Sensor Networks for Computer Scientists

S17 Senior Seminar – January 2017

Phillip J. Curtiss, Assistant Professor Department of Computer Science

Montana Tech

Sensor Networks

• Collection of highly-distributed grid of devices that collect data, possibly transform and/or anonymize the data, and then transmit this data to one or more central processing sites for analysis and action.

• Such networked sensors can be orchestrated, working collectively or work independently. • Code must be developed to manage the devices and their sensors.

• Network code (stacks) must be developed and used to connect devices to networks – LANs, WANs, MANs – different media types – Cu, Wireless, OC, Cell, etc.

• Data Structures and Data Management – Protecting against data loss, data integrity, privacy and security concerns

• etc.

Sensor Networks - Example

• Meteorology – METAR • Managed by the National Climate Data Center (NCDC), part of NOAA

• Network of sensors to provide the data that feeds the weather and climate models that produce Aerodome Forecast Information – Weather predictions

• Automated Weather Observing System (AWOS) • Over 600 AWOS sites

• Automated Surface Observing System (ASOS) • Largest Distribution of Sensors for Weather Collection

Sensor Networks - Examples

Sensor Networks - Example

• Border Control– Watchkeeper • Managed by General Atomics (Defense Contractor) • Network of autonomous sensors to provide reconnaissance data related to

boarder protection and surveillance.

• Automated Autonomous Network of Sensors – Aerial and Ground • Self-Learning and Self-Healing Network

• Sensor Compliment • Proximity and Mass Detection • Sound – Recording and Transmission • Video – Recording and Transmission • Proprioception – Notion of Where it is in relation to goal space

Sensor Networks - Example

• Automated Traffic Cones – Department of Transportation • Automated Deployment and Configuration of Traffic Patterns

• Saves Hundreds of Human Road Workers’ Lives per Year

• Cost Effective Alternative

• Automated Autonomous Network of Sensors –Ground • Self-Learning and Self-Healing Network

• Sensor Compliment • Proximity

• Proprioception – Notion of Where it is in relation to goal space

Sensor Networks - Example

• Waste Collection – Government Application of IoT • Just In Time Servicing of Trash Receptacles

• Saves Hundreds of money – As much as 40% per year

• Environmental impact

• Automated Autonomous Network of Sensors – Wireless • Route and Fleet Management

• Sensor Compliment • Weight, Smell, Composition

• Enables effective use of Resources for Fleet Management

Sensor Networks - Example • Other examples

• Access Control Systems – Sanction Access to facilities via Keycard, RFID, etc. • Medical Telemetry Systems • Aircraft Navigation Systems • Earthquake Detection Systems • Energy Monitoring and Metering Systems • SCADA Systems – for Large Scale Monitoring, Reporting, and Control • Transportation Systems – Camera, Day/Night, Light Sensors, etc. • Agriculture and Farming • etc.

• Great reference for Application of Wireless Sensor Networks, Antoine Bagula, UCT

Human Perception

The organization, identification, and interpretation of sensory information in order to represent and understand the environment.

Two Broad Categories of Processes involved in Perception: • Processing Sensory Input – Transforming Low-Level Information to High-Level

Information External distal stimulus excites one or more senses to create a proximal stimulus where the brain creates a percept of the external object – transduction.

• Cognitive Processing – Processing connected with a person’s concepts and expectations and selective mechanisms that affect perception.

Human brains are structured in a modular way, with different areas of the brain processing different sensory information.

The ``Jailed Brain’’

Your Brain is in a Jail - your Cranium

• A Jail without any Light

• A Jail without anything to Taste

• A Jail without anything to Smell

• A Jail without anything to Touch

Your Brain “knows” the world through its perception of the world. It forms its perception through your senses, the transduction process.

Perception is not Perfect

Perception is not Perfect

Perception is not Perfect

Perception is not Perfect

Perception is not Perfect

Perception is not Perfect

• Shepard Tone – An Auditory Asymptote

• Use of Prior Information to Bias Perception

How does our brain make sense of the world?

• One of the oldest questions in philosophy – Descartes’ Term “The Veil of Darkness”

• Neurophysiologists, Neuroanatomists, Cognitive Scientists, Philosophers, Linguists, etc. – Trying to answer this question, or at least understand constituents that are required to answer the question.

• Perception seems to be sensor fusion comingled with knowledge, memory, and attention to form a holistic view – or at least dimensionally continuous.

What about a computer’s perception of its world? • Aren’t computer’s “jailed” in an analogous way as our brains?

• Aren’t such computers, even if distributed, relying on inputs as the basis of transforming state, and thereby computing?

• Are the inputs reliable representations of the world? • Do we have digital transduction correct?

• Are we guilty of confirmation bias in the formation of sensor networks?

• In the end, is perception required, or just sensor fusion?

• What’s the difference?

Want to build a computer system to take over the tasks of driving a car? • We need to sense the world around us?

• Do we need more than this? Do we need a notion of perception?

• Is mere “digital transduction” sufficient for making moral decisions about the occupants of the car? Bystanders on the road?

• Today – Each car contains about 200 individual sensors. (Car & Driver 2016)

• Driverless Cars – About the same number.

