├── AI4I-Literacy_in_AI_Certificate_41862757.pdf ├── README.md └── AI4I-Literacy_in_AI-Quiz_and_Azure_ML_Studio_Links.txt /AI4I-Literacy_in_AI_Certificate_41862757.pdf: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/JNYH/AI4I_Literacy_in_AI/HEAD/AI4I-Literacy_in_AI_Certificate_41862757.pdf -------------------------------------------------------------------------------- /README.md: -------------------------------------------------------------------------------- 1 | # AI4I_Literacy_in_AI 2 | This is a memo to share what I have learnt in AI4I - Literacy in AI 3 | 4 | Literacy in AI is a 5-hour course which introduces learners into the world of machine learning. It also explores the topic of ethics in AI. 5 | 6 | Upon completion of this course, the learner would be exposed a range of introductory topics related to AI. 7 | 8 | ## Learning Outcomes 9 | At the end of the course, you will: 10 | 11 | Understand common machine learning terms 12 | 13 | Understand what machine learning models are 14 | 15 | Understand the basic workings of deep learning 16 | 17 | Be able to identify possible ethical issues in a typical AI project 18 | 19 | Build a simple demand forecasting model using Microsoft Azure ML Studio (Classic) 20 | 21 | ## Course Scope 22 | 23 | This course consists of 10 modules with 5 quizzes. To successfully complete the course, a learner will have to achieve 100% scores for all quizzes with unlimited attempts. 24 | 25 | The total estimated course learning time is 5 hours. 26 | 27 | Upon successful completion of the course, you will receive the "AI for Industry – Literacy in AI" certificate. 28 | 29 | ## Azure ML Studio 30 | 31 | The course includes the use of Azure Machine Learning Studio: 32 | ![image](https://user-images.githubusercontent.com/52286325/142727418-c2dd2952-1192-4e10-bb58-2c7462b0582f.png) 33 | 34 | ## Some links for further practice 35 | 36 | Demand Forecast of Time Series Data 37 | 38 | https://gallery.azure.ai/Experiment/Demand-Forecast-of-Time-Series-Data 39 | 40 | A model recommending e-commerce products to users 41 | 42 | https://gallery.azure.ai/Experiment/E-commerce-Product-Recommender 43 | 44 | A model which segments customers 45 | 46 | https://gallery.azure.ai/Experiment/Marketing-Customer-Segmentation 47 | 48 | A model which predicts the credit risk for suppliers and customers 49 | 50 | https://gallery.azure.ai/Experiment/Corporate-Credit-Risk-Prediction 51 | 52 | A model which predicts the credit risk of individual borrowers 53 | 54 | https://gallery.azure.ai/Experiment/Finance-Credit-Risk-Classification 55 | 56 | A model which predicts customer churn 57 | 58 | https://gallery.azure.ai/Experiment/Telco-Churn 59 | 60 | ![image](https://user-images.githubusercontent.com/52286325/142729301-f0f7f36a-ecf8-4614-9175-879bbf9af812.png) 61 | 62 | -------------------------------------------------------------------------------- /AI4I-Literacy_in_AI-Quiz_and_Azure_ML_Studio_Links.txt: -------------------------------------------------------------------------------- 1 | AI4I - Literacy in AI 2 | 3 | 4 | 5 | Machine Learning For Everyone Quiz 6 | 7 | Machine learning finds patterns in existing data and applies the patterns to new data. Is this true or false? 8 | True <-answer 9 | False 10 | 11 | What is a target variable? 12 | The most important variable for predicting the output. 13 | The variable that you do not have data for and still need to obtain. 14 | The value that you are trying to predict. <-answer 15 | 16 | Which type of machine learning is normally used for customer segmentation? 17 | Supervised Learning 18 | Unsupervised Learning <-answer 19 | 20 | What is a confusion matrix? 21 | A way to evaluate classification models. <-answer 22 | A way that computers use to create classification models. 23 | An advanced way for computers to organize machine learning data. 24 | 25 | What are hyperparameters? 26 | Inputs to a machine learning algorithm that can be changed. <-answer 27 | A type of classification model. 28 | 29 | What is one disadvantage of deep learning? 30 | It is difficult to explain how the AI arrives at its predictions. The model becomes a 'black box'. <-answer 31 | It is less accurate. 32 | It only works for a few types of machine learning problems. 33 | 34 | 35 | 36 | AIET-3 Ethics Issues In Product and Model Design Quiz 37 | 38 | Suppose we have evaluated an AI product to have a likely and severe negative impact when the model fails. What can we do to reduce negative impact caused by AI? 39 | Introduce a human-out-of-the-loop model, ie. allow the AI full control over the decision-making process, with no easy human override. 40 | Introduce a human-in-the-loop model, ie. require a human to make affirmative action on all decisions, with the AI making recommendations. <-answer 41 | Make the model creation open-source. 