Applying Geospatial Representation and Forecasting Models forImproving Chem-Bio-Rad Defense in Battlefield and CounterterrorismField Operations Martin Dudziak, PhD TETRAD Technologies Group, Inc. June 21, 2005
Introduction - Objectives• Smooth transition and integration of methods and systems for CBRNE in both combat, post-combat, and civilian environments• Integration of CBRNE prediction, forecasting, detection, countermeasures with geospatial representation and analysis (more than GIS)• Incorporation of several maturing technologies and familiar methodologies: – Mobile, wireless, wearable, portable – Platform-independence, “plug and play” – Commercial, conventional, cheap, familiar, cast-away – Inverse methods, nonlinear methods, hybrid probabilistic reasoning• Adaptation of CBRNE and GIS to changing models of conflict, warfare and military-civilian discipline/collaboration• “Reusable and reconfigurable” is not only about cost-savings Copyright 2005 2 TETRADTechnologies Group, Inc.
Introduction - Objections• Too ambitious a goal and too many differences between CBRNE situations in the combat field and homeland sectors - “apples and oranges”• Too difficult to attempt assimilation of high-noise sources and low-sensitivity sensors• Consumer-grade technology not sufficiently specialized or robust• Problem of false-positives, esp. in bio and rad domains• Requires massive deployment and training - too expensive and too long Copyright 2005 3 TETRADTechnologies Group, Inc.
Foundations• Nomad Eyes architecture for open-ended deployment of sensor-analyzers• Use of inverse methods (from wave scattering and subsurface imaging) with Bayesian and RETE reasoning for analysis of distributed array data• Focus on a few target problems and technical (sensing) solutions – radiation sensors – chemical (organo-phosphate) sensors• Role of the GS and GIS is threefold: – Locate sensor reports over time and provide correlation – Locate both at-risk and risk-potential humans, machines, resources – Predict likely targets and movements• High-speed real-time database “ETL” and other VLDB processing is necessary to keep track of changes in data collection and geospatial object movement Copyright 2005 4 TETRADTechnologies Group, Inc.
Nomad Eyes - Mobile Wireless Portable/Handheld Nets for an Asymmetric, Dynamic Countermeasure SystemFor Rad Terrorism but also for othertypes and necessarily looking for all,not only one Mobile units using both cellular and wireless internet/intranets Freeform but adhering to industry standards Incorporating the General Public Incorporating the commercial sector (advertising and consumer products) Asynchronous, Atypical, Asymmetric Sensor Fusion Copyright 2005 5 TETRADTechnologies Group, Inc.
Nomad Eyes™ Architecture and Principles (I)Prevention by Detection of Terrakt Planning OperationsMovement of multiple types of components (CBRNE) or likely source/deliveryTime-matching and space-matching of logically connective, supportive events“Sensor Fusion” of the Unordinary (Необычный) Kind - e.g., tracer readings perhaps not individually remarkable Photos of suspicious individuals and vehicles that have some “matches” Exceptional shipping orders, out-of-sequence, special-route, handling Parallel transit/shipment/transaction of non-contraband components useful in an RDDGoal toward Inverse Reasoning and Abductive Assimilation with other KBs / XsysFall-Back Value: Emergency public alerts and First-Responder capabilities Copyright 2005 6 TETRADTechnologies Group, Inc.
Nomad Eyes™ Architecture and Principles (II) EVENT ! Class (x) objects received by servers results in generation of n graphs representing hypothetical x y… relational maps; the majority are discarded, but events of interest trigger feedback to both autonomous and human-based nodes for additional collection and reorienting. No node or subset of nodes is reliant and the whole may be considered as a dynamic-geometry cellular automata. EVENT ! Copyright 2005 7 TETRADTechnologies Group, Inc.
