TDR System - A Multi-Level Slow Intelligence System for Personal Health Care Shi-Kuo Chang, JunHui Chen, Wei Gao and Qui Zhang University of Pittsburgh chang@cs.pitt.edu

Slow Intelligence System The slow intelligence system is an approach to design environment aware systems. The general characteristics of a slow intelligence system include enumeration, propagation, adaptation, elimination, concentration and multiple decision cycles. In our previous work, an experimental test bed was implemented that allows designers to specify interacting components for slow intelligence systems.

An experimental TDR system The TDR system consists of multi-level super-components each with different computation cycles specified by an abstract machine model. The TDR system has three major super-components: Tian (Heaven), Di (Earth) and Ren (Human), which are the essential ingredients of a human-centric psycho-physical system following the Chinese philosophy. This experimental TDR system provides a platform for exploring and integrating different applications in personal health care, emergency management and social networking.

Outline Multi-Level Slow Intelligence System An Abstract Machine Model Compiler for Abstract Machine TDR System Architecture Tian Super-Component Web GUI Discussion

Outline Multi-Level Slow Intelligence System An Abstract Machine Model Compiler for Abstract Machine TDR System Architecture Tian Super-Component Web GUI Discussion

Multi-Level Slow Intelligence System

Multi-Level Slow Intelligence System

Outline Multi-Level Slow Intelligence System An Abstract Machine Model Compiler for Abstract Machine TDR System Architecture Tian Super-Component Web GUI Discussion

An abstract machine Model An abstract machine for slow intelligence systems: Msis = { P, S, Po, cycle1, cycle2, …, cyclen} The problem space P is a nonempty, enumerable set of problem elements p1, p2, …, pm. A problem set Pk is a finite subset of P. The solution space S is a nonempty subset of the problem space P. A solution set S is a finite subset of S. Starting from an initial problem set Po, the objective of the abstract machine is to derive a problem set Pj that is also a solution set, by applying one or more of the computation cycles cycle1, cycle2, …, cyclen.

An abstract machine Model A computation cycle is a sequence of slow intelligence operators to transform problem sets. Two elementary slow intelligence operators: Enumerator: P1 -enum< P2 is the enumerator that takes each problem element of P1 to generate a number of new elements, which are put into P2. Therefore the cardinality of p2 is usually larger than that of P1. Eliminator: P1 >elim- P2 is the eliminator that eliminates non-solution elements of P1 and put the rest into P2. Therefore the cardinality of P2 is usually smaller than that of P1.

An abstract machine Model cycle1: [guard1,2] Po -enum< P1 >elim- P2 A guard for cyclei is a predicate of the form guardi,j defined on problem sets and evaluated whenever a problem set is generated. If the predicate is evaluated to be true then control is transferred to cyclej. If the predicate is evaluated to be false then Msis remains in the current cycle. If this is the last problem set then the machine halts. The default guard is “problem set is empty”.

Puzzle Solving Example

Short Computation Cycle Short computation cycle is: cycle1: P0 -enum< P1 >elim- P2

Another Puzzle Solving Example

Long Computation Cycle

Long Computation Cycle Long Computation Cycle is: cycle2: P0 -enum< P3 -enum< P4 -enum< P5 >elim- P6 Result is: P6= {<123,804,765>}

An Exhaustive Search Machine cycle1: [guard1,1] P0 -enum< P1 >elim- P2 A slow intelligence exhaustic search abstract machine can always be constructed to find a solution, but it may not terminate.

More Slow Intelligence Operators Concentrator: P1 >conc= P2 is the concentrator that selects some problem elements of P1, which are put into P2. Adaptor: P1 +adap= P2 is the adaptor that inputs information from the environment and modifies elements of P1 to produce P2 according to the adaptation rule. Propagator: P1 =prop+ P2 is the propagator that outputs information to the environment and modifies elements of P1 to produce P2 according to the propagation rule. In the simplest case P2 is identical to P1.

An Adapt-to-Environment Example A machine that constantly adapts to the environment can be defined as follows: cycle1: [guard1,1] P0 +adap= P1 -enum< P2 >elim- P3 An adapt-to-environment abstract machine can always be constructed. Other machines for divide-and-conquer, follow-the-leader can also be constructed.

Discussion A theoretical foundation is provided for the slow intelligence approach by introducing the abstract machine model with multiple computation cycles. From practical standpoint, the computation cycle is specified by the controller super-component .

Discussion (continued) Each time a super component is added, a controller is created to control the timings of this computation cycle. Thus the SIS systems supports multiple computation cycles controlled by multiple controllers. Other types of super components (SI patterns) can also be specified.

Discussion (continued) Fast Cycle﹕Super components are not involved in a computation cycle (intuitive﹑involuntary) Slow Cycle﹕Super components are involved in a computation cycle (deliberate﹑voluntary) Environment variables and system variables are used to classify﹕ conceptual domain (such as high blood pressure, etc.) spatial domain (such as home, public area, etc.) temporal domain (such as morning, after lunch, etc.) Cycle switching rules are defined based upon environment variables and system variables.

Outline Multi-Level Slow Intelligence System An Abstract Machine Model Compiler for Abstract Machine TDR System Architecture Tian Super-Component Web GUI Discussion

The TDR Compiler

Slow Intelligence Patterns Monitor Advertiser Uploader, Alerter, GUI Coordinator Controller

Input to TDR Compiler

Temperature Settings GUI

Temperature/Blood Pressure Alerter

Controller for Temperature/ Blood Pressure Super- Component

Fall Detection SuperComponent 1/2

Fall Detection SuperComponent 2/2

Hospital Finder SuperComponent 1/2

Hospital Finder SuperComponent 2/2

Outline Multi-Level Slow Intelligence System An Abstract Machine Model Compiler for Abstract Machine TDR System Architecture Tian Super-Component Web GUI Discussion

TDR System

Multi-Level Computation Cycles The Tian super-component has computation Cycle1: Cycle1 [guard1,2]: P10 +adapA1= P11 -enum< P12 >elim- P13 =propA2+ P14 Likewise, the Di super-component has computation Cycle2: Cycle2 [guard2,3]: P20 +adapA2= P21 -enum< P22 >elim- P23 =propA3+ P24 Finally, the Ren super-component has computation Cycle3: Cycle3 [guard3,1]: P30 +adapA3= P31 -enum< P32 >elim- P33 =propA1+ P34 Notice the three computation cycles together form a higher-level computation cycle

Outline Multi-Level Slow Intelligence System An Abstract Machine Model Compiler for Abstract Machine TDR System Architecture Tian Super-Component Web GUI Discussion

Parrot Plant Sensor

Tian Super-Component

Three Layers Tian Super-Component Top Layer: Tian coordinator Middle Layer: Tian1/Tian2 controllers Bottom Layer: Flower1/Flower2 monitors, GUI1/GUI2, Uploader1/Uploader2 advertisers

Outline Multi-Level Slow Intelligence System An Abstract Machine Model Compiler for Abstract Machine TDR System Architecture Tian Super-Component Web GUI Discussion

Results visualized on smart phone

Outline Multi-Level Slow Intelligence System An Abstract Machine Model Compiler for Abstract Machine TDR System Architecture Tian Super-Component Web GUI Discussion

Sentient Systems A Sentient System is a multiple-level slow intelligence system TDR system provides an experimental platform for exploring different sentient systems such as personal health care, emergency management and social networking, to name but a few.

Slow Intelligence Patterns Patterns for monitors, advertisers, controllers and coordinators Patterns for Tian, Di and Ren Slow cycles vs. fast cycles A computation theory for slow intelligence

Q&A