1. Emergency Communications, Floating Content
(and IoT)
Gianluca Rizzo
University of Applied Sciences of Western Switzerland – HES-SO Valais
Universita’ di Foggia – Italy
gianluca.rizzo@hevs.ch
gianluca.rizzo@unifg.it
May 21st, 2021
Summer school UNI-INTERNATIONAL SCHOOL ON IOT – Universita’ di Salerno
2. Goals
• Be able to recognize the various dimension of a
(natural) disaster, and their implications for the
provisioning of content, computing, and
communication
• Be able to recognize the specificities of demand for
content, computing and communications in the
aftermath of a disaster
2
3. What is a disaster?
1. A serious problem
2. Causing human, material,
economic or environmental
loss
3. Exceeding the ability of the
affected community to cope
using its own resources.
Q1: are all of these three elements
necessary?
Ruins from the 1906 San Francisco earthquake(Wikipedia)
Source: WHO, IFRC
4. Disasters take a heavy toll on world
economy and people’s lives
Source: UNISDR – Human and Economic Impact of Disaster in last 10 years
5. Disasters Prone regions are common
in every continent
Source: UNISDR – 2015 Disasters in numbers *Epidemics and Infections not included
6. Number of disasters by type (2012-2018)
The vast majority of natural disasters are tightly coupled to
local environmental and geological features
7. A Disaster Manangement strategy is the
combination of risk management and
crisis management
7
8. What is the role of ICT in Disaster
management?
• Ensuring an adequate diffusion of key information is
vital for both crisis management and risk management
• Both risk and crisis management have a near-term
(immediate aftermath) and medium-long term
component
• The most challenging (and largely open) issue is how
ICT can cater for crisis management in the immediate
aftermath of a disaster
• Key for saving lives and mitigating overall impact of disaster
8
9. Earthquakes
• Area covered: a city or a small region
• Impact on population: migration (partially
definitive), disruption of almost all human
activities
• Difficult to predict (sometimes, a few
seconds ahead)
• However, seismic risk level is well known and
mapped in detail.
• Importance of early support (first
hours/days)
• Entrapped people: Life expectancy drops to zero
after 3 days
• Hazard reduction: gas leaks, floods, fires, electricity
10. A forecasted disaster:
Hurricane Dorian
• The most intense tropical cyclone on
record to strike the Bahamas
• worst natural disaster in the country's
history
• Peak on Abaco Island on Sept 1st 2019, over
by Sept 3rd.
• Damages:
• badly damaged roofing
• power lines cut off
• roads impassable
• fatalities: 63 direct, 7 indirect, 280 dispersed
• damage ≥ 8.28 billion USD
• Local community required massive
interventions from non-local actors
• Huge deployment of external rescue teams,
humanitarian organizations
11. If entirely predictable, how was
Dorian a disaster?
• Forecasted well in advance, in a
region usually struck by hurricanes
• However:
• big differences in amount of damage
between the area directly covered
and neighboring areas
• hard to predict exact path more than
1-2 days in advance
• serious damage is relatively rare:
• Last hurricane which made
comparable damages in the same
areas struck 50 years before
• Uncertainty made perceived risk
too low to take drastic preventive
measures
12. Key
properties
of a
disaster
Disruption is limited in time (and space)
• Short or long period
• Permanent disaster is not a disaster
Directly related to the choice of:
• A model of the environment, of
operational conditions
• A risk management strategy (Implicit or
Explicit)
• A tradeoff between amount of tolerated
risk and costs of
• Ordinary operations
• Disruptions due to disasters
12
13. Elements of a
disaster
management
strategy
Goal: Minimize consequences of
disasters, given:
a. A model of the environment
b. Budget constraints
c. Cost of (ordinary) operations
d. Cost of disaster management
• before and after the disaster
e. Available technologies
13
14. Elements of a
disaster
management
strategy
Goal: Minimize consequences of
disasters, given:
a. A model of the environment
b. Budget constraints
c. Cost of (ordinary) operations
d. Cost of disaster management
• before and after the disaster
e. Available technologies
14
15. How can an inadequate model of the
environment affect system fragility?
