Machine Learning Concepts PDF

Machine Learning Rodrique Kafando, PhD Research Scientist 1 [email protected] AI Concepts L3 Rodrique Kafando, PhD Research Scientist 2 [email protected]  Who am I...

Machine Learning Rodrique Kafando, PhD Research Scientist 1 [email protected] AI Concepts L3 Rodrique Kafando, PhD Research Scientist 2 [email protected]  Who am I ? I’m Rodrique KAFANDO PhD in Computer Science, Data Mining & AI, University of Montpellier, France Research Scientist at the Virtual University of Burkina Faso Contributor at CITADEL Burkina Faso Key competencies : Spatio-temporal & Complex event monitoring Heterogeneous Data mutual Enrichment & Analysis Multi-Agent Systems for Smart Cities planning Disambiguation of Spatial information Rodrique Kafando, PhD Research Scientist 3 [email protected]  Strategic Research Areas Rodrique Kafando, PhD Research Scientist 4 [email protected] FORMATION : Master Fouilles de Données et IA de l’UV-BF https://citadel.bf/ [email protected] Rodrique Kafando, PhD Research Scientist 5 [email protected] AI Concepts I - Define what’s AI II - Give problem types and problem-solving paradigms III - Give Intuition of concepts that are applicable to AI IV - Cite some Uses for AI algorithms - Real world problems You won’t be AI expert at the end of this presentation! Rodrique Kafando, PhD Research Scientist 6 [email protected] AI Concepts I - Definition of IA Rodrique Kafando, PhD Research Scientist 7 [email protected] AI Concepts I - Definition of IA What is Artificial Intelligence ? “The true sign of intelligence is not knowledge, but imagination.” Albert Einstein “Intelligence without ambition is a bird without wings.” “Intelligence is the ability to Salvador Dalí adapt” Stephen Hawking Intelligence is a concept that has no agreed-upon definition! Rodrique Kafando, PhD Research Scientist 8 [email protected] AI Concepts I - Definition of IA What is Artificial Intelligence ? AI - Synthetic system that exhibits “intelligent” behavior. Rishal Hurbans “AI is whatever hasn’t been done yet.” e.g : a self-driving car Douglas Hofstadter Humans as being the dominant (and most intelligent) species! Rodrique Kafando, PhD Research Scientist 9 [email protected] AI Concepts I - Definition of IA Examples of things that What is Artificial Intelligence ? A system that succeeds at playing many types of complex games exhibit AI-likeness A cancer tumor detection system A system that generates artwork based on little input A self-driving car Etc. Rodrique Kafando, PhD Research Scientist 10 [email protected] AI Concepts I - Definition of IA Data - the core of AI algorithms Data are mandatory for all AI algorithms Understanding data is one the key of success of AI algorithms Quality of data determine the outcome of the algorithms ○ E.g of problems : incorrect choice of data, poorly represented data, or missing data Rodrique Kafando, PhD Research Scientist 11 [email protected] AI Concepts I - Definition of IA Data - the core of AI algorithms Data are everywhere with different forms (we can’t even sense!) Rodrique Kafando, PhD https://www.manning.com/ Research Scientist 12 [email protected] AI Concepts I - Definition of IA Algorithms Concepts - Set of instructions An algorithm is a set of instructions and rules provided as a specification to accomplish a specific goal. Algorithm Inputs Output Set of finite steps ********Read a book******** 1- Find the book. 2-Open the book. 3 - While unread pages remain, 1- Read page. 2- Turn next page. 3- Think about what you learned. 4- Think about how you can apply your learnings in the real world. Rodrique Kafando, PhD Research Scientist 13 [email protected] AI Concepts I - Definition of IA A brief history harvard.edu Rodrique Kafando, PhD Research Scientist 14 [email protected] AI Concepts I - Definition of IA A brief history https://connectjaya.com/ai-timeline/ Rodrique Kafando, PhD Research Scientist 15 [email protected] AI Concepts II - Problem types and problem-solving paradigms Rodrique Kafando, PhD Research Scientist 16 [email protected] AI Concepts II - Problem types and problem-solving paradigms Identifying the problem Depending on the need, we may have insights that allow us to identify the associated problem(s). This will guide us towards the choice of algorithms to use/prioritize to solve the problem. Problems are classified based on the context and the goal. Rodrique Kafando, PhD Research Scientist 17 [email protected] AI Concepts II - Problem types and problem-solving paradigms Search problems: Find a path to a solution >> A search problem involves a situation that has multiple possible solutions, each of which represents a sequence of steps (path) toward a goal. - Some solutions contain overlapping subsets of paths; - Some are better than others; and some are cheaper to achieve than others. Better Solution determined by the specific problem at hand. Cheaper Solution means computationally cheaper to execute. Rodrique Kafando, PhD Research Scientist 18 [email protected] AI Concepts II - Problem types and problem-solving paradigms Optimization problems: Find a good solution >> An optimization problem involves a situation in which there are a vast number of valid solutions and the absolute-best solution is difficult to find. Local Best solution best solution within a specific area in the search space Global Best solution best solution in the entire search space Enormous number of possibilities, and each one solves differently the problem Rodrique Kafando, PhD Research Scientist 19 [email protected] AI Concepts II - Problem types and problem-solving paradigms Prediction & Classification problems: Learn from patterns in data >> Prediction problems are problems in which we have data about something and want to try to find patterns. ??? https://www.dailymail.co.uk/ Predict the fuel consumption of a new model of vehicle, given its engine size? What condition must be satisfied between the variables for a good Rodrique Kafando, PhD prediction? Research Scientist 20 [email protected] AI Concepts II - Problem types and problem-solving paradigms Prediction & Classification problems: Learn from patterns in data >> Classification problems are similar to prediction problems, but instead of trying to find an exact prediction such as fuel consumption, we try to find a category of something based on its features. Given the dimensions of a vehicle, its engine size, and the number of seats, can we predict whether that vehicle is a motorcycle, sedan, or sport-utility vehicle! https://www.coxautoinc.com/ Classification problems require finding patterns in the data that group examples into categories. Rodrique Kafando, PhD Research Scientist 21 [email protected] AI Concepts II - Problem types and problem-solving paradigms Clustering problems: Identify patterns in data >> Clustering problems include scenarios in which trends and relationships are uncovered from data. Different aspects of the data are used to group examples in different ways. https://www.javatpoint.com Clustering aims to find relationships in data even when a precise question is not being asked. Rodrique Kafando, PhD Research Scientist 22 [email protected] AI Concepts II - Problem types and problem-solving paradigms Deterministic models: Same result each time is calculated >> Deterministic models are models that, given a specific input, return a consistent output. Kelvin = Celsius + 273.15 E.g : Simple linear regression Y= a + bX If something is deterministic, you have all of the data necessary to predict (determine) the outcome with 100% certainty. Rodrique Kafando, PhD Research Scientist 23 [email protected] AI Concepts II - Problem types and problem-solving paradigms Stochastic/Probabilistic models: Potentially different result each time it’s calculated >> Probabilistic models are models that, given a specific input, return an outcome from a set of possible outcomes. - The weather (X) in a given place at time (t) E.g : - Number of customers (X) in a checkout line at time (t) - Traffic congestion (X) at time (t). A stochastic model represents a situation where uncertainty is present. In other words, it’s a model for a process that has some kind of randomness Rodrique Kafando, PhD Research Scientist 24 [email protected] AI Concepts III - Intuition of concepts that are applicable to AI Rodrique Kafando, PhD Research Scientist 25 [email protected] AI Concepts III - Intuition of concepts that are applicable to AI Levels of AI Science fiction movies so far! E.g : Postapocalyptic worlds, in which all machines are connected, are able to Super Intelligence reason about things beyond our Unknown understanding, and dominate human. General Intelligence Human-like solutions Human Humanlike intelligence - able to learn from various experiences and interactions in the world and apply that understanding from one problem to