C O M P U T E R 0 0 1 8 - 9 1 6 2 / 1 5 / $ 3 1 . 0 0 © 2 0 1 5 I E E E J U LY 2 0 1 5 37

COVER FEATURE SELF-AWARE AND SELF-EXPRESSIVE SYSTEMS

The Benefitsof Self-Awarenessand Attention in Fogand Mist ComputingJürgo S. Preden and Kalle Tammemäe, Tallinn University of Technology

Axel Jantsch, TU Wien

Mairo Leier and Andri Riid, Tallinn University of Technology

Emine Calis, TU Wien

Self-awareness facilitates a proper assessment of cost-

constrained cyber-physical systems, allocating limited

resources where they are most needed. Together, situation

awareness and attention are key enablers for self-awareness

in efficient distributed sensing and computing networks.

Theoretical neurobiologist Bernard Baars observed that “like any other biological adaptation, consciousness is functional.”1 The same can be said about self-awareness

in computing—a system that is aware of its own state can become robust and dependable, even with radical environmental changes and drastically diminished capabilities. This idea has resulted in a proliferation of research on self-awareness and other self-* proper-ties such as organization, configuration, optimization, protection, and healing (see the “Self-Aware and Auto-nomic Systems” sidebar).

Self-awareness improves system behavior, reducing processing, communication, and energy requirements. However, it is not feasible to design and implement self-awareness in an ad hoc manner for every new sys-tem. Introducing self-awareness as a separate concept in the cyber-physical system (CPS) infrastructure—rather than as part of the application functionality—promises to simplify CPS development and operation. As CPSs are typically systems of systems, self-awareness must be coherent and consistent across the component systems.

We believe that self-awareness, situation awareness, and attention are key enablers of efficient fog and mist

38 C O M P U T E R W W W . C O M P U T E R . O R G / C O M P U T E R

SELF-AWARE AND SELF-EXPRESSIVE SYSTEMS

computing (see the “Fog and Mist Com-puting” sidebar). Here, we explore two central aspects of self-awareness— situation awareness and attention—and how they facilitate the assessment of human physiological data in a proto-type self-aware health-monitoring CPS.

SITUATION AWARENESS AND ATTENTIONSelf-awareness monitors overall sys-tem performance in a dynamically changing environment. It includes self-monitoring, situation awareness and attention because the system must understand both its own state and the environmental conditions. A system that tracks only its own state has a very limited view of its situation.

Situation awareness enables the continuous interpretation of data col-lected from the environment in the context of the CPS’s goals and objec-tives.2 A situation is defined by the values and interpretation of a set of sit-uation parameters.3 A situation param-eter can be monitored or computed independently and represents a prop-erty of the situation of interest. In our example, the information for generat-ing situation awareness is exchanged via proactive middleware, which is inde-pendent of the application functional-ity and can be considered part of the CPS platform.

Attention helps balance the com-peting tasks of collecting, process-ing, and responding to the data by

prioritizing scarce system resources for the CPS’s tasks and objectives. These priorities dynamically change depending on the situation and sys-tem state.

SITUATION-DEPENDENT INTERPRETATIONThe concept of situation awareness originated in psychology, but its con-cepts are applicable to embedded sys-tems. Just as humans process data from their senses to develop situation awareness, a CPS must be aware of its situation to perform optimally, as the “correct” behavior is dependent on the current situation. For instance, the meaning of a low fuel warning light in a vehicle is different when that vehicle

SELF-AWARE AND AUTONOMIC SYSTEMS

More than a century ago, the concepts of self-awareness and consciousness were

explored by pioneer psychologists William James1 and Sigmund Freud,2 among others. Since the 1960s, attempts to structure and categorize self-awareness into levels, degrees, and scope have proliferated, but this also has led to a fair amount of confusion due to lack of coherence with earlier work. To clear the fog, Alain Morin proposed a framework of nine neurocognitive models of self-awareness with an analysis of their respective differences and similarities.3 Morin’s framework distinguishes—from lower to higher levels—unconsciousness, consciousness of external stimuli and events, self-awareness of public and private self-aspects, and meta self-awareness. Interesting aspects and nuances are due to “perception of self in time and com-plexity of self-representations.”3

Beyond the categorization of self-awareness phenomena, numerous theories have been devel oped about self-awareness, consciousness, and attention in the human brain. Bernard Baars described the Global Workspace Theory (GWT) of conscious processing, which is supposed to occur in the extended reticular-thalamic activation

system.4 GWT’s salient feature is that only one of the many parallel, subconscious processing modules has access to information at any given time and can spread it globally, thus controlling activation of large parts of the brain. Essen-tially, consciousness serves as a global resource allocator. Many GWT-predicted phenomena could be confirmed in simulations and experiments, although more recent research downplays the importance of consciousness while attention and goals assume more prominent roles.5 Goals affect behavior by modulating attention: when people try to attain them, attention helps maintain a balance between the focus and flexibility of actions.