Nomad Eyes = Compound Eyes Multiple TYPES of sensor data Multiple INSTANCES at multiple TIMES INVERSE Methods applied “as if” in surface/subsurface imaging: the task is to find what events and processes may be the modifiers of known or deducible behaviorsUSING•Abductive rules•Bayesian probabilistic inference•Fuzzy inference•Heuristics and “common sense” rulesFor all the value of sophisticated detectors, an “outlier” element or two could make all the difference:Requests for building or water/sewer line plans Repeat-visits of unusual vehicle or people“Non-sequitur” orders of shielding-quality materials Unusual change in shipping order or pickup Copyright 2005 8 TETRADTechnologies Group, Inc.
I3 Foundations Inverse, Nonlinear, Counter-Intuitive (sometimes)Source The Object causes diffusion and scattering of the Beam but the laws governing propagation and movement in different media are known or can be ascertained. Working backwards from the Result, one computes and estimates the Object on the basis of how the Beam must have changed in order to produce the Result instead of a pattern, computable, for what there would have been if no Object had been present. Now, transfer this Inverse Model ought of imaging and into the world of semiotics and intensions. Now, one can do inverse thinking from something Sensed and Observed, in actuality, to determine what were some of the intervening steps and processes out of the usual and ordinary process that would have produced something different, most likely less complex. Copyright 2005 9 TETRADTechnologies Group, Inc.
I3 Examples:Problem 1: Radioactive or chemical compounds are passing through a shippingport or through the public waterworks - how to identify a pattern and link a set ofevents and detections into a pattern that shows a natural or deliberate processwhich can be detected, localized, and treated with countermeasures?Problem 2: Small tumors or microscopic probes or nanosized drug delivery agentsare in the liver - how to accurately track, compare, recognize, and localize whenthe patient is moving and the body is constantly changing?The IRM (Inverse Relational Map) approach is one of several using inverseproblem modeling plus other nonlinear dynamic structures and functions in orderto produce not only usable answers but answers in real-time. Many of theunderlying mathematics and algorithms have been known and used before in otherdisciplines. Our approach is to try something new, primarily in the short cuts andspeed-ups gained through applying higher-level representations and heuristics thatcan significantly reduce the compute-cycle and delays. Copyright 2005 10 TETRADTechnologies Group, Inc.
Making Sense of the Data (I)• Basic diffusion equation - usable as starting point for inverse problems ∂ 2u 1 ∂u Particular credits - Roger Dufour, MIT 2 = u( x ,0) = f ( x ) u(0, t ) = u(a, t ) = 0 ∂x k ∂t• Time-transition is accomplished in Fourier domain ∞ x 2 a x f ( x ) = ∑ fn sin πn fn = ∫ f ( x ) sin πn dx n =1 a a 0 a ∞ n u( x , t ) = ∑ fne 2 −k ( πn a ) t sin π n =1 a• Transition backwards in time requires amplification of high frequency components - most likely to be noisy and skewed Copyright 2005 11 TETRADTechnologies Group, Inc.
Making Sense of the Data (II)• Heuristic and a priori constraints needed to maintain physical realism and suppress distortions from inverse process Particular credits - Roger Dufour, MIT• First-pass solution best match or interpolation among a set of acceptable alternatives ˆ x = arg min Ax − y s.t. x∈X x• Final solution may minimize the residual error and the regularization term 2 2 ˆ x = arg min Ax − y 2 + λ L( x − x ) 2 x Regularization offers fidelity to the observed data and an a priori determined (e.g., higher-scale-observed) solution model Copyright 2005 12 TETRADTechnologies Group, Inc.
Making Sense of the Data (III) • Diffusion _ Attraction • Modeling situations and schemas Particular credits - J. P. Thirion, INRIA as composite “images” in n-D • Iterative process with exploration of parallel tree paths – Speculative track; not required for Nomad Eyes sensor fusion to be useful to analysts – Purpose is to enable automation of the analysis and forecasting post-collection process – Area of active current research Copyright 2005 13 TETRADTechnologies Group, Inc.
Making Sense of the Data (IV) - I3BAT • Multiple modalities Sensor 1 Sensor 2 – Acoustic, EM, Optical, Text, NLP, SQL, AI-reasoning… • All looking at the same topic of interest (aka “region”) • Each sensitive to different physical/logical properties Property 3 – “Trigger” data – Contiguity (space/time) – Inference relations Property 2 – “Hits” with conventional DBProperty 1 queries (immigration, known associations, other Background investigations) • Compare with Terrorist Cadre Tactic models (schemas, maps) Copyright 2005 14 TETRADTechnologies Group, Inc.
CONFIG SETUP (ETLJOB ADB (Initialize) ADaM - making it real-time ETLSPEC) System & Meta Data Agent-Driven Data Mover ORCHESTRATOR (ORCH)• If you cannot collate, coordinate and efficiently MONITO R ETLP access the collected data, in Generato real-time, free-form (with functional r space respect to views and users) and without blocking users Transformer during backup and archiving Transformer periods, then you have a very Transformer inefficient database and it is Extractor Insrtor not conducive to the open- Extractor Insertor Extractor Loader ended purposes of BioScan Control Monitor or Nomad Eyes. Memory Memory• The ADaM software Docs outperformed that from NCR- Data Thread Docs and and Teradata with their own Files Memory Pool product as a data Files Databas Databas warehouse. It outperformed es Machine es ab Initio, a leader in the field space of Extract-Transfer-Load for Fortune 100 VLDB ADaM runtime modules Setup and configuration modules applications. ADaM runtime components Internal elements External data flow sources/destinations Copyright 2005 15 TETRADTechnologies Group, Inc.
Sensor Device Family• 1. OPA ™ Organo-Phosphate Analyzer – Nitrates, Organophosphates (e.g., Sarin, VX) (OPA ™) – OPA in beta development with matching-fund opps• 2. MagnetEyes ™ – Thin-film based magneto-optic sensing and imaging devices for desktop, industrial, and micro-scale applications in security, anti-counterfeiting, structural engineering, and biomedicine. Deployment-Ready• 3. BioScan ™ – Handheld wireless base for plug-compatible interface-standardized sensors and imaging• 4. Radiation sensors – Gamma and neutron detection – Compatible for GPS-locatable mobile wireless (telephony and wi-fi) devices Copyright 2005 17 TETRADTechnologies Group, Inc.
OPA ™ Portable Version The assay of OPs and other BChE inhibitors is achieved due to the use of nanostructured films based on polyelectrolytes and the bi-enzyme system cholineoxidase / butyrylcholinesterase (ChO/BChE). Conventional nerve agent organo- phosphates (Sarin, VX. GB) and carbamate type ChE-inhibitors can be detected at extremely low levels. Sensitivity for organophosphates (DFP, paraoxon, trichlorfon) is achievable @ 10 pM/L.• Automated version processes up to 24 samples For classical nerve agents the in sequence detection limits will be an order of• Portable unit can be adapted magnitude better; for instance, with air sampling and carbamates (carbofuran, carbetamid, condenser carbaryl) at @ 0.1 -1.0 nM/L. Copyright 2005 18 TETRADTechnologies Group, Inc.
OPA Comparative Sensitivity (1)Parameters Gas chromatograph GC with mass- PolyEnergetics spectrometer portable handheldSensitivity (SN – 1.0 0.5 0.1sanitary norm)System price (USD) 10K – 20K 150K – 400K 400Test cost (USD) 12 15 4Microchip sensor n/a n/a 1element cost (USD)Time to perform test hours hours 30-70 min.Sample preparation hours hours 10-20 min.Field analysis Not possible Not possible YesOrganic solvents Necessary Necessary NoReagent consumption High High LowSample volume No No Yes Copyright 2005 19 TETRADTechnologies Group, Inc.