another. Narrow Intelligence Pong-playing program Map-routing program Fraud detection program Cancer detection program Specific-purpose solutions Etc. Solve problems in a specific context or Rodrique Kafando, PhD domain. Research Scientist 26 [email protected] AI Concepts III - Intuition of concepts that are applicable to AI Old AI and New AI >> Old AI is often understood as being systems in which people encoded the rules that cause an algorithm to exhibit intelligent behavior—via in-depth knowledge of the problem or by trial and error. Manually creating a decision tree and the rules and options in the entire decision tree. >> New AI aims to create algorithms and models that learn from data and create their own rules that perform as accurately as, or better than, human-created rules. Use AI algorithms for robust decision tree. https://towardsai.net/ Rodrique Kafando, PhD Research Scientist 27 [email protected] AI Concepts III - Intuition of concepts that are applicable to AI Categorization of concepts within AI Search algorithms Biology-in Useful for solving problems in which several actions Deep spired are required to achieve a goal. Learning Algorithms Biology-inspired algorithms Machine Learning By observing and learning from various phenomena, we’ve gained knowledge of how some organic systems operate and Search of how simple rules can result in emergent intelligent ms Algorith behavior. Artificial Intelligence Rodrique Kafando, PhD Research Scientist 28 [email protected] AI Concepts III - Intuition of concepts that are applicable to AI Categorization of concepts within AI Machine Learning algorithms ML algorithm takes a statistical approach to training models to learn from data. Three main approaches: Biology-in Deep spired Learning Algorithms Supervised Learning Unsupervised Learning Reinforcement Learning Training data has Uncovers hidden Inspired by behavioral known outcomes for a relationships and psychology - it describes question being asked. structures within the rewarding an individual if a Machine Learning data that guide us in useful action was performed asking the dataset and penalizing that individual relevant questions. if an unfavorable action was Search performed. ms Algorith e.g : determining the type e.g : It may find patterns of fruit if we have a set of in properties of similar fruits and group them e.g : a robot that is tasked to open data that includes the accordingly, which can doors; it is penalized when it Artificial Intelligence weight, color, texture, and inform the exact questions doesn’t open a door and rewarded fruit label for each we want to ask the data. when it does. example. Rodrique Kafando, PhD Research Scientist 29 [email protected] AI Concepts III - Intuition of concepts that are applicable to AI Categorization of concepts within AI Deep Learning algorithms Stems from machine learning, DL is a broader family of approaches and algorithms that are used to achieve narrow intelligence and strive toward general intelligence. Biology-in Deep spired [Xianlin Wang, al 2021] Learning Algorithms Machine Learning Very much inspired by the structure of the brain. Search ms Algorith Artificial Intelligence Rodrique Kafando, PhD Research Scientist 30 [email protected] AI Concepts IV - Uses for AI algorithms - Real world problems Rodrique Kafando, PhD Research Scientist 31 [email protected] AI Concepts IV - Uses for AI algorithms - Real world problems Where there are data and problems to solve, there are potential applications of AI. Some examples…. Agriculture : Optimal plant growth Crops grow differently based on the type of crop, the nutrients in the soil, the water content of the soil, the bacteria in the water, and the weather conditions in the area, among other things. manning.com Rodrique Kafando, PhD Research Scientist 32 [email protected] AI Concepts IV - Uses for AI algorithms - Real world problems Where there are data and problems to solve, there are potential applications of AI. Some examples…. Banking : Fraud detection Fraud occurs when someone is dishonest or does something illegal to acquire something for themselves. With the vast amounts of transaction data available, we can, in real-time, find patterns of transactions specific to an individual’s spending behavior that may be out of the ordinary. nexocode.com Rodrique Kafando, PhD Research Scientist 33 [email protected] AI Concepts IV - Uses for AI algorithms - Real world