Biological examples have long inspired computer engineers. The 1998 DARPA Broad Agency Announcement on Self-Adaptive Soft-ware triggered a plethora of research on topics like self-adaptive, autonomic, and self-aware computing.6 IBM picked up the thread and devel-oped a powerful vision of autonomic computing, leading to an abundance of research papers and products.7 Since then, work on self-* topics has flourished in the context of both large software systems and constrained embedded systems, as evidenced by recent research on self-healing,8,9

J U LY 2 0 1 5 39

is in the middle of a desert than when it is close to a gas station. Although the sensor value is identical, the interpre-tation of the sensor reading and conse-quent actions can be very different.

Complex phenomena require using data from several sensors with diverse modalities to generate an adequate level of situation awareness. The sen-sors may be attached to distinct, phys-ically disjointed computing nodes, which presents the challenge of dis-tributing the computation among individual nodes.

One example of distributed sensing and processing involves monitoring the human body during everyday activities. Evaluating the body’s state requires measuring physiological parameters

(such as heart rate) and interpreting their meaning (for example, the person is sleeping or running). Thus, sensors of different modalities must be attached to different areas of the body, leading to a distributed sensor system. However, wiring the human body is impractical, necessitating a network of autonomous wireless sensors.

When both sensing and process-ing are distributed, a hierarchical abstraction of situational information becomes necessary. As Figure 1 shows, this hierarchy extends from sensing (internal properties and the physical world), to perceived context (low-level sit-uations), to comprehension (higher-level situations), to projection (estimating future situations). In most computing

nodes, the highest level of situational data abstraction is perceived context, while more capable computing nodes also perform comprehension. Projec-tion requires a thorough understand-ing of the application domain as well as complex prediction methods and may not be possible in many CPSs.

The concept of such a hierarchy was introduced by Mica Endsley in the con-text of human situation awareness.2 Jürgo S. Preden extended this concept to the CPS domain by allowing for sit-uational information exchange at the lower levels, which maps well to the computational architecture in fog and mist computing.3

Situation parameters3 enable the representation of relevant abstracted

on-chip self-monitoring,10 bio-inspired hardware design,11 situation identification techniques,12 pattern-based engineering approaches,13 and self-awareness.14 Frameworks and platforms for self-aware computing are also proliferating.15–17

References1. W. James, The Principles of Psychology: Volume One, Dover

Publications, 1950.

2. S. Freud, Dora: An Analysis of a Case of Hysteria (Collected

Papers of Sigmund Freud), Touchstone, 1997.

3. A. Morin, “Levels of Consciousness and Self-Awareness: A

Comparison and Integration of Various Neurocognitive Views,”

Consciousness and Cognition, vol. 15, no. 2, 2006, pp. 358–371.

4. B.J. Baars, A Cognitive Theory of Consciousness, Cam-

bridge Univ. Press, 1989.

5. A. Dijksterhuis and H. Aarts, “Goals, Attention, and (un)

Consciousness,” Ann. Rev. of Psychology, vol. 61, 2010,

pp. 467–490.

6. R. Laddaga, “Active Software,” Proc. 1st Int’l Workshop

Self-Adaptive Software (IWSAS 00), 2000, pp. 11–26.

7. J.O. Kephart and D.M. Chess, “The Vision of Autonomic

Computing,” Computer, vol. 36, no. 1, 2003, pp. 41–50.

8. D. Ghosh et al., “Self-Healing Systems—Survey and Synthe-

sis,” J. Decision Support Systems, vol. 42, no. 4, 2007, pp.

2164–2185.

9. H. Psaier and S. Dustdar, “A Survey on Self-Healing Sys-

tems: Approaches and Systems,” Computing, vol. 91, no. 1,

2011, pp. 43–73.

