Machine Learning : why we should know and how it works
0 likes172 views
This document provides an overview of machine learning, including: - An introduction to machine learning and why it is important. - The main types of machine learning algorithms: supervised learning, unsupervised learning, and deep neural networks. - Examples of how machine learning algorithms work, such as logistic regression, support vector machines, and k-means clustering. - How machine learning is being applied in various industries like healthcare, commerce, and more.
![Python codes for k-means
#import ML algorithm
from sklearn.cluster import KMeans
#prepare datasets of X whose size is (100,2)
X = …
# build k means model with 2 clusters
kmeans = KMeans(n_clusters = 2, tol=0.0001, max_iter=100)
kmeans.fit(X)
#print the results of the value of centroids
print(kmeans.cluster_centers_) # two centroids
---- array([[-0.94665068, -0.97138368],[ 2.01559419, 2.02597093]])
#print the results of the value of centroids
print(kmeans.labels_) # 100 labels based on two centroids.
---- array([ 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1,
1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1])](https://image.slidesharecdn.com/machinelearning-whyweshouldknowandhowitworks-200503184559/85/Machine-Learning-why-we-should-know-and-how-it-works-36-320.jpg)
Machine Learning : why we should know and how it works
- 1. Machine Learning – Why we should know and how it works Kevin Lee, AVP of ML & AI Consultant
- 2. The Agenda ➢ Why Machine Learning? ➢ Introduction of Machine Learning – Supervised, Unsupervised, Deep Neural Network ➢ Machine Learning Implementation ➢ PhUSE Machine Learning Working Group ➢ Questions and Discussion
- 3. Kevin, do you know about Machine Learning?
- 4. Why did people ask / expect me if I know about Machine Learning? • Programming • Statistics / modeling • Working with data all the times
- 5. What is ML? An application of artificial intelligence (AI) that provides systems the ability to automatically learn and improve from experience without being explicitly programmed.
- 7. How does Human Learn? - Experience
- 8. How does Machine learn? Algorithm Input Data
- 9. How does Machine learn? Algorithm Test Data (cats) Real Data cat
- 10. How does Machine Learning work? X0 X1 X2 … Xn Y • Hypothesis Function - hθ(x) = θx + b • Minimize Cost Function, J(θ) = hθ(x) – Y, using Gradient Descent Input data Algorithm
- 11. Machine Learning Algorithms • Hypothesis function • Model for data • Y = θ0x0 + θ1x1 + θ2x2 + θ3x3 + …. + θnxn (Y = 2X + 30) • Cost function • measures how well hypothesis function fits into data. • Difference between actual data point and hypothesized data point. • Gradient Descent • Engine that minimizes cost function
- 12. Machine Learning Training Process 1. Select Data and hypothesis function 2. Start with random parameters of hypothesis function 3. Process first sets of data in hypothesis function 4. Find predicted values from hypothesis function 5. Calculate cost function ( actual – predicted) 6. Minimize cost function using Gradient Descent 7. Update parameters of hypothesis function based on cost function 8. Repeat steps ( 3 and 7) with next sets of data until the cost function becomes 0 or at the lowest point
- 13. Machine finds best model using input data X Y hθ(x) = x + 30 hθ(x) = 2x + 30
- 14. Model building in Machine Learning Hypothesis Function New Iteration Cost Function Calculation Applying Gradient Descent Updating Parameters
- 15. Best model can provide best predicted value X Y Xi Yi
- 16. More data, Better model, Better prediction X Y X Y Different data, Different model, Different prediction
- 17. X Y Garbage in Garbage out Bad data, Bad model, Bad prediction
- 18. Machine Learning Type • Supervised - we know the correct answers • Unsupervised – no answers • Artificial Neural Network – like human neural network
- 19. Machine Learning Type • Supervised • Input data labeled : has correct answers • Specific purpose • Types • Classification for distinct output values • Regression for continuous output values • Unsupervised • Input data not-labeled : No correct answers • Exploratory • Types : Clustering (clusters) X0 X1 X2 … Xn Y X0 X1 X2 … Xn
- 20. Classification X1 • Categorical target • Binary or Multiple Classification • Example : Yes/No, 0 to 9, mild/moderate/severe • Logistic Regression, SVM, K-nearest Neighbor, Naïve Bayes, Decision Tree, Random Forests, XGBoost X2
- 21. Support Vector Machine (SVM) • SVM is one of the most powerful classification model (both binary & multi), especially for complex, but small/mid-sized datasets. • Best for binary classification • Easy to train • It finds out a line/ hyper-plane (in multidimensional space that separate outs classes) that maximizes margin between two classes. • parameters – Kernel (linear, polynomial), gamma, margin(a separation of line to the closest class points) *** SAS codes; proc svmachine data=x_train C=1.0; kernel linear; input x1 x2 x3 x4 / level=interval; target y; run;
- 22. SVM in SAS Visual Data Mining and Machine Learning SAS Machine Learning portal can provide an interactive modeling.