problems Where there are data and problems to solve, there are potential applications of AI. Some examples…. CyberSecurity : Attack detection & Handling AI can identify and block potential attacks on servers using logs. We send and receive sensitive information over the internet all the time. mit.edu Rodrique Kafando, PhD Research Scientist 34 [email protected] AI Concepts IV - Uses for AI algorithms - Real world problems Where there are data and problems to solve, there are potential applications of AI. Some examples…. Health Care : Diagnosis of patients Advances in deep learning have improved the detection of tumors in images generated by scans. Now doctors can detect cancer earlier, which means that a patient can get the required treatment in time and have a higher chance of recovery. AI can be used to find patterns in symptoms, ailments, hereditary genes, mit.edu geographic locations, and the like. Rodrique Kafando, PhD Research Scientist 35 [email protected] AI Concepts IV - Uses for AI algorithms - Real world problems Where there are data and problems to solve, there are potential applications of AI. Some examples…. Logistics: Rooting and Optimization Traveling-salesperson problem Logistics industry is a huge market of different types of vehicles delivering different types of goods to different locations, with different demands and deadlines. ODSC Through AI, we can find optimal routes between locations in terms of time and distance. Rodrique Kafando, PhD Research Scientist 36 [email protected] AI Concepts IV - Uses for AI algorithms - Real world problems Where there are data and problems to solve, there are potential applications of AI. Some examples…. Telecoms : Networks Optimization - monitor behavior on a network and optimize routing. - to optimize the networks based on known load from certain individuals, areas, and specific local networks. [Khaled B, and al, 2021] The network data can also be instrumental for understanding where people are and who they are, which is useful for city planning. Rodrique Kafando, PhD Research Scientist 37 [email protected] AI Concepts IV - Uses for AI algorithms - Real world problems Where there are data and problems to solve, there are potential applications of AI. Some examples…. Marketing : Recommendation systems harvard.edu User data and company well-specified goal are necessary. Rodrique Kafando, PhD Research Scientist 38 [email protected] AI Concepts IV - Uses for AI algorithms - Real world problems Where there are data and problems to solve, there are potential applications of AI. Some examples…. Marketing : Recommendation systems Directly compares products and finds other products that are similar to what the user likes. harvard.edu harvard.edu Suggests products that are favored by other people with similar tastes to the user. Rodrique Kafando, PhD Research Scientist 39 [email protected] AI Concepts IV - Uses for AI algorithms - Real world problems Where there are data and problems to solve, there are potential applications of AI. Some examples…. Games : Creating AI agent Art : Creating masterpieces manning economictimes Rodrique Kafando, PhD Research Scientist 40 [email protected] AI Concepts Human-level AI - Yann LeCun Rodrique Kafando, PhD Research Scientist 41 [email protected] AI Concepts Top@100 industry wise companies https://www.analyticsinsight.net/artificial-intelligence-100-top -industry-wise-companies-listing/ Rodrique Kafando, PhD Research Scientist 42 [email protected] AI Concepts Thank you! Rodrique Kafando, PhD Research Scientist 43 [email protected] Machine Learning Rodrique Kafando, PhD Research Scientist 44 [email protected] Machine Learning What is Machine Learning ? Let's say you want to rent a house in the city: - A one-bedroom apartment in the city center (close to work) costs $5,000 per month. - A one-bedroom apartment in the city center with a garage costs $6,000 per month. - A one-bedroom apartment outside the city center, where you will need to travel to work, costs $3,000 per month. - A two-bedroom apartment outside the city center costs $4,500 per month. - A one-bedroom apartment outside the city center with a garage costs $3,800 per month. Rodrique Kafando, PhD Research Scientist 45 [email protected] Machine Learning What is Machine Learning ? Let's say you want to rent a house in the city: This example shows how we use data to find patterns and make decisions. Rodrique Kafando, PhD Machine