Radiation sensor specs (targets) Parameter Range -25 -1 -80 -1Gamma sensitivity 200 +80 s (µSv/h) 2cps(µR/h) to 100 -25 s (µSv/h) 1cps(µR/h) +300 -1 +200 -1Neutron sensitivity 200 -25 s (µSv/h) 2cps(µR/h) to 100 -25 s (µSv/h) 1cps(µR/h)Gamma energy range 0.04 – 3.0 MeVNeutron energy range 0.03 – 3.0 MeVDose equiv. rate 1 – 5000 µR/hDose equiv. error +/- 30%False alarms < 1 per hourResponse time (gamma) < 2.5 sU detection 15g at 0.5m, velocity <= 0.5 m/s, background rad < 25 µR/hPu detection 0.5g at 0.5m, velocity <= 0.5 m/s, background rad < 25 µR/hIsotopes and materials U-235, U-238, Np-237, Puy-239, Pu-241, Cr-51, Ga-67, Pd-103, In-detectable 111, I-131, Tl-201, Xe-133, Co-57, Co-60, Ba-133, Cs-137, Ir-192, Se-75, Ra-226, Am-241 and othersBattery lifetime > 20 hrs. with average cell-phone usage (i.e., reduction of cell phone battery life to not less than one typical day)Weight < 100gDimensions smaller than 150mm x 50mm x 20mmCost per unit feasible to manufacture for under $50.00 in quantities > 10,000 Copyright 2005 21 TETRADTechnologies Group, Inc.
Today’s consumer-class RAD components Our simple conversion with Nomad Eyes™ Existing mobile phoneLi-ion A/D logic Nomadiks logic or other Rad-sensor element mProc Interface logic to wireless internet Copyright 2005 22 TETRADTechnologies Group, Inc.
23 CerviScan HEAD NT1004 Video Chip (*) TLWA1100 (*) NT1004 or LED NT1003 options (Array) TETRADTechnologies Group, Inc. CerviScan STEM Copyright 2005 Image Cam/LED Cam/LED Data Recognitio Control CollectionVersion 1 BioScan Architecture n / Processor Processor Processor Classifier Module (*) Module (*) Processor Module (*) (*) ST-20/40. ST FIVE, ARM7, StrongARM 20/40. ST FIVE, ARM7, StrongARM (Dragonball), CY8C2xxxx, xX256, TE502 (SoC or 16/32 micro + Flash + SRAM chipset solutions for 16/32 micro + Flash + SRAM chipset solutions for each logical module function each logical module function CerviScan BASE Belkin USB USB Cable Interface VideoBus II Logic Charger Li ion Lucent/ Interface Battery Proxim Wireless Logic
Conclusions• GSR / GIS databases can adapt to handling data produced by a Nomad Eyes type network• In each C-B-R-N-E category there exist today sensors with capability for inclusion in a distributed network of mobile wi-fi devices• Inverse methods can be successfully for accuracy and computational performance) be applied to the problem of analyzing massive amounts of low- accuracy, high-noise data from reporting sources• Interpretation of sensor-analyzer data will benefit from adjunct and meta data about the environment, such as provided by today’s GSR / GIS products• Universality and reusability of network collection and transmission devices simplifies human interface, training, time-lag and reduces errors. Copyright 2005 24 TETRADTechnologies Group, Inc.
Current Project Status Corporate-partnerships established and underway IRAD program (internally funded 1999 – 2004) I3BAT Phase 1 Sensors, phones, wi-fi networks ready ADaM prototype completed OpenNet prototype completed Collaboratory prototype completed Nomad Eyes Global “Peace Proliferation” Copyright 2005 25 TETRADTechnologies Group, Inc.
Current Technology Development Status • The electronics hardware for the mobile wireless image capture and collection has been radically simplified. • Pre-contract agreements with suppliers and partners in the electronics hardware domain have been established. • Matching fund agreements for phase-1 work have been obtained. • The software development has proceeded extensively during 2001-2004 and includes work using SOAR, GeNie, BNJ, JESS, and PNL, plus extensive work in the application of inverse method models. • Project work can be resumed and a substantial team of technical personnel can be activated within 1 to 3 months. Copyright 2005 26 TETRADTechnologies Group, Inc.