- 23. SVM in SAS Visual Data Mining and Machine Learning SVM in SAS Visual Data Mining and Machine Learning
- 24. Python codes for SVM #import ML algorithm from sklearn.svm import SVC #prepare train and test datasets x_train = … y_train = …. x_test = …. #select and train model svm = SVC(kernel=‘linear’, C=1.0, random_state=1) svm.fit(x_train, y_train) #predict output predicted = svm.predict(x_test)
- 25. Decision Trees The widely used machine learning algorithm that use tree to visually and explicitly represent decision and its conditions. It is used for both classification (more popular) and regression. Why Decision Trees? • Easy to understand and interpret • Require a very little data preparation • Possible to validate model using statistical tests.
- 26. Python codes for Decision Tree #import ML algorithm from sklearn.tree import DecisionClassifier #prepare train and test datasets x_train = … y_train = …. x_test = …. #select and train model d_tree = DecisionClassifier(max_depth=4) d_tree.fit(x_train, y_train) #predict output predicted = d_tree.predict(x_test)
- 27. Decision Tree in SAS Visual Data Mining and Machine Learning
- 28. Python in Machine Learning Most popular programming language in Machine Learning • Scikit-Learn • simple and efficient ML tool, open-source • Classification, Regression, SVM, Clustering, Dimensionality Reduction, Decision Trees, Random Forests • TensorFlow • developed by Google, open source since Nov.’2015 • Artificial Neural Network, Convolutional NN, Recurrent NN, Autoencoder, Reinforcement Learning
- 29. Logistic Regression • Output: Binary target (Yes/No) • Input: Continuous/categorical variables • Example : predicting if the tumor is malignant based on tumor size Tumor size (mm) Yes (1) No (0) Yes (1) No (0) Tumor size (mm) Malignant?Malignant? Threshold : 0.5
- 30. Logistic Regression Architecture • Hypothesis function : hθ(x) = 1 / (1 + e-z) where z= ( θ0x0 + θ1x1 + θ2x2 + θ3x3 +,, + θnxn ) x1 x2 x3 xn ∑ f(z) θ1 θ2 θ3 θn x0 z hθ(x) • What are the parameters?θ0
- 31. Python codes for Logistic Regression #import ML algorithm from sklearn.linear_model import LogisticRegression #prepare train and test datasets x_train = … y_train = …. x_test = …. #select and train model lr = LogisticRegression() lr.fit(x_train, y_train) #predict output predicted = lr.predict(x_test)
- 32. Regression x • To predict values of a desired target quantity when the target quantity is continuous. • Output: Continuous variables • Example : predicting house price per sqft • Types: Linear Regression, Polynomial Regression, Decision Tree, Random Forest • Hypothesis function: • hθ(x) = θ0x0 + θ1x1 + θ2x2 + θ3x3 +,, + θnxn • y = 10 + 2x • Accuracy of Regression model • Square Error(SE): sum (y – pred)2 • Mean Square Error (MSE) = (1/n)*SE • Root Mean Square Error (RMSE) = square root of MSE • Relative MSE (rMSE) = MSE / Var(y) • R2 = 1 - rMSE y
- 33. Python codes for ML Linear Regression #import ML algorithm from sklearn import linear_model from sklearn.metrics import mean_squared_error #prepare train and test datasets x_train = … y_train = …. x_test = …. #select and train model linear = linear_model.LinearRegression() linear.fit(x_train, y_train) #predict output predicted = linear.predict(x_test) #evaluate model (MSE) Mean_squiared_error(y_test, predicted)
- 34. Unsupervised Machine Learning • Not-labeled input data : no correct answers • Exploratory • Clustering : the assignment of set of observations into subsets (clusters) so that data in the same clusters are more similar to each other than to those from different clusters. • Algorithms : k-means • Industry implementation - the grouping of documents, music, movies by different topics, finding customers that share similar interests based on common purchase behaviors. • Customer segmentation • Document clustering • Image segmentation • Recommendation engines
- 35. k-means clustering • The simplest and popular unsupervised machine learning algorithm • To group similar data points together and discover underlying patterns. To achieve this objective, K-means looks for a fixed number (k) of clusters in a dataset. • the K-means algorithm identifies k number of centroids, and then allocates every data point to the nearest cluster, while keeping the centroids as small as possible. How it works 1. Randomly pick k centroids as initial cluster centers. 2. Assign each sample to the nearest centroid. 3. Move the centroids to the center of the samples that were assigned to it. 4. Repeat 2 and 3 until the cluster assignments do not change under a tolerance.
