Classification Fundamentals Python ML and AI libraries Tensorflow Theano Pytorch Keras Edward PyMC3 NLTK Anthony FAUSTINE [email protected] pythontz 05th OCTOBER 2017 53 / 55 

Introduction to Machine Learning Typical ML task : Linear Regression Typical ML task : Classification Fundamentals Data Science Platform Kaggle : helps you learn, work, and play. Data set : • Academic Torrents • UCI Machine learning repository Anthony FAUSTINE [email protected] pythontz 05th OCTOBER 2017 54 / 55 

Introduction to Machine Learning Typical ML task : Linear Regression Typical ML task : Classification Fundamentals THANK YOU Anthony FAUSTINE [email protected] pythontz 05th OCTOBER 2017 55 / 55