transversal delivery pipeline by mike nescot and nick grace
TRANSCRIPT
Transversal Delivery Pipeline
Mike Nescot, Director of DevOps & Cloud SolutionsNick Grace, Director of UX & Web Development
JBS International, Inc,
● Data Science● Machine Learning/Deep Learning● Natural Language
Processing/Conversational Agents● Internet of Things ● Virtual Reality/Augmented Reality
Web: Transformations
● Intelligent Applications● Bots● Multimodal Systems● Industrial Product Design● Hardware● Health● Public Safety
DevOps: New Opportunities
● A state of development, preparation, or production: “several projects in the pipeline”; also: the system for such processes: “a strong product pipeline”
● A route, channel, or process along which something passes or is provided at a steady rate; means, system, or flow of supply or supplies:“Freighters and cargo planes are a pipeline for overseas goods.”
● Workflow, Lifecycle● Continuous Delivery, Deployment Pipeline● Contrast to silos
Pipeline
Development Test Production
Simple Web Development Pipeline
DevOps Pipeline
Transverse
● Velocity● Infrastructure as Code● Automation● Collaboration● Repeatable Process● Agility
DevOps Principles
● Big Data (3 V’s)● Data Driven Decisionmaking (Moneyball)● Predictive Analytics● Algorithms in Control
Rise of Data Science
Data Science Venn Diagram (Drew Conway)
● Programming● Data Analysis● Data Visualization
Data Science: Udacity
CleanStore
DeliveryTransform
Scrape
Scrapy MongoDB Pandas
Matplotlib
Flask
Jupyter
Explore,Analyze, Process
Pandas
D3
Data Science Pipeline
● Traditional statistics: statistical distributions (normal distribution/Bell curve, exponential distribution, binomial distribution). Linear and logistic regression to predict the data based on these numerical techniques.
● Machine learning: using the data to build the model itself with the aid of computers
Traditional Stats vs Artificial Intelligence/Machine Learning
● Recommendation Systems● Autonomous Vehicles● Physics● Real Estate● Finance
Machine Learning Applications
● “To make progress, every field of science needs to have data commensurate with the complexity of the phenomena it studies”
● Sciences that were data poor are now data rich:○ Sociology: graph databases ○ Neurology: connectomes
Pedro Domingos: The Master Algorithm
● Knowledge Engineers vs. Machine Learners● Five Tribes of Machine Learning:
○ Symbolists: Expert Systems○ Connectionists: Backpropagation, Neural
Networks○ Evolutionaries: Genetic Programming○ Bayesians: Bayes theorem, uncertainty○ Analogizers: Similarity, SVM
The Master Algorithm
ML Algorithms
● More neurons than previous networks● More complex ways of connecting layers
(RNN, CNN)● More computing power to train● Automatic feature extraction
Deep Learning: Deep Neural Networks
Neuron
Artificial Neural Networks
Deep Neural Network
● 2012: AlexNet: Superhuman visual pattern recognition
● 2012: Google Speech Recognition● 2015: Amazon Echo● 2016: AlphaGo● 2016: Google Search
Deep Learning Milestones
● Parameters● Layers● Activation functions● Loss functions● Optimization methods● Hyperparameters● Backpropagation
Deep Neural Networks Architecture
Sampling(Datasets)
Data Processing
Model Selection
TrainingValidation
Data CollectionTransformation
(Feature Engineering)
Standard Machine Learning Pipeline
● Apache Kafka● Apache Spark● Apache Hadoop● Apache Zookeeper● Apache Mesos● Torch, Caffe, TensorFlow
New Big Data/ML Open Source Technologies
● Real-time, Distributed (Streams)● More Data, Bigger Networks● “End to End Deep ML”● Serving Models, Continuous Delivery
New ML Workflow
Real-Time Deep ML Pipeline: PipelineIO
● Data Science is Application Development● Ops Teams in Data Science● Continuous Feedback● Use of Stats/ML in DevOps (e.g., Anomaly
Detection)
DevOps and Data Science Converge
Previously: “a software application that runs automated tasks (scripts) over the Internet.”