• Difference – Sample Rate – order of magnitude greater for Driverless

• Difference – Synthesis – More Computing for Driverless v. Microcontrollers

Want to build a computer system to operate a modern farm? • We need to sense the world around us?

• Do we need more than this? Do we need a notion of perception?

• Is mere “digital transduction” sufficient for making moral decisions about what to plant for the different markets? Food, Energy, Textile, Medical?

• Need to predict market forces for all these markets at time of crop maturity; early enough to course correct and plant different compliment

• Don’t my neighbors’ decisions affect mine? How do I avoid homogeneity?

• Today – Modern Farms use Drones, Satellite imagery, Field Sensors on Equipment

• Need Predictive analytics; Need Algorithms; Data Science on a Large Scale

• Need Models for How to do this synthetically – “Agricultural Simulacrum”

Internet of Things/Internet of Everything

34B devices connected to the Internet by 2020 IoT devices will account for 24B of that total $6T will be spent on IoT solutions over the next 5yrs Business Proposition:

Lowering Operating Costs Increasing Productivity Expanding/Developing New Markets

Governments:

Improving Citizens Quality Measures Second largest adopter of IoT/IoE Solutions

Consumer Market:

Nest, Google Home, Amazon Alexa – Nexus devices

John Greenough and Johnathan Camhi, Business Insider, Tech Insider 8/29/16

Case Study: ID Genomics • Solving ``Broad Spectrum Antibiotic’’ Paradox

• Broad spectrum antibiotics protect against a wide variety of bacteria - the so-called gram(+)(-) bacteria. • Tetracycline, Ciprofloxacin, Levofloxacin, Penicillin, Cephalexin

• Narrow spectrum antibiotics exist as well – generally more effective, more of the time, and fewer side-effects.

• The effectiveness of using broad spectrum is enticing – no need to perform additional culture studies to understand which narrow spectrum to use; works most of the time, one or the other can be switched out if resistance is observed; side-effects can be managed

• The obscene reliance on broad spectrum antibiotics has rendered sensitivities to narrow spectrum counterparts, and has caused mutations in bacterial populations – superbugs.

Case Study: ID Genomics • Using genomics – genetic pattern matching based on genome

sequencing – ID Genomics has developed a test that identifies the sensitivity of certain bacteria to all known narrow (and wide) spectrum antibiotics in about 30 minutes • No culturing, no wait time • The result is a heat-map graph showing the efficacy of each antibiotic on a

given bacterial strain

• Outcomes: • Quells the overprescribing of wide-spectrum antibiotics (CloNet) • Generates tremendous data and meta-data (CloNet) • Doesn’t address bacterial mutation due to treatment sensitivities (BactNet)

Case Study: ID Genomics • Goals of BactNet

• Use data analytics and machine learning to work at the “population” or “cluster” level to reason (infer) trends and avoid sensitivities

• Applications: • Syndromic Surveillance

• Reduction in Bacterial Mutations as a response to treatment sensitivities

• Research into: • Outcome Driven Treatments

• Remediation of socio-economic factors relating to healthcare

• Correlation - understand between disease processes and treatments

Case Study: ID Genomics • BactNet – Barrier is information, computation, algorithms

• IoT/IoE – dsRTU (data-science Remote Terminal Unit) • Data Collection Models – Algorithms – for:

• Data Ingestion – Data Models, Network Algorithms, Cybersecurity • Data Tracking – Anonymization, Data Structures, Privacy v Security • Data Analytics – Machine Learning, Predictive Modeling • Cluster-based – geographic, risk populations, Dx, Tx, Rx populations, etc.

• Computational Science, Computer Science, Software Engineering • All at play in a project such as this

• See Also Mount Sinai Icahn School of Medicine, Arnhold Institute for Global Health

Case Study: Hoplite Industries

• Rethinking the Cybersecurity Model • Modeling end-user behavior as a means of detection and remediation

• Ability to move security into the “software defined network” layer and not just at the border of the network

• Allows for the protection of network boundaries and not just network borders

• Target Markets: • Data Center Networks – Global Tier-1 Providers

• Telecommunications Networks - Cellular, VoIP, Satellite, Wireless

• Enterprises Networks

• Government and Military Networks

Case Study: Hoplite Industries

• Barriers to Achieving Goals • Extracting information from network flows at wireline speeds

• Developing Methods for mining the extracted information for actionable trends

• Developing a repository of “bad behavior” that can inform and classify existing and future trends of interest

• Opportunities: • Behavioral Model Development – Domain Specific, Tasks Specific, Early

Detection of “Imposters” (Behavioral Spoofing)

• Data and Storage models for quickly culling information under some semantic from large data repositories

Concluding Remarks

• Likely whatever industry in which you find yourself, Sensor Networks will be associated in one way or the other.

• Even if you are not directly interfacing with the Senor Network, you are likely using data from it, or possibly producing code for it.

• The Distributed nature of computing, and the fusion of sensor information as a source of real-time data, causes a need for analytics, models, data management, data structures, and algorithms to manage the autonomous nature, and to construct systems out of these distributed components. We have just begun how to build such systems - especially in a scalable, extensible, and heterogeneous way.

Questions?

Phillip J. Curtiss, Assistant Professor Department of Computer Science Montana Tech pcurtiss@mtech.edu 406-496-4807, pcurtiss@mtech.edu