10. G. Kornaros and D. Pnevmatikatos, “A Survey and Taxonomy

of On-Chip Monitoring of Multicore Systems-on-Chip,” ACM

Trans. Design Automation Electronic Systems, vol. 18, no. 2,

2013, article 17.

11. P. Cong-Vinh, ed., Autonomic Networking-on-Chip: Bio-In-

spired Specification, Development, and Verification, CRC

Press, 2011.

12. J. Ye, S. Dobson, and S. McKeever, “Situation Identif i-

cation Techniques in Pervasive Computing: A Review,”

Pervasive and Mobile Computing, vol. 8, no. 1, 2012,

pp. 36–66.

13. T. Chen et al., The Handbook of Engineering Self-Aware and

Self-Expressive Systems, arXiv:1409.1793, 2015; http://arxiv

.org/abs/1409.1793.

14. P. Lewis et al., “A Survey of Self-Awareness and Its Appli-

cation in Computing Systems,” Proc. 5th IEEE Conf. Self-

Adaptive and Self-Organizing Systems Workshops (SASOW

11), 2011, pp. 102–107.

15. A. Jantsch and K. Tammemäe, “A Framework of Awareness

for Artificial Subjects,” Proc. 2014 Int’l Conf. Hardware/

Software Codesign and System Synthesis (CODES+ISSS 14),

2014, pp. 1–3.

16. H. Hoffmann et al., SEEC: A Framework for SElf-awarE Com-

puting, tech. report MIT-CSAIL-TR-2010-049, MIT Computer

Science and Artificial Intelligence Lab, 2010.

17. S. Sarma et al., CyberPhysical-System-on-Chip (CPSoC):

Sensor-Actuator Rich Self-Aware Computational Platform,

tech. report CECS TR-13-06, Center for Embedded Com-

puter Systems, Univ. of California, Irvine, 2013.

40 C O M P U T E R W W W . C O M P U T E R . O R G / C O M P U T E R

SELF-AWARE AND SELF-EXPRESSIVE SYSTEMS

sensory data. Parameter types are highly domain and application depen­dent. It is critical that a single ontol­ogy and common semantics are used for the situation parameters, and

that the parameter values are valid in the required temporal and spatial intervals. A specific data item might be valid and useful at a certain time and location, but could be invalid

and misleading a few seconds later or at another location. Thus, metadata must be associated with the parameter values computed by distinct nodes.

In the case of human health moni­toring, the minimum spatial validity criterion is that the situation para­meters should characterize the same person. The temporal validity criteria specify the temporal interval in which the parameter values are valid. For example, the human activity assess­ment must reflect the activity within the last 10 seconds to correctly evalu­ate a human’s heart rate.

Situation parameter values reflect phenomena of interest; by fusing them, we can compute values of higher­level parameters. The types and accuracy of situation parameters, how they are computed, and their validity criteria are highly application dependent. The system must be able to cope with inac­curacies as the values could be impre­cise due to inherent challenges in pre­cisely monitoring physical processes, limited fidelity of the sensing hard­ware, or software approximations. For example, image processing and acous­tic signal processing results are almost always approximations.

One methodology that allows for combining data with varying levels of certainty is fuzzy logic.4 By associ­ating situation parameter types with fuzzy sets and situation parameter values with the degree of member­ship to a given set, the set member­ships can be computed at any comput­ing node. Fuzzy rules can be used to derive higher­ level situation param­eter values from lower­level fuzzy set membership levels. The membership functions for some fuzzy sets can be quite complex—for example, a set representing the amount and qual­ity of sleep. Comprehensive solutions

FOG AND MIST COMPUTING

Self-awareness in computing nodes can be achieved using several architectures.

In cloud computing scenarios, sensor data is communicated to central servers for analysis. This is a powerful method, as global knowledge from all relevant sources is available for evaluation but results in inefficiencies due to high bandwidth consumption and long delays.

In fog computing,1 computation is performed at the edge of the network at the gateway devices, reducing bandwidth requirements, latency, and the need for communicating data to the servers. Due to its distributed and localized architecture, fog computing is a natural platform for a variety of critical Internet of Things applications such as connected vehicles, smart grids, smart cities, and wireless sensor and actuator networks.2 To that end, several programming models and application frameworks have been developed for fog computing.3,4 However, in a strict defini-tion of fog computing, the devices at the edges are not involved in computation but only in data acquisition, while the interpretation occurs in the gateway. Thus, network delay and inefficient band-width utilization are still present.

Mist computing pushes processing even further to the network edge, involving the sensor and actuator devices. This decreases latency and increases subsystems’ autonomy. In such scenarios, self-awareness of every device is critical as the computation and actuation are dependent on the device’s perception of the situa-tion. The challenge with implementing mist computing systems lies in the complexity and interactions of the resulting network, which must be managed by the devices themselves as central management of such systems is not feasible.

References1. Cisco Technology Radar, “Fog Computing,” Mar. 2015; https://techradar.cisco

.com/trends/Fog-Computing.

2. F. Bonomi et al., “Fog Computing and Its Role in the Internet of Things,” Proc.

1st MCC Workshop Mobile Cloud Computing (MCC 12), 2012, pp. 13–16.

3. I. Satoh, “A Framework for Data Processing at the Edges of Networks,” Database

and Expert Systems Applications, H. Decker et al., eds., Springer, 2013, pp. 304–318.

4. K. Hong et al., “Mobile Fog: A Programming Model for Large-Scale Applications

on the Internet of Things,” Proc. 2nd ACM SIGCOMM Workshop Mobile Cloud

Computing (MCC 13), 2013, pp. 15–20.

J U LY 2 0 1 5 41

exist for automatically evaluating sleep. For instance, the Beddit Sleep Monitor (www.beddit.com) takes into account sleeping time, breathing pat-terns, heart rate, movement, and sev-eral other factors. Thus, the function for estimating the quality of sleep is complex. Membership functions for other fuzzy sets can also be simple, such as providing a value of the situ-ation parameter reflecting a person’s heart rate.

A SELF-AWARE HEALTH MONITORFigure 2 shows the conceptual archi-tecture of a self-aware health moni-tor that makes use of the middleware ProWare and was developed at the Research Laboratory for Proactive Technologies at Tallinn University of Technology. The prototype monitor abstracts and classifies inputs from various sensors including altitude, location, heart rate, accelerometers, temperature, and oxymeter and then compares the identified input pat-tern class with a prebuilt or dynam-ically updated model. In the case of a mismatch, an anomaly signal is generated that induces attention. An attention control mechanism trig-gers the collection of complementary data or additional analysis steps if an anomaly appears and the analysis is not conclusive. Depending on the intensity and duration of the anoma-lous situation, the monitor alerts the person or changes the health goal, thereby adapting to a new situation. In highly anomalous cases, it alerts other higher- level devices, emulating an emergency call.

In previous work, we showed how situation parameters for assessing the physical world’s state can be used in the context of an Intelligence,

Surveillance, and Reconnaissance ap-plication (ISR).3 Abstracting data in the sensor nodes reduces band-width and helps the system cope with changes to its structure.

We apply the same mist comput-ing principles to monitor vital human body parameters in the context of the individual’s activity. In addition to the interpretation of sensor data dif-fering in various situations, the fidel-ity of individual sensor data acquisi-tion and processing is dependent on the person’s activity—for example,

monitoring requirements for sleep-ing are different from those for run-ning. Monitoring requirements can be guided by a health practitioner, who could instruct the system in cer-tain situations to increase the fidelity of monitoring or to log the data with finer granularity.

The situation parameter reflecting the heart rate is only meaningful in context. To evaluate whether a person’s heart rate at a given moment is within a safe range, the algorithm must—at minimum—consider the specific

t°

O2

Feedback, alerts

OK

Introspection

Attention

Model

Select

Goals and behaviors:

• Standby• Classify• Emergency

Internal and external control signals

Learning

Slee

ping

Walking

Anomaly≠?

Data

abs

tract

ion

Categorization:Running, walking, stairs,

sleeping, and so on

FIGURE 2. Architecture of a prototype self-aware health monitor.

Projection: estimating future situations

Comprehension: higher-level situations

Perceived context: low-level situations

Internal properties

Sensing

PerceptionSystem

Other systems

Physical world

FIGURE 1. Hierarchical abstraction of situational information in cyber-physical systems with distributed sensing and processing, inspired by work on human situation awareness.

42 C O M P U T E R W W W . C O M P U T E R . O R G / C O M P U T E R

SELF-AWARE AND SELF-EXPRESSIVE SYSTEMS

current activity and the immediate history of activities, as it takes time for the human body to adapt to or recover from a specific activity. The larger con-text is also relevant: how well rested the person is, how much food has been consumed, and so on. In a health- monitoring application, parameter types can reflect a person’s activity (such as resting, walking, or running), the state of the body (stressed, rested, or tired), and the current physical load (high or medium). The parameter values can be computed centrally or through separate computing nodes, as depicted in Figure 3a.

Monitoring a person’s daily activ-ities has been an active research area for some time. Liang Dong and his col-leagues placed several accelerometers on an individual, using a Kalman filter to track and classify the individual’s daily physical routines.5 The body seg-ment status was categorized into static and dynamic, and further differenti-ated into periodical and nonperiodical status using discrete Fourier trans-form. A hidden Markov model was used for training data and periodical movement modeling. The researchers reported overall classification accu-racy at about 90 percent.

Davide Curone and his colleagues developed a similar physical activ-ity assessment system for emergency intervention rescuers.6 They inte-grated wearable electronics into tex-tile fabrics to automatically identify potentially dangerous conditions for the monitored subject. The system

achieved overall classification accu-racy of 88.8 percent.

As these examples illustrate, the evaluation of a person’s activity might use a range of sensors as input, and the sensor data must be interpreted differ-ently for different activities. A heart rate of 130 might be normal for climb-ing stairs or jogging, but the same heart rate is worrying when the person is at rest or working at a desk.

ProWare facilitates the effective exchange of situation parameters. Fig-ure 3b depicts the system elements that are involved in data exchange. Pro-Ware enables the dynamic establish-ment of communication partnerships (as service subscriptions) for exchang-ing situation parameters, enabling the consumer to specify the parameters’ temporal and spatial constraints.3

The ProWare components, located in every computing node, check the validity criteria and ensure that only data that satisfies the validity crite-ria are delivered to the analysis algo-rithms. The components are packaged with a clustered mesh protocol in a compact wireless module that enables fast integration with sensor devices and is pin compatible with Raspberry Pi, Arduino, and Bluehex.

EXPERIMENTAL EVALUATIONAlthough the complete architec-ture shown in Figure 2 has not yet been realized in our prototype sys-tem, we conducted a series of health- monitoring experiments to validate key assumptions and show the viability of

identifying self-awareness properties in a mist computing approach. In our experiments, a test subject engaged in the following activities: resting on a couch, sitting, driving a car, walking slowly indoors, climbing and descend-ing stairs indoors, and walking at a rapid pace outdoors. Relevant data for these activities was collected from three sensors—namely, an accelerometer, an altitude meter, and a heart rate monitor.

We analyzed the collected data to determine whether assessing the local situation by combining data from indi-vidual embedded sensor nodes is fea-sible. Although we analyzed the data offline, the applied algorithms were suf-ficiently lightweight to be executable in embedded low-power computing nodes.

The subject’s heart rate data was logged using a BM innovations GmbH chest strap BM-CS5 (http://bm -innovations.com). The pulse rate was communicated once per second using the BlueRobin wireless protocol to the Texas Instruments eZ430- Chronos watch (www.ti.com/tool/ez430-chronos). The watch was equipped with an inter-nal pressure sensor for altitude measure-ments. The heart rate and altitude were temporarily saved with a full timestamp in the watch’s internal memory. We used a smartphone accelerometer to evalu-ate the subject’s activity, and collected accelerometer measurements using the G-Sensor Logger Android application.

Data logs from the experiments were communicated to a PC for analy-sis via the wireless SimpliciTI protocol (http://processors.wiki.ti.com). We used

Provider

Provider

Consumer

Consumer

Subscription

Subscription

Data

Data

Health assessment

Health assessmentTypical physiological

parameter values

Current activity

(b)

ProWare

Node 1SensingFusion

ProWareSensingFusion

Node 3

Node 2

ProWare SensingFusion

Activity assessment

Accelerometer iBeacon interface(a)

GPS Heart rate

Pulse oxymeter

....

Current parameter values

Physiologicalparameter assessment

FIGURE 3. Hierarchy of situation parameters in a health monitoring scenario and exchange of parameter values facilitated by ProWare. (a) Hierarchies of situation parameters computed by individual computing nodes. (b) ProWare components in individual computing nodes.

J U LY 2 0 1 5 43

Matlab to analyze the data logged from all sensing devices during the tests. Our goal was to investigate whether the performed activities can be detected (and whether the situation parameter values can be determined) from indi-vidual data streams. The sampling rate for pulse rate and altitude estimate was 1 Hz, while the average sampling rate for the accelerometer was 16 Hz, mean-ing that the raw sensor data was syn-chronized before analysis. In a mist computing implementation, ProWare would perform the synchronization.

We employed the modulus of the acceleration vector

a a ax y2 2= +

when data from two sensors was used, and

a a a ax y z2 2 2= + +

for data from three sensors to esti-mate the subject’s activity level, which proved to be sufficient.

Figure 4a depicts the average values of observed pulse rates and accelera-tion over 10-second periods.

The results from the different experiments populate distinct areas in the pulse rate/activity space. The activities of sitting, driving, indoor walking, and outdoor walking can be well categorized using only two sensors: the pulse rate meter and the accelerometer. However, stair climb-ing forms a rather large area in the pulse rate/accelerometer space, trig-gering the attention mechanism to seek further data from the altitude sensor. The additional data allows the health monitor to identify the activity as stair climbing and to distinguish between moving upward (green) and downward (blue).

More generally, the results illus-trate the benefit of attention-directed

data collection and analysis. If data from a few sensors, processed with a simple analysis algorithm, leads to an unambiguous conclusion, unnec-essary data collection and processing are avoided. Only cases when anom-alies are detected or the analysis is inconclusive warrant more elaborate and expensive data collection, com-munication, and computing. Thus,

attention- based sensing and analysis have the potential to save significant time and energy in CPS applications. Future work should quantify this potential in various applications.

Note that the states in Figure 4 can be considered the steady states for given activities, although some tran-sitional states are also observed (the samples between the boxes). Pulse rate

Acce

lera

tion

(m/s

2 )

70

0.5

1

1.5

2

2.5

3

3.5

4

4.5

80 90 100 110 120 130 140 150

Puls

e ra

te (b

pm)

Altit

ude

(m)

Acce

lera

tion

(m/s

2 )

50

30

40

50

0

5

10

Sitting Stairs Sitting SittingRunning

50100

100

150

150

200

200

250 350300 400 450 500 550

50 100 150 200 250 350300 400 450 500 550

50 100 150 200 250 350300 400 450 500 550

(a)

(b)

Pulse rate (bpm)

Time (s)

FIGURE 4. Experimental evaluation results. (a) Pulse rate versus activity, sitting or resting (red), driving a car (black), indoor slow walking (dark blue), outdoor rapid walking (magenta), walking upstairs (green), and walking downstairs (light blue). (b) The modes are correctly tracked as the subject performs different activities.

r7tam.indd 43 6/25/15 4:25 PM

44 C O M P U T E R W W W . C O M P U T E R . O R G / C O M P U T E R

SELF-AWARE AND SELF-EXPRESSIVE SYSTEMS

changes are never instantaneous, as shown in Figure 4b, which depicts a series of exercises and illustrates why the temporal aspect of human phys-iology must be considered as it also depicts the transitional phases from one activity to another.

Our experiments show that rel-atively simple sensors can be used to determine a person’s activity and

correlate physiological parameters to individual activities. Most of the time, only a subset of the available sensors must be employed, making it a lean monitoring approach. For a more accu-rate estimation, however, additional phenomena must be measured. Natu-rally, such a monitoring system must adapt to an individual; once the adap-tion phase is complete, the CPS can

monitor the person and determine whether the physiological parameters are within the typical range for a given activity.

Self-awareness has the poten-tial to achieve high efficiency in various sensor and actuator

networks. Inspired by biological exam-ples, studies and proposals that touch on various aspects of computational self-awareness have proliferated in recent years. Still, its potential is hardly understood and not yet exploited. Con-tributing to this effort, we explored situation awareness and attention in a prototype health monitoring CPS and found that both are critical to self-awareness. A system must under-stand its inner state and what its envi-ronment looks like to make a proper assessment of its own state and perfor-mance; a system can be self-aware only if it understands its context.

In ProWare, situation awareness is jointly achieved by a group of sensors that distribute the burden over sev-eral nodes, leading to mist computing. Fairly simple algorithms, executed by resource-constrained embedded nodes, compute the situation parame-ters. ProWare facilitates the informa-tion exchange of situation parameters within a resource-constrained network: using a standalone RF hardware mod-ule that contains ProWare, it is easy to connect any device to an existing CPS.

We used three sensors (for pulse rate, acceleration, and altitude) in our exper-iments to show how the different mea-surements complement one another to allow for a precise assessment of typ-ical activities, and how attention can streamline data collection and process-ing to make the system more lean and efficient. Multiple sensors distribute

ABOUT THE AUTHORS

JÜRGO S. PREDEN is a senior researcher and head of the Research Labora-

tory for Proactive Technologies at Tallinn University of Technology, Estonia, and

CEO of Thinnect (www.thinnect.com). His research interests include distributed

computing systems, specifically the situation awareness of such systems. Pre-

den received a PhD in systems engineering from Tallinn University of Technol-

ogy. He is a member of IEEE. Contact him at jurgo.preden@ttu.ee.

KALLE TAMMEMÄE is a professor in the Department of Computer Engineering

at Tallinn University of Technology. His research interests include hardware−

software codesign and system-on-chip (SoC) design as well as brain-inspired

self-aware systems. Tammemäe received a PhD in systems engineering and

informatics from Tallinn University of Technology. He is a senior member of

IEEE, the IEEE Estonia Section chair, and a member of ACM. Contact him at

kalle.tammemae@ttu.ee.

AXEL JANTSCH is a professor in the Institute of Computer Technology at TU

Wien, Austria. His research interests include dependability, robustness, and

self-awareness in SoCs and embedded systems. Jantsch received a PhD in

computer science from Vienna University of Technology. He is a member of

IEEE. Contact him at axel.jantsch@tuwien.ac.at.

MAIRO LEIER is a PhD student and researcher in the Department of Computer

Engineering at Tallinn University of Technology. His research interests include

optical pulse wave analysis methods for increased signal quality, detection of

sleep diseases in children, and body area networks. Leier received an MSc in

informatics from Tallinn University of Technology. He is a member of IEEE. Con-

tact him at mairo.leier@ttu.ee.

ANDRI RIID is a senior research scientist at the Research Laboratory for Pro-

active Technologies at Tallinn University of Technology. His research interests

include computational intelligence, fuzzy systems, classification, and control

and modeling algorithms. Riid received a PhD in systems engineering from Tal-

linn University of Technology. Contact him at andri.riid@ttu.ee.

EMINE CALIS is a PhD student in the Institute of Computer Technology at TU

Wien. Her research interests include self-awareness in SoCs. Calis received an

MSc in power engineering from Vienna University of Technology. She is a mem-

ber of IEEE. Contact her at calis@ict.tuwien.ac.at.

J U LY 2 0 1 5 45

data collection and processing based on fog and mist computing paradigms.

We are currently extending our CPS prototype to monitor more of the human body’s parameters and to dynamically learn normal sensor patterns, facilitating the detection of deviations and anomalies.

REFERENCES1. B.J. Baars, A Cognitive Theory of

Consciousness, Cambridge Univ. Press, 1989.

2. M.R. Endsley, “Design and Eval-uation for Situation Awareness Enhancement,” Proc. Human Factors and Ergonomics Society Ann. Meeting, vol. 32, no. 2, 1988, pp. 97–101.

3. J. Preden et al., “On-line Data Vali-dation in Distributed Data Fusion,” Proc. SPIE 8742, Ground/Air Multisen-sor Interoperability, Integration, and Networking for Persistent ISR IV, 2013; doi: 10.1117/12.2016249.

4. A. Riid and E. Rustern, “An Inte-grated Approach for the Identifica-tion of Compact, Interpretable and Accurate Fuzzy Rule-based Classifi-ers from Data,” Proc. 15th IEEE Int’l Conf. Intelligent Eng. Systems (INES 11), 2011, pp. 101–107.

5. L. Dong, J. Wu, and X. Chen, “Real-Time Physical Activity Monitoring by Data Fusion in Body Sensor Net-works,” Proc. 10th Int’l Conf. Informa-tion Fusion, 2007, pp. 1–7.

6. D. Curone et al., “Heart Rate and Accelerometer Data Fusion for Activ-ity Assessment of Rescuers During Emergency Interventions,” IEEE Trans. Information Technology in Biomedicine, vol. 14, no. 3, 2010, pp. 702–710.

Selected CS articles and columns are also available for free at http://ComputingNow .computer.org.

Call

Publish Your Paper

in IEEE Intelligent Systems

IEEE Intelligent Systems seeks papers on all aspects

of artifi cial intelligence,

focusing on the development

of the latest research into

practical, fi elded applications.

For guidelines, see

www.computer.org/mc/

intelligent/author.htm.

The #1 AI Magazine www.computer.org/intelligent

Be on the Cutting Edge of Artifi cial Intelligence!

IEEE

Articlesfor