- 36. Python codes for k-means #import ML algorithm from sklearn.cluster import KMeans #prepare datasets of X whose size is (100,2) X = … # build k means model with 2 clusters kmeans = KMeans(n_clusters = 2, tol=0.0001, max_iter=100) kmeans.fit(X) #print the results of the value of centroids print(kmeans.cluster_centers_) # two centroids ---- array([[-0.94665068, -0.97138368],[ 2.01559419, 2.02597093]]) #print the results of the value of centroids print(kmeans.labels_) # 100 labels based on two centroids. ---- array([ 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1])
- 37. Artificial Neural Network (ANN) • Most powerful ML algorithm • Game Changer • Works very similar to human brain – Neural Network • Types – DNN, CNN, RNN • Architecture • Artificial Neurons • Input Layers • Hidden Layers • Output Layers
- 38. Human Neural Network vs Artificial Neural Network (ANN) ➢ Human Neurons : Neural Network – 100 billions ➢ Artificial Neurons : Artificial Neural Networks
- 39. Artificial Neuron • Input (x1, x2, x3,,,,xn) – numeric values that goes into neuron model • Weight (w1, w2, w3,,,,wn) – weight of each input values • Bias (b1) – bias of (input * weight) • Artificial neuron • Model (input values * weights + bias): z1 = x1w1,1 + x2w1,2 + x3w1,3 +,,,, + xnw1,n + b1 • Activation function : a1 = f(z1) x1 x2 x3 xn ∑ Activation function W1,1 W1,2 W1,3 W1,n b1 z1 a1
- 40. Neural Network in SAS using proc nnet Proc nnet data=Train; architecture mlp; hidden 4; hidden 2; input x1 x2 x3 ; target Y; Run;
- 41. Neural Network in SAS Visual Data Mining and Machine Learning
- 42. Python codes for DNN #import ANN - TensorFlow Import tensorflow as tf X = tf.placeholder(..) Y = tf.placeholder(..) hidden1 = tf.layer.dense(X, 4, activation=tf.nn.relu) hidden2 = tf.layer.dense(hidden1, 2, activation=tf.nn.relu) logits = neuron_layer(hidden2, 2) …. loss = tf.reduce_mean(….) optimizer = tf.train.GradientDescentOptimezer(0.1) traing_op = optimizer.minimizer(loss) tf.Session.run(training_op, feed_dict={X:x_train, Y:y_train})
- 44. Where is SAS in ML? SAS Visual Data Mining and Machine Learning • Linear Regression • Logistic Regression • Support Vector Machine • Deep Neural Networks ( limited layers)
- 45. How ML is being used in our daily life
- 47. Recommendation • Amazon • Netflix • Spotify
- 49. Amazon Go – Cashless Shopping
- 50. AlphaGO
- 52. Why is AI(ML) so popular now? • Cost effective • Automate a lot of works • Can replace or enhance human labors • “Pretty much anything that a normal person can do in <1 sec, we can now automate with AI” Andrew Ng • Accurate • Better than humans • Can solve a lot of complex business problems
- 53. Now, how Pharma goes into AI/ML market • GSK sign $43 million contract with Exscientia to speed drug discovery • aiming ¼ time and ¼ cost • identifying a target for disease intervention to a molecule from 5.5 to 1 year • J&J • Surgical Robotics – partners with Google. Leverage AI/ML to help surgeons by interpreting what they see or predict during surgery
- 54. Now, how Pharma goes into AI/ML market • Roche • With GNS Healthcare, use ML to find novel targets for cancer therapy using cancer patient data • Pfizer • With IBM, utilize Watson for drug discovery • Watson has accumulated data from 25 million articles compared to 200 articles a human researcher can read in a year.
- 55. • Novartis • With IBM Watson, developing a cognitive solution using real-time data to gain better insights on the expected outcomes. • With Cota Healthcare, aiming to accelerate clinical development of new therapies for breast cancer. • New CEO claimed Novartis as Medicine and Data Science company. Now, how Pharma goes into AI/ML market
- 56. • First FDA approval for medical device using AI/ML • FDA is open for machine learning algorithm based on devices. How FDA sees ML/AL
- 57. ML application in Pharma R&D • Drug discovery • Drug candidate selection • Clinical system optimization • Medical image recognition • Medical diagnosis • Optimum site selection / recruitment • Data anomality detection • Personized medicine
- 58. Adoption of AI/ML in Pharma • Slow • Regulatory restriction • Machine Learning Black Box challenge – need to build ML models, statistically or mathematically proven and validated, to explain final results. • Big investment in Healthcare and a lot of AI Start up aiming Pharma.
- 59. Healthcare AI/ML market • US - 320 million in 2016 • Europe – 270 million in 2016 • 40% annual rate • 10 billion in 2024 • Short in talents
- 60. PhUSE Machine Learning & AI Project Goals • To raise awareness in Machine Learning in Pharma Industry • To get more helps (volunteers and expertise) • To Introduce Machine Learning Education • To curate Machine Learning Materials • White papers • Projects
- 61. Machine Learning & AI Team • PhUSE Wiki in http://www.phusewiki.org/wiki/index.php?titl e=Machine_Learning_/_Artificial_Intelligence • Project of “Educating For the Future”
- 62. Why Machine Learning & AI Team?
- 64. • Kevin Lee at kevin.kyosun.lee@gmail.com • Nicolas Dupuis at ndupuis@protonmail.com • Wendy Dobson at wendy@phuse.com How to Volunteer