Now (AI/ML):
● Conversational Interface● Personal Assistant● Digital Agent
Bots
● Advances in AI/ML● Advances in NLP● Mobile● Messaging● Social Networking
Rise of Bots
● Personality● Name● Purpose (God vs. special purpose)● Interaction
Bots
● Context● Generative vs. Retrieval Based● Purpose● Diversity● Tone● Interaction● AI (Language, Image)
Conversational Interface Characteristics
● Question Answering● Recommender Systems● Summarization● Human Augmentation● Sentiment Analysis
Bot/NLP Applications
Parts of Speech Tagging
Quotation speaker identification Character name
clustering
Lemmatization
Dependency parsingNamed entity recognition
Text (Corpus) Tokenization
Pronominal coreference resolution
NLP Pipeline
Speech Synthesis
Processing
Natural Language Generation
Speech Recognition
Natural Language Understanding
Translation
Service Delivery
Conversational Agent Pipeline
● Complexity● Context● Ambiguity● Slang● Humor, sarcasm● Dynamics
NLP Challenges
● Visual● Auditory● Haptic● Kinesthetic● Proxemic
Multimodal/Multisensory Systems
InputOutputMerge
Transfer
Translate
Substitute
Visual
Haptic
Auditory
Kinesthetic
Proxemic
Deliver Visual
Haptic
Auditory
Kinesthetic
Proxemic
Multimodal Delivery Pipeline
● Home Appliances ● Industrial Equipment ● Medical Devices● Vehicle Components● Soil Sensors
IoT Smart Products
● Physical Components: mechanical and electrical parts ● Smart Components: Sensors, microprocessors, data
storage, controls, software, embedded operating system, digital user interface
● Connectivity Components: ports, antennae, protocols, and networks that enable communication between the product and the product cloud, which runs on remote servers and contains the product’s external operating system
Smart Products
● Instrumented: Sensors detect conditions and changes in their surroundings. Light, radiation, motion, heat, humidity, vibration, sound, magnetic fields. Data collectors. Actuators, control system or mechanism that acts on environment.
● Intelligent: Embedded microprocessors, knowledge bases, user profile information. Make decisions, optimize outputs, adapt to environment, trigger actions, customize UX.
● Interconnected: Wi-Fi or other, share data and decisions with people or other products. Smart networks.
Smart Products
● Product Cloud● Devices● Registry● Messaging ● Data Collection/Analysis● Digital Twin/Shadow
IoT Architecture
● Continuous Verification● Collaboration Across Engineering Disciplines● Open Data● Link Product, Market● Extension to V Model (DoD ITS, DoT, Germany)● Focus on Running Systems or Virtual Models● Engineering Data Analytics
IBM: Continuous Engineering
V Model
Abstract mechanics, electronics, and software entities to create a virtual prototype to test your system before you build. Create executable models that enable early analysis and tests of the functionality, behavior, architecture, structure, performance, reliability, and safety of the system early in the development process.
Virtual Models
Electrical/Electronics Design
Mechanical Design Requirements
Design
Testing
Manufacturing
Software Design
Virtual Model/Digital
Twin
Prototype
SmartProduct
Consumer
Continuous Engineering Pipeline
● Virtual Reality (Oculus, Vive)● Augmented Reality (Pokemon
Go, HoloLens, Project Tango)● Mixed Reality
Virtuality Continuum
● 3D-World: Surfaces, Objects, Boundaries.● Virtual Objects: Shapes, Textures, Position
in the real world.● Motion Tracking● Depth Perception● Area Learning
Augmented Reality
● Marker vs Markerless AR● GPS: Pokemon Go● Simultaneous Localization and Mapping (SLAM):
3D Buildings● BIM (Building Information Model) and CAVE
(Computer Augmented Virtual Environment)
Simple AR vs Complex AR
BIM and CAVE
Virtual Realm
Physical Realm
Mediated Realm
MutimodalSystems
MutimodalSystems
User
VR/AR Pipeline
● Connectivity● Web Capabilities: WebVR, WebRTC, Web
Workers, Device access● Device Capabilities: Embedded operating
systems (RTOS, Linux, Docker)● Security● Protocol: HTTP vs. MQTT
IoT and VR/AR: Cloud/Web vs Mobile/Device
● Cathy O’Neill: “Weapons of Math Destruction:○ Predictive Policing○ Employment○ Education○ Finance
● Unreproducible research
Data Driven Society: Problems
● Platform for Interactive Science● Web IDE● Computational Narratives● Multimedia● Executable Papers● Algorithm Collaboration● Data Journalism● JSON
Jupyter Notebook
Jupyter Notebook
Publish Clone/Fork Validate/Update
Pull RequestResearcher
Researcher
Notebook Pipeline
● Precision Medicine Initiative● Wearables: IoT Sensors
○ Blood Pressure○ Pulse○ Sleep○ Calories○ Activity○ Glucose
Health/Medicine
● SDN: Software Defined Networking● WAN: Wide Area Network ● LAN: Local Area Network ● MAN: Metropolitan Area Network● PAN: Personal Area Networks● CAN: Car Area Networks● BAN: Body Area Networks
Delivery Pipelines: Networks
Ye Sun: BAN and CAN
● Unified Data Warehouse● Pipeline Options (service vs platform)● Life is just a just a collection of
microservices, we’re all headed towards the final deployment
Transversal Delivery Pipeline