References • Early Nomad Eyes prototype including online co-development experiment http://tetradgroup.com/nomad/ • Early overview document (product oriented, high-level) http://tetradgroup.com/library/bioscan.doc • Technical documents and notes available, on archived CDs • Early published paper on the neural net component http://tetradgroup.com/library/bistablecam_ijcnn99.doc • ADaM extract-transfer-load system, critical for the super-fast movement of image data, triggering of agents, and coordination of images within patient-specific and feature-specific database views http://tetradgroup.com/library/ADaM_Design_Description1-1.doc • ADaM performance optimization, a key part of the system enabling massive throughput and parallelism for high-density imaging (not only for BioScan but more for MRI, CT, PET, 3d-ultrasound, digital x-ray) http://tetradgroup.com/ADaM_PerfOpt.doc Copyright 2005 27 TETRADTechnologies Group, Inc.
AcknowledgementsBy no means do we claim that those from whom we learn or collaborate will agreewith all of this, but thanks is due to more than a few… • J. P. Thirion, INRIA • Eric Klopfer, Henry Jenkins, MIT • Eric Miller, David Castanon, Clem Karl, CENSSIS • Tamara Koval, Forte Horizons LLC • Peter Sharfman, MITRE • ST Microelectronics SA • T-Mobile (Deutsche Telekom) • Attometrix • Lee Arzt • Jack Singer Copyright 2005 28 TETRADTechnologies Group, Inc.
Contacts• Martin Dudziak, PhD – (804) 740-0342 – (202) 415-7295 – firstname.lastname@example.org (also email@example.com) TETRAD Technologies Group, Inc. 28 Chase Gayton Circle, Suite 736 Richmond, VA 23238-6533 Copyright 2005 29 TETRADTechnologies Group, Inc.
BACKUP Material Copyright 2005 30 TETRADTechnologies Group, Inc.
OPA BACKUP Material Copyright 2005 31 TETRADTechnologies Group, Inc.
Basic Principles of OPA Operation (1)Amperometric analysis of organophosphates (OPs), carbamates and other specificand nonspecific inhibitors of butyrylcholinesterase (BChE).BChE activity) is inversely related to inhibitor concentrations.The analytical principle is based on the detection of hydrogen peroxide, releasedas a result of two consequent enzymatic processes: BChE Butyrylcholine + H2O → Choline + Butyryc acid (1) ChO Choline + 2O2 + H2O → Betain + 2H2O2 (2)Hydrogen peroxide is released at the final step and is detected through theelectrode. Copyright 2005 32 TETRADTechnologies Group, Inc.
Basic Principles of OPA Operation (2)Enzymes are fixed on a graphite support in the microelectrodes using layer-by-layer self-assembling nanofilm technology. At present, single-enzyme electrodesmodified by oxidoreductases (cholineoidase of tyrosinase) are available forsensitive chemical analysis of choline and phenol.The first prototype of the hand-portable measuring unit was developed and testedfor simple analyte detection: hydrogen peroxide, glucose, choline.This system is based upon the prior and currently available automated desktopsystem capable of processing up to 24 liquid samples per removable tray. Thissystem can be adapted to an air condenser system for processing upwards of 450L volumes into 10ml samples within approx. 10 minutes. Copyright 2005 33 TETRADTechnologies Group, Inc.
OPA Sensitivity (2)Numerous analytical approaches describing anticholinesterase detection arepublished every year in the scientific literature, but they remain distant frompractical commercial application that can meet the demands of widespreaddeployment, transit and movement, operations within intolerant physicalenvironments and conditions, and operation by personnel who are not experttechnicians. These are but a few of the problems that other systems face and thatour solution overcomes.A possible reason for the difficulties with other technical approaches andarchitectures is that primary attention is paid to the development of the sensitiveelement but not both the sensitive element and the measuring device. Because atpresent, water quality assays are based mainly on gas chromatography/gaschromatography with mass-spectrometry techniques. A brief comparison ofperformance characteristics with those that can be realized uniformly from thehandheld analyzer follows: Copyright 2005 34 TETRADTechnologies Group, Inc.
