Introduction to Machine Learning. Laura Seletos

Transcription

1 Introduction to Machine Learning Laura Seletos

2 INTERACTIVE DEMO I m in an awesome machine learning talk and I wanted to tell you

3 WHY Should You Care?

4 1 Autonomous Cars WHY Should You Care?

5 1 Autonomous Cars WHY Should You Care? 2 Search Engines

6 1 Autonomous Cars WHY Should You Care? 2 Search Engines 3 Spam Filters

7 1 Autonomous Cars WHY Should You Care? 2 Search Engines 4 3 Healthcare Spam Filters

8 1 Autonomous Cars 5 WHY Should You Care? 2 Search Engines Financial Trading 4 3 Healthcare Spam Filters

9 Your Smart Phone 6 1 Autonomous Cars 5 WHY Should You Care? 2 Search Engines Financial Trading 4 3 Healthcare Spam Filters

10 Your Smart Phone 6 1 Autonomous Cars Financial Trading 5 2 Search Engines 4 3 Healthcare Spam Filters

11 BIO: HELLO WORLD Career IT Security Engineer and Consultant ReliaQuest Information Security System Administrator Raymond James Financial Enterprise Technical Account Manager Qualys Inc. Education Stetson University Degree in Computer Information Systems Dual minors in Management Information Systems (MIS) and Business Administration Community Involvement Member of InfraGard Current Member and Advisor of ISSA Tampa Bay Organized ISSA Tampa Bay s 2016 "Women in Security" forum Organized and Mentor the ISSA Tampa Bay CTF 2016 Competition Speaker/Presenter at University of Tampa, Stetson University, ISSA South FL, ISSA Tampa Bay, and IANS Charlotte Notable Hobbies Home lab, crypto currencies, machine learning, automation, CTFs, video games, etc.

12 AGENDA Evolution of Machine Learning (ML) Key Terminology Theory, Structures, & Examples Cyber Security Use Cases AI & ML Failures Tools and Resources Questions & Discussions

13 EVOLUTION OF ML Source: ( ficial-intelligence-strategy)

14 KEY TERMS part 1 of 5 Artificial Intelligence (AI) The science of getting computers to act without being explicitly programmed Machine Learning (ML) A specific scientific method for building AI where an output is based on given goals and input data <Note> All Machine Learning is AI, but not all AI is Machine Learning </Note>

15 KEY TERMS part 2 of 5 Supervised Learning Training on a pre-defined set of examples Ensures more accurate conclusion with new data Unsupervised Learning No training Must find patterns & relationships in datasets Reinforcement Learning (RL) Reward feedback is given to teach behavior; this is known as the reinforcement signal Representation Learning (Feature Learning) Datasets that are not mathematically convenient Raw, real-world datasets like images, video, and sensor data Translates based on features or representations through examination without relying on explicit algorithms

16 KEY TERMS part 3 of 5 Classification (Supervised learning) The prediction variable takes class labels Example Predict the type of tumor (e.g. "benign" or "malignant") Regression (Supervised learning) The prediction variable takes continuous values Example Help with questions of How much? or How many? like house price (a real value) Clustering (Unsupervised learning) Analysis of data not included in pre-labeled classes Identifies & groups similar instances Example Pattern recognition, image analysis, information retrieval, bioinformatics, data compression, etc. Association (Unsupervised learning) Rule-based algorithm for discovering interesting relationships between variables in large datasets Example Cross-marketing & Customer behavior analysis Market Basket Analysis = Association between items chosen by shoppers

17 Source: Image via Abdul Rahid (

18 KEY TERMS part 4 of 5 Neural Networks (Neural Net) Just like any other network; Comprised of interconnected web of nodes, called neurons, and the edges that connect them together Main Functions: 1. Receive a set of inputs 2. Preform increasingly complex calculations 3. Use the outputs to solve a problem Nodes are assigned a number known as a weight In Training: Weights & thresholds are continually adjusted until labels consistently yield similar outputs Well suited for ML problems with gigantic inputs Neural Networks are used for lots of applications but this presentation focuses on Classification Source: uggets.com/2015 /01/deep-learning -explanation-what -how-why.html

19 Source: NEURAL NET EXAMPLE The classification is determined by the score of each node The act of passing an input value from one layer to another = Forward Propagation (aka Forward Prop)

20 KEY TERMS part 5 of 5 Deep Learning Process of applying deep neural network technologies, with multiple layers of neurons, to solve problems Deep learning is a specialized form of machine learning Source:

21 LAYERS OF AI Artificial Intelligence Machine Learning Representation Learning (Also known as Feature learning) Deep Learning Source: Ian Goodfellow s (Scientist at Google Brain) Book: Deep Learning (

22 ML STRUCTURE Source: research/artificial-intelligence-strategy

23 ML STRUCTURE part 2 Neural Network Classification Attributes: Source: es/47/types+of+neural+networks.jpg

24 ML WORKFLOW Source: research/artificial-intelligence-strategy

25 ML EXAMPLES

26 CYBER SECURITY EXAMPLES Insider Threat Detection Malware Analysis (variants) Network Analytics (abnormal) Incident Response Fraud Prevention Spam Detection Situational Awareness Support Chat / AI Training Structuring unstructured data Defensive & Offensive Pentesting (SQL Ex)

27 AI FAILS Facebook Shut down program after developers discovered AI had created its own language There was no reward to sticking to English language Target Predicted teen girl was pregnant before her family did Sent coupons for baby items according to client s pregnancy scores Microsoft s AI Chatbot Twitter teaches it how to be racist in less than a day Source: Source: /16/how-target-figured-out-a-teen-girl-waspregnant-before-her-father-did/# Source: ay-microsoft-chatbot-racist

28 AI FAILS personal favorite Banana Toaster Google researchers developed a psychedelic sticker the effectively tricks deep learning systems into classifying the image as a toaster. Source: -simple-sticker-can-trickneural-networks-intothin

29 ML TOOLS & RESOURCES Top 8 Programming Languages for ML Python, Java, R, C++, C, JavaScript, Scala, Julia Free YouTube Training Video Deep Learning SIMPLIFIED: The Series Episodes watch?v=b99uvkwzytq List of Beginner ML Projects with Tutorials m/machine-learningprojects-for-beginners m/tag/test-ai-coding-skillsprogramming-challenge/ Google.com! Source:

30 QUESTIONS? Laura Seletos Technical Account Manager Qualys, Inc Bridge Parkway Redwood City, CA M