learning aims to find patterns in data for useful applications in the real world. Research Scientist 46 [email protected] Machine Learning What is Machine Learning ? Detection of potential cancerous cells. Rodrique Kafando, PhD Machine learning aims to find patterns in data for useful applications in the real world. Research Scientist 47 [email protected] Machine Learning What is Machine Learning ? predict Rodrique Kafando, PhD Research Scientist 48 Machine learning aims to find patterns in data for useful applications in the real world. [email protected] Machine Learning What is Machine Learning ? Rodrique Kafando, PhD Research Scientist 49 [email protected] Machine Learning REFRESH - Categorization of concepts within AI Machine Learning algorithms ML algorithm takes a statistical approach to training models to learn from data. Three main approaches: Biology-in Deep spired Learning Algorithms Supervised Learning Unsupervised Learning Reinforcement Learning Training data has Uncovers hidden Inspired by behavioral known outcomes for a relationships and psychology - it describes question being asked. structures within the rewarding an individual if a Machine Learning data that guide us in useful action was performed asking the dataset and penalizing that individual relevant questions. if an unfavorable action was Search performed. ms Algorith e.g : determining the type e.g : It may find patterns of fruit if we have a set of in properties of similar fruits and group them e.g : a robot that is tasked to open data that includes the accordingly, which can doors; it is penalized when it Artificial Intelligence weight, color, texture, and inform the exact questions doesn’t open a door and rewarded fruit label for each we want to ask the data. when it does. example. Rodrique Kafando, PhD Research Scientist 50 [email protected] Machine Learning What is Supervised Learning ? Rodrique Kafando, PhD Research Scientist 51 [email protected] Machine Learning What is Supervised Learning ? From daily usage… Observe and direct the execution of a task, project or activity. Rodrique Kafando, PhD Research Scientist 52 [email protected] Machine Learning What is Supervised Learning ? finxter Look at data, understand the patterns and relationships among the data, and predict the results if we are given new examples of different data in the same format. Rodrique Kafando, PhD Research Scientist 53 [email protected] Machine Learning What is Supervised Learning ? Two types of Supervised Learning Rodrique Kafando, PhD Research Scientist 54 [email protected] Machine Learning What is Supervised Learning ? Classification Aims to predict categories of examples based on their features. Can we determine whether a cell is benign or malignan based on its Size, Shape, Norm, etc ? Rodrique Kafando, PhD Research Scientist 55 [email protected] Machine Learning What is Supervised Learning ? Regression Involes drawing a line through a set of data points to most closely fit the overall shape of the data. ??? Can we determine CO2 emission of a given car based on its Engine Size, Number of Cylinders, Fuel Consumption, etc ? Rodrique Kafando, PhD Research Scientist 56 [email protected] Machine Learning What is Supervised Learning ? Regression Rodrique Kafando, PhD Research Scientist 57 [email protected] Machine Learning What is Supervised Learning ? Regression Rodrique Kafando, PhD Research Scientist 58 [email protected] Machine Learning What is unsupervised Learning ? Rodrique Kafando, PhD Research Scientist 59 [email protected] Machine Learning What is unsupervised Learning ? finxter finding underlying patterns in data that may be difficult to find by inspecting the data manually. Rodrique Kafando, PhD Research Scientist 60 [email protected] Machine Learning Training, Validation, and Test set 1- Train and evaluate the model on the whole dataset - Test set will be a portion of the training set - High “training accuracy” - Low “out-of-sample” accuracy medium.com overfitting → which happens whenever the model is too adapted to training data. Rodrique Kafando, PhD Research Scientist 61 [email protected] Machine Learning Training, Validation, and Test set 2- Split data in Train/Test set - Mutually exclusive - More accurate evaluation on “out-of-sample” accuracy - Highly dependent on which dataset the model is trained and medium.com tested Rodrique Kafando, PhD Research Scientist 62 [email protected] Machine Learning Training, Validation, and Test set 3- Train, validation and test set - Mutually exclusive - Accurate evaluation on “out-of-sample” accuracy - Test set highly independent on which dataset the model is medium.com trained and tested Rodrique Kafando, PhD Research Scientist 63 [email protected] Machine Learning Simplified Machine Learning Workflow Rodrique Kafando, PhD Research Scientist 64 [email protected] Machine Learning Collecting and understanding data: Know your context Collecting and understanding the data Each domain has its specific terminology. you’re working with is paramount to a successful machine learning endeavor. DATA Data may comes from various Domain knowledge is very important. sources…to be complementary. Every success AI designed with domain experts. Rodrique Kafando, PhD Research Scientist 65 [email protected] Machine Learning Collecting and understanding data: Know your context Case study : Diamond measurement https://www.manning.com/ Rodrique Kafando, PhD Research Scientist 66 [email protected] Machine Learning Case study : Diamond measurement Rodrique Kafando, PhD The diamond dataset Research Scientist 67 [email protected] Machine Learning Preparing the data: Clean and wrangle The diamond dataset with missing data In real life, your data will not always be clean and ready to use: therefore, some pre-processing and cleaning is required. There are always missing data, inconsistent data, and data in a format that is difficult to work with for the algorithms that we want to use. Rodrique Kafando, PhD Research Scientist 68 [email protected] Machine Learning Training a model: Predict with linear regression Goal : Find the best line that fits (all) the points Rodrique Kafando, PhD Research Scientist 69 [email protected] Machine Learning Training a model: Predict with linear regression Goal : Find the best line that fit all the points 1- Find the mean for each feature. $3,431 for price 1,229 for carats Rodrique Kafando, PhD Research Scientist 70 [email protected] Machine Learning Training a model: Predict with linear regression Goal : Find the best line that fit all the points 1- Find the mean for each feature. $3,431 for price 1,229 for carats Some regression lines might be better than others at fitting the data. Rodrique Kafando, PhD Research Scientist 71 [email protected] Machine Learning Training a model: Predict with linear regression Goal : Find the best line that fit all the points 1- Find the mean for each feature. $3,431 for price 1,229 for carats How do you find the best line ? We can use the method of least squares. Rodrique Kafando, PhD Research Scientist 72 [email protected] Machine Learning Training a model: Predict with linear regression Goal : Find the best line that fit all the points 2- Finding regression lines with the least-squares method. Ituition behind the regression lines. Rodrique Kafando, PhD Research Scientist 73 [email protected] Machine Learning Training a model: Predict with linear regression Goal : Find the best line that fit all the points 2- Finding regression lines with the least-squares method. A straight line is represented by the equation : y = c + mx y : The dependent variable x : The independent variable m : The slope of the line c : The y-value where the line intercepts the y axis Rodrique Kafando, PhD Research Scientist 74 [email protected] Machine Learning Training a model: Predict with linear regression Goal : Find the best line that fit all the points 2- Finding regression lines with the least-squares method. The basic workflow for calculating a regression line. Rodrique Kafando, PhD Research Scientist 75 [email protected] Machine Learning Training a model: Predict with linear regression Goal : Find the best line that fit all the points 3- plot the regression line Y = 3039.4406 + 0.3186X Rodrique Kafando, PhD Research Scientist 76 [email protected] Machine Learning Testing a model: Predict with linear regression Goal : Find the best line that fit all the points Find the mean squared error for this example! -Predict the price for Carat = 1250. Rodrique Kafando, PhD Research Scientist 77 [email protected] Machine Learning Tools Rodrique Kafando, PhD Research Scientist 78 [email protected] Machine Learning A map of popular machine learning algorithms. Rodrique Kafando, PhD Research Scientist 79 [email protected] Machine Learning Thank you! Rodrique Kafando, PhD Research Scientist 80 [email protected]