Rad Sensor BACKUP Material Copyright 2005 35 TETRADTechnologies Group, Inc.
BioScan BACKUP Material Copyright 2005 36 TETRADTechnologies Group, Inc.
Resources• The following images and charts give a snapshot introduction to a few of the tool components that were developed and applied in the BioScan R&D process. Not all of these images reflect BioScan directly, cervical cancer, or skin-related imaging.• These images are provided to show some of what was produced and can be deployed now to either a new Bioscan initiative or to other projects, unrelated to BioScan, for which the same expertise (including mathematical modeling, image analysis, electronics design and testing, database and knowledgebase implementation) can be very easily applied. Wireless Telemed Interface Macromolecular Networks Simulation Verite interactive pattern detection/classification Copyright 2005 37 TETRADTechnologies Group, Inc.
Resources (More) e-Presents conferencing andAnother Verite application, with EKG muilti-channel video streaming ADaM’s exceptional performance 16000 14000 12000 Typical Fastload Typical Tpump Typical Mixed 10000 Peak Fastload Rows/Sec Peak Tpump 8000 Peak Fstld & Tpump Transparent FastL Transparent Tpump 6000 Special FastL "Kitchen Sink" 4000 Peak ETL 2000 0 ed L p " p tL ad ad p tL p nk ET m m as um m ix as t lo t lo pu u Si pu M lF ak Tp SQL (Oracle) Data Server tF Tp s as T n tT al Fa ia Pe he en ak al F c & ec en pi c ak it c al ar Pe t ld pi Ty Sp c ar "K Pe sp Ty piinterface for image data mining Fs sp Ty an ak an Tr Tr Pe Te st Type Copyright 2005 38 TETRADTechnologies Group, Inc.
Resources (Still More) Screenshots of SOAR-based production-rule system Copyright 2005 39 TETRADTechnologies Group, Inc.
Nomad Eyes BACKUP Material Copyright 2005 40 TETRADTechnologies Group, Inc.
Where are the likely targets and means? In the public mind’s-eye and Angst And the less-likely form for many reasons Psycho-Shock is the Aim and Nuclear Radiation is Powerful even in non-lethal doses Mass-dispersion with uncertain contact and degree will create the most widespread fears Copyright 2005 41 TETRADTechnologies Group, Inc.
Newport-Norfolk (Hampton Roads) Copyright 2005 42 TETRADTechnologies Group, Inc.
Port of Baltimore > 30M tons per year, mainly containers2M+ residents in Baltimore and surrounding urban centerMain East-Coast rail and interstate highways traverse region Copyright 2005 43 TETRADTechnologies Group, Inc.
Deployment - Where and How• Static but ad-hoc – Passage locations and nexus points for cargo and transfer vehicles – Likeliest places but not limited to one configuration• Pseudo-random• Personal mobile units – Assigned to staff personnel – Personal cell phones• Unpredictable - a “two-edged sword” that cuts in in favor of the Defenders – Inverse predictive models can be applied better to the data “mass” – Al Qaeda (or “X”) cannot predict where are our eyes and ears• Sun Tzu (“Art of War”) - Always Make Your Enemy Nervous Copyright 2005 44 TETRADTechnologies Group, Inc.
First Responder Capability as well Notify Maximum Numbers of People ASAP after Terrakt Redirect Survivors Keep Other People Away Assist People Finding Loved Ones Provide Essential Life-Saving Information Real-TimeCoordinate and Inform First-Responder Teams Locations of People Active Sensor Array including useful data from public Coordinate with volunteers Copyright 2005 45 TETRADTechnologies Group, Inc.