1. Learning Paradigms
1.1. Supervised
1.1.1. Regression
1.1.1.1. Learning Algorithm
1.1.1.1.1. Non Parametric
1.1.1.1.2. Parametric
1.1.2. Classification
1.1.2.1. Fitting Models
1.1.2.1.1. Linear Regression
1.1.2.1.2. Logistic Regression
1.1.2.2. Multi i/p case
1.1.2.2.1. 2- dimension
1.1.2.2.2. 3- dimension
1.1.2.2.3. ...
1.1.2.2.4. infinite dimension
1.2. Unsupervised
1.2.1. Clustring
1.2.1.1. examples
1.2.1.1.1. Image Processing
1.2.1.1.2. Organizing computer Clusters
1.2.1.1.3. Social Network Analysis
1.2.1.1.4. Market Segmentation
1.2.1.1.5. Astronomical data analysis
1.2.1.1.6. Understanding genes data
1.3. ReInforcement
1.3.1. Applied to
1.3.1.1. Web-Crawling
1.3.1.2. Robotics
1.3.1.2.1. teaching Robot to get over obstcales
1.3.1.2.2. teaching Car drive off road and avoiding obstacles
1.3.1.2.3. robotic snake
1.3.1.2.4. 4-legged robotic dog
1.3.2. idea
1.3.2.1. Reward function
2. Learning Resources
2.1. Deeplearning
2.1.1. first commercial-grade, open-source, distributed deep-learning library written for Java and Scala
2.1.2. Integrated with Hadoop and Spark, Deeplearning4j is designed for business environments and includes a distributed multithreaded deep-learning framework and a single-threaded deep-learning framework
3. Keywords
3.1. Machine only understand numbers. It starts with obvious things and extends to subtle things
3.2. Feature engineering
3.2.1. Finding connections between variables and packing them into a new discreet variable
3.3. The Power to Predict Who Will Click, Buy, Lie, or Die. [Predictive Analytics is) technology that learns from experience [i.e. data] to predict the future behavior of individuals in order to drive better decisions
3.4. Pairing human workers with machine learning and automation will transform knowledge work and unleash new levels of human productivity and creativity
4. Practical Applications
4.1. Shopping Recommendation
4.1.1. better sales automation, lead generation, efficient marketing, predictive hiring, algorithmic trading
4.2. Churn Prediction
4.3. Face tagging
4.4. Sentiment Analysis
4.4.1. Understand emotions regardless of language written
4.5. Youtube uses a deep neural network model to generate higher-quality thumbnails for videos
4.6. Skype real-time translation
4.7. Video monitoring to identify interesting objects such as dogs and trucks
4.8. Google Translate, Voice, Photos. Google driverless car predict appropriate driving actions
4.9. Numenta's Grok predict future energy requirements and prices
4.10. LinkedIn predicts who you want to connect with
5. Future Applications
5.1. Medical imaging
5.1.1. simple chip utilizing cloud computing and deep learning models
5.2. Baidu Deep Speech
5.2.1. transcribe voice queries in Mandarin
6. Purpose
6.1. Increase Sales Performance
6.1.1. 1) Don’t worry if data isn’t 100% accurate to begin with. As long as it’s directionally correct it will stimulate the right discussions. Data quality will improve naturally with use, feedback, updating, and iterative cleansing. 2) Drive excitement and adoption by making the application simple and engaging for the field, with easy-to-understand, interactive visualizations. 3) Integrate predictive analytics into the visualization and discovery process on a self-service basis so that new insights are intuitively delivered as the underlying data and attributes change. This will keep the insights from the application relevant. 4) Use iterative techniques to design and deliver a working app quickly and then adapt it based on user feedback. 5) Partner with IT through this process so that the users receive the desired self-service and flexibility while leveraging the business intelligence platform to maintain data governance, security, and control
6.2. Increase number of customers
6.2.1. reducing attrition/churn using historical data and look for likelihood of churn
6.2.2. acquiring new customers by lead scoring and optimizing marketing campaigns
6.3. Serve customer better
6.3.1. cross-selling products
6.3.2. optimizing products and pricing by mapping product characterizations to no. of sales
6.3.3. Increasing engagement by observing customer behavior and mapping customer-item pairs to interest indicators
6.4. Serve customer more efficienctly
6.4.1. Predicting demand. Observe high variability for services/products
6.4.2. Automating tasks such as scoring credit applications and insurance claims
6.4.3. Making enterprise apps predictive in prioritize things, use adaptive workflows (route customer support requests to best available person), adapt the interface, set configurations and preferences automatically
7. Enterprise
7.1. Security/Fraud
7.1.1. BrightPoint Sentinel automate threat detection and risk analysis
7.2. HR/Recruiting
7.2.1. Textio analyzed job text and outcomes data using listings from tens of thousands of companies
7.2.2. hiQ People Analytics helps employee selection, development and retention by modeling historical data to predict future outcomes
7.3. Sales
7.3.1. Sentient Aware uses visual search to help shoppers quickly find the products they want to buy just like a store associate, connecting the right products to every customer
7.4. Marketing
7.4.1. LiftIgniter improves CTR, engagement and conversion by providing personalization using recommendation in real-time
7.5. Customer Support
7.5.1. Clarabridge collects customer feedback from various sources and provide actionable insights
7.5.2. Quantifind tells what's most important in driving people to buy your products by introducing brand strategy. Explanatory analytics
7.5.2.1. potential replacement for survey-based consumer research, brand health studies, focus groups, strategic consulting engagements, etc.
7.6. Internal Intel
7.6.1. Using the combination of machine learning and crowdsourcing from experts identified by their usage of tables, Alation centralizes the knowledge on data and ensures it’s always up-to-date
7.7. Market Intel
7.7.1. Mattermark mines and crunches public Internet data to provide investors, sales teams and others with search tools and other business intelligence,
8. Algorithms
8.1. Deep Boltzman Machine (DBM)
8.2. Deep Belief Network (DBN)
8.3. Convolutional Neural Networks
8.4. Stacked Auto Encoders
8.5. Hierarachical Temporal Memory
9. Deep Learning
9.1. Definition
9.1.1. Deep Learning is an algorithm which has no theoretical limitations of what it can learn; the more data you give and the more computational time you provide, the better it is. (Geoffrey Hinton)
9.2. Domains
9.2.1. Image
9.2.1.1. Faces
9.2.1.2. Machine vision
9.2.2. Sound
9.2.2.1. Speech to text
9.2.2.2. Machine translation
9.2.3. Text
9.2.3.1. Search
9.2.3.2. Information retrieval
9.2.4. Time Series
9.2.4.1. Weather
9.2.4.2. Biodata
9.2.4.3. Stocks
9.3. Why
9.3.1. CIO looking for highest performance in ML
9.3.2. Advance the state of the art in pattern recognition and natural language processing
9.3.2.1. attempts to model high-level abstractions in data
9.4. Differences
9.4.1. "Normal" neural networks usually have one to two hidden layers and are used for SUPERVISED prediction or classification.
9.4.2. SVMs are typically used for binary classification, but occasionally for other SUPERVISED learning tasks.
9.4.3. Deep learning neural network architectures differ from "normal" neural networks because they have more hidden layers. Deep learning networks differ from "normal" neural networks and SVMs because they can be trained in an UNSUPERVISED or SUPERVISED manner for both UNSUPERVISED and SUPERVISED learning tasks
9.5. Data Scientist seek to process huge amount of unstructured data
10. Definition
10.1. computers learning to predict from data
10.1.1. learning implies improvement through gaining experience or knowledge
10.2. A (machine learning) computer program is said to learn from experience E with respect to some class of tasks T and performance measure P, if its performance at tasks in T, as measured by P, improves with experience E. (Tom Mitchell)
11. Applications
11.1. Natural Langauge Processing
11.1.1. deep LSTM (long short-term memory)
11.2. Syntactic Pattern Recognition
11.3. Search Engines
11.3.1. from lexical matching (matching terms) to latent semantic analysis (semantic matching) to deep neural network to extract high-level semantic representations
11.4. Medical Diagonsis
11.5. Detection Credit Card fraud
11.6. Stock Market Analysis
11.7. Classifying DNA Sequence
11.8. Speech & handwriting sequences
11.9. Object Recognition in Computer Vision
11.10. Game Playing
11.11. Robot Locomotion
11.12. Multimedia Signal Processing
11.12.1. Image
11.12.2. Speech and Audio
12. Why M.L ?
12.1. because we need to make machines ....
12.1.1. think like humans
12.1.2. notice similarties betwen things and generate new ideas
12.1.3. learn from mistakes
12.1.4. give explanation why things went wrong
12.1.5. Solve problems difficult or impossible for human to solve
12.1.5.1. problems
12.1.5.1.1. Phenomena are changing rapidly
12.1.5.1.2. Application need to be customized for each user separately
12.1.5.1.3. No human experts
12.1.5.1.4. experts unable to explain thier experience
13. SAP HANA
13.1. an in-memory platform that runs analytics applications smarter, business processes faster, and data infrastructures simpler
13.2. Predictive Analysis Library (PAL)
13.2.1. Association
13.2.2. Classification
13.2.2.1. To predict a binary answer – i.e. Is this transaction fraudulent or not?
13.2.3. Regression
13.2.3.1. To predict or score an amount that is a non-binary value - i.e. Determining the insurance risk factor this this driver
13.2.4. Cluster
13.2.4.1. To find groups in your dataset – i.e. Who are all the people likely to buy my product today?
13.2.4.2. To predict future values based on previously observed values – How likely are flight cancellations in winter vs. summer months?
13.2.5. Time Series
13.2.6. Probability
13.2.7. Outlier
13.3. Automated Predictive Library (APL)
13.3.1. customers, developers, and partners do not need to be data scientists to use the SAP APL – they simply need to feed the APL what they have and tell it what they need.
13.3.2. Classification, Regression, Clustering, TIme Series, Key Influencers
13.4. SAP Lumira
13.4.1. an agile data discovery tool designed to expedite data preparation and enable data to be presented in a visual, easily digestible form