42 | 43 | Suppose we have evaluated an AI product which uses facial recognition to identify criminal fugitives. We believe that it is a very negative outcome if innocent civilians are being wrongly identified and taken into custody. Which of the following is true? 44 | The negative impact of false positives is high, so we should lower the confidence needed to identify suspected fugitives. 45 | The negative impact of false negatives is high, so we should lower the confidence needed to identify suspected fugitives. 46 | The negative impact of false positives is high, so we should increase the confidence needed to identify suspected fugitives. <-answer 47 | The negative impact of false negatives is high, so we should increase the confidence needed to identify suspected fugitives. 48 | 49 | 50 | 51 | AIET-4 Ethics Issues in Data Collection and Access Quiz 52 | 53 | What is the ethical concern when a dataset is not representative? 54 | The model is more likely to be inaccurate. 55 | The model is more likely to be less transparent because it is missing data. 56 | The model is more likely to be unfair, performing poorly for under-represented user groups. <-answer 57 | The model is likely to have more false negatives. 58 | 59 | Which of the following is NOT a way to address the privacy and consent concerns of the people we collect data on? 60 | De-identify data by masking and deleting information to make it difficult or impossible to re-identify the subjects of the data. 61 | Give users the option to opt out of their data being used. 62 | Test the data make sure it is fair to different user subgroups. <-answer 63 | Restrict access to the data and take security measures to protect the data. 64 | 65 | 66 | 67 | AIET-5 Ethics Issues in Model Evaluation Quiz 68 | 69 | What is the main ethics issue that can be identified in the model evaluation process? 70 | The issue of privacy: whether user data can lead to their re-identification. 71 | The issue of accountability: whether AI engineers take responsibility for their role in the model evaluation. 72 | The issue of fairness: whether different user subgroups are treated equitably or whether any of them are being disadvantaged. <-answer 73 | The issue of consent: whether users give permission to the use of their data in the training of AI models. 74 | 75 | 76 | 77 | AIET-7 Resources and readings on AI Ethics 78 | 1. Singapore: PDPC Model AI Governance Framework and Self-Assessment Tool 79 | https://www.pdpc.gov.sg/-/media/Files/PDPC/PDF-Files/Resource-for-Organisation/AI/SGIsago.pdf 80 | 81 | 2. World Government Summit 2019: AI Ethics: The Next Big Thing In Government 82 | https://www.worldgovernmentsummit.org/docs/default-source/default-document-library/deloitte-wgs-report-en-lq.pdf?sfvrsn=1acfc90b_0 83 | 84 | 3. Microsoft: research paper on Fairness in AI 85 | https://pdfs.semanticscholar.org/58bb/221c1e375f254826b7b7341f74057e87676c.pdf 86 | 87 | 4. Google AI: Perspectives on Issues in AI Governance 88 | https://ai.google/static/documents/perspectives-on-issues-in-ai-governance.pdf 89 | 90 | 91 | 92 | AI4I-Basic-Q: Practice Quiz 93 | 94 | What is the objective of the AI project? 95 | To determine the right number of bicycles to have ready. 96 | To forecast demand for bicycle rentals. <-answer 97 | To forecast weather. 98 | To forecast temperature. 99 | 100 | What type of model was used initially in the project? 101 | Decision tree. 102 | Support vector machines. 103 | K-Nearest Neighbors. 104 | Linear regression. <-answer 105 | 106 | What was the model changed to? 107 | Support vector machines. 108 | K-Nearest Neighbors. 109 | Boosted Decision Tree. <-answer 110 | Linear regression. 111 | 112 | Why was the type of model changed? 113 | To increase its accuracy. <-answer 114 | The first model cannot work for this problem. 115 | To increase the speed of model training. 116 | 117 | What are the features of a model? 118 | The inputs used to make the prediction. <-answer 119 | The variable to be predicted. 120 | The advantages and disadvantages of the model. 121 | The evaluated accuracy scores of the model. 122 | 123 | 124 | 125 | AI4I-Basic: Additional Practice 126 | 127 | Demand Forecast of Time Series Data 128 | https://gallery.azure.ai/Experiment/Demand-Forecast-of-Time-Series-Data 129 | 130 | A model recommending e-commerce products to users 131 | https://gallery.azure.ai/Experiment/E-commerce-Product-Recommender 132 | 133 | A model which segments customers 134 | https://gallery.azure.ai/Experiment/Marketing-Customer-Segmentation 135 | 136 | A model which predicts the credit risk for suppliers and customers 137 | https://gallery.azure.ai/Experiment/Corporate-Credit-Risk-Prediction 138 | 139 | A model which predicts the credit risk of individual borrowers 140 | https://gallery.azure.ai/Experiment/Finance-Credit-Risk-Classification 141 | 142 | A model which predicts customer churn 143 | https://gallery.azure.ai/Experiment/Telco-Churn 144 | 145 | --------------------------------------------------------------------------------