15
Two factors (among many):
Cognitive biases
Networking effects
Things are getting worse as technology
progresses…
17. Rare
events
matter!
• Extrapolating from data: The Turkey
• A nonzero probability of ruin
• The era of AI: Are we all getting turkeys?
• The more we know, the more we get
confident on our risk model
• Gaussian vs fat tail domains: «Rare» events
17
18. Complex systems,
interconnectedness
and fragility
Interconnectedness among its parts makes
a system «complex» and more fragile
• Directly: e.g. power distribution network
and telecommunication network
• Other examples: Financial
markets…
• Indirectly: Harder to model
• Interrelationships make collapses and
cascading failures more dangerous
18 A graph visualisation of the topology of network connections of the core of the Internet, December 11 2000. (Source: Bill Cheswick, http:// www.lumeta.com)
19. Elements of a
disaster
management
strategy
Goal: Minimize consequences of
disasters, given:
a. A model of the environment
b. Budget constraints
c. Cost of (ordinary) operations
d. Cost of disaster management
• before and after the disaster
e. Available technologies
19
20. In many disasters, disruptions in network
operations are a consequence of network
design choices
• Before: Reassure, minimize.
• Use media to minimize impact of disaster (and of lack of investments)
on their image
• After: be unresponsive, hide issues from the news, disclose
nothing.
• Hurricane Dorian post disaster communication disruptions (relatively hard to find):
• By September 11, cellular services had resumed in most affected areas except for eastern Grand Bahama and northern parts of Abaco,
including Cooper’s Town and Treasure Cay. Aliv, one of the country’s telecommunications providers, continues to conduct repairs and
estimates that up to 80 percent of the network in Abaco will be restored within one week; re-establishing cellular service in eastern
Grand Bahama is anticipated to take longer. As telecommunications services improve, Télécoms Sans Frontières (TSF) noted plans to
reduce the quantity of free humanitarian calling operations for isolated communities beginning September 11; TSF continues to provide
satellite connection to facilitate communication for humanitarian operations in Abaco and Nassau.
• Electricity infrastructure in Abaco, particularly in Marsh Harbour and across the island’s northern areas, remains extensively damaged,
leaving affected communities reliant on generators for power and increasing the demand for fuel, NEMA reports. Equipment and
personnel from the Bahamas Power and Light Company (BPLC) arrived in Abaco on September 11 and 12 to begin repairs, which will
initially focus on restoring power in southern Abaco where damage is less severe. NEMA is coordinating with relief partners to facilitate
the acquisition and transport of additional equipment—including high-capacity generators—to supplement BPLC efforts.
22. What is a
reasonable
goal for
research on
post-disaster
technologies?
Goal: Minimize consequences of
disasters, given:
a. A model of the environment
b. Budget constraints
c. Cost of (ordinary) operations
d. Cost of disaster management
• before and after the disaster
e. Available technologies
22
23. What is a
reasonable
goal for
research in
post-disaster
technologies?
• Lowering the cost (and the amount)
of pre/post disaster interventions
• Increasing their effectiveness
Information diffusion is a key enabler
23
24. Why are disasters a
problem for a
telecommunication
networks?
Disaster: A set of events which
• disrupts the services provided by a communication
network
• changes their operating conditions, and
• not taken into account in network design phase.
• in a way which the network is not able to satisfy
demand
• in part, at least
Fallen telephone poles resulting from Hurricane Katrina. Photo by USACE.
25. Consequences of a disaster on
communications
• Some services /QoS levels cannot be guaranteed
• Service demand changes
• Disaster changes (sometimes drastically) the
composition and distribution of the population
• It introduces new communication needs, which have to
be satisfied (rapidly)
• It introduces new needs for data/content, and for
computing resources
• The outcome is the emergence of new demand
patterns, which cannot be satisfied adequately by the
network
26. An agenda for post-disaster
communications research
Before a disaