Machine Learning 2 deep Learning: An Intro
- 1. Machine Learning to Deep Learning: An Introduction S. I. Krishan Integrated Knowledge Solutions www.iksinc.online/home
- 2. Agenda • What is machine learning? • Machine learning terminology • Overview of machine learning methods • Machine learning to deep learning • Summary and Q & A [email protected]
- 3. What is Machine Learning? [email protected]
- 4. What is Machine Learning? • Machine learning deals with making computers learn to make predictions/decisions without explicitly programming them. Rather a large number of examples of the underlying task are shown to optimize a performance criterion to achieve learning. [email protected]
- 6. Buzz about Machine Learning "Every company is now a data company, capable of using machine learning in the cloud to deploy intelligent apps at scale, thanks to three machine learning trends: data flywheels, the algorithm economy, and cloud-hosted intelligence." Three factors are making machine learning hot. These are cheap data, algorithmic economy, and cloud-based solutions. [email protected]
- 7. Data is Getting Cheaper For example, Tesla has 780 million miles of driving data, and adds another million every 10 [email protected]
- 10. Cloud-Based Intelligence Emerging machine intelligence platforms hosting pre-trained machine learning models-as-a- service are making it easy for companies to get started with ML, allowing them to rapidly take their applications from prototype to production. Many open source machine learning and deep learning frameworks running in the cloud allow easy leveraging of pre- trained, hosted models to tag images, recommend products, and do general natural language processing tasks. [email protected]
- 13. Feature Vectors in ML • A machine learning system builds models using properties of objects being modeled. These properties are called features or attributes and the process of measuring/obtaining such properties is called feature extraction. It is common to represent the properties of objects as feature vectors. Sepal width Sepal length Petal width Petal length [email protected]
- 14. Learning Styles • Supervised Learning – Training data comes with answers, called labels – The goal is to produce labels for new data [email protected]
- 15. Supervised Learning Models • Classification models – Predict customer churning – Tag objects in a given image – Determine whether an incoming email is spam or not [email protected]
- 16. Supervised Learning Models • Regression models – Predict credit card balance of customers – Predict the number of 'likes' for a posting – Predict peak load for a utility given weather information [email protected]
- 17. Learning Styles • Unsupervised Learning – Training data comes without labels – The goal is to group data into different categories based on similarities Grouped Data [email protected]
- 18. Unsupervised Learning Models • Segment/ cluster customers into different groups • Organize a collection of documents based on their content • Make product Recommendations [email protected]
- 19. Learning Styles • Reinforcement Learning – Training data comes without labels – The learning system receives feedback from its operating environment to know how well it is doing – The goal is to perform better [email protected]
- 21. Overview of Machine Learning Methods [email protected]
- 22. Walk Through An Example: Flower Classification • Build a classification model to differentiate between two classes of flower [email protected]
- 23. How Do We Go About It? • Collect a large number of both types of flowers with the help of an expert • Measure some attributes that can help differentiate between the two types of flowers. Let those attributes be petal area and sepal area. [email protected]
- 24. Scatter plot of 100 examples of flowers [email protected]
- 25. We can separate the flower types using the linear boundary shown next. The parameters of the line represent the learned classification model. [email protected]
- 26. Another possible boundary. This boundary cannot be expressed via an equation. However, a tree structure can be used to express this boundary. Note, this boundary does better prediction of the collected data [email protected]
- 27. Yet another possible boundary. This boundary does prediction without any error. Is this a better boundary? [email protected]
- 28. Model Complexity • There are tradeoffs between the complexity of models and their performance in the field. A good design (model choice) weighs these tradeoffs. • A good design should avoid overfitting. How? – Divide the entire data into three sets • Training set (about 70% of the total data). Use this set to build the model • Test set (about 20% of the total data). Use this set to estimate the model accuracy after deployment • Validation set (remaining 10% of the total data). Use this set to determine the appropriate settings for free parameters of the model. May not be required in some cases. [email protected]
- 29. Measuring Model Performance • True Positive: Correctly identified as relevant • True Negative: Correctly identified as not relevant • False Positive: Incorrectly labeled as relevant • False Negative: Incorrectly labeled as not relevant Image: True Positive True Negative Cat vs. No Cat False Negative False Positive [email protected]
- 30. Precision, Recall, and Accuracy • Precision – Percentage of positive labels that are correct – Precision = (# true positives) / (# true positives + # false positives) • Recall – Percentage of positive examples that are correctly labeled – Recall = (# true positives) / (# true positives + # false negatives) • Accuracy – Percentage of correct labels – Accuracy = (# true positives + # true negatives) / (# of samples) [email protected]
- 31. Sum-of-Squares Error for Regression Models For regression model, the error is measured by taking the square of the difference between the predicted output value and the target value for each training (test) example and adding this number over all examples as shown [email protected]
- 32. Bias and Variance • Bias: expected difference between model’s prediction and truth • Variance: how much the model differs among training sets • Model Scenarios – High Bias: Model makes inaccurate predictions on training data – High Variance: Model does not generalize to new datasets – Low Bias: Model makes accurate predictions on training data – Low Variance: Model generalizes to new datasets [email protected]
- 33. The Guiding Principle for Model Selection: Occam’s Razor [email protected]
- 34. Model Building Algorithms • Supervised learning algorithms – Linear methods – k-NN classifier – Neural networks – Support vector machine – Decision tree – Ensemble method [email protected]
- 35. Illustration of k-NN Model Predicted label of test example with 1-NN model : Versicolor Predicted label of text example with 3-NN model: Virginica Test example [email protected]
- 36. Illustration of Decision Tree Model Petal width <= 0.8 Setosa Yes Petal length <= 4.75 Versicolor Virginica Yes No No The decision tree is automatically generated by a machine learning algorithm. [email protected]
- 37. Model Building Algorithms • Unsupervised learning – k-means clustering – Agglomerative clustering – Self organization feature maps – Recommendation system [email protected]
- 38. K-means Clustering K- “by far th cluste nowadays industri Choose the number of clusters, k, and initial cluster centers
- 39. K-means Clustering K- “by far th cluste nowadays industri K-means clusterinK-means clusterinK-means clusterin Assign data points to clusters based on distance to cluster centers
- 40. K-means Clustering K- “by far th cluste nowadays industri K-means clusterinK-means clusterinK-means clusterin Update cluster centers and reassign data points.
- 43. Machine Learning to Deep Learning [email protected]
- 44. Machine Learning Limitation • Machine learning methods operate on manually designed features. • The design of such features for tasks involving computer vision, speech understanding, natural language processing is extremely difficult. This puts a limit on the performance of the system. [email protected] Feature Extractor Trainable Classifier
- 45. Processing Sensory Data is Hard How do we bridge this gap between the pixels and meaning via machine learning?
- 46. Sensory Data Processing is Challenging
- 47. Sensory Data Processing is Challenging
- 48. Sensory Data Processing is Challenging
- 49. End-to-End Learning Coming up with features is often difficult, time consuming, and requires expert knowledge. So why not build integrated learning systems that perform end-to-end learning, i.e. learn the representation as well as classification from raw data without any engineered features. Feature Learner Trainable Classifier An approach performing end-to-end learning, typically performed through a series of successive abstractions, is in a nutshell deep learning
- 50. Deep Learning • It’s a subfield of machine learning that has shown remarkable success in dealing with applications requiring processing of pictures, videos, speech, and text. • Deep learning is characterized by: – Extremely large amount of data for training – Neural networks with exceedingly large number of layers – Training time running into weeks in many instances – End to end learning (No human designed rules/features are used) [email protected]
- 51. Deep Learning Models [email protected] Convolutional Neural Networks
- 52. Deep Learning Models [email protected] Recurrent Neural Networks
- 53. Deep Q Network (DQN) [email protected]
- 54. CNN Application Example: Object Detection and Labeling [email protected]
- 57. Training Deep Neural Networks • The work horse is the backpropagation algorithm based on chain-rule differentiation • Training consists of optimizing a suitable loss function using the stochastic gradient descent (SGD) algorithm to adjust networks parameters typically running into millions. [email protected]
- 58. Deep Learning Examples: Automatic Description Generation of Images [email protected] Training Data Examples Generated Captions
- 59. Deep Learning Examples: Predicting Heart Attacks [email protected]
- 61. Why Deep Learning Now? [email protected]
- 62. Transfer Learning • Trained models for one task can be used for another task via transfer learning – Faster training – Data needs are low [email protected]
- 63. Learning Bias [email protected] There is an urban legend that back in the 90’s, the US government commissioned for a project to detect tanks in a picture. The researchers built a neural network and used it classify the images. Once the product was actually put to test, it did not perform at all. On further inspection they noticed that the model had learnt the weather patterns instead of the tanks. The trained images with tanks were taken on a cloudy day and images with no tanks were taken on a sunny day. This is a prime example of how we need to understand the learning by a neural net.
- 66. Summary • Machine learning and deep learning are growing in their usage • Many of the ideas have been around for a long time but are being put into practice now because of technology progress • Several open source software resources (R, Rapid Miner, and Scikit-learn, PyTorch, TensorFlow etc.) to learn via experimentation • Applications based on vision, speech, and natural language processing are excellent candidates for deep learning • Need to filter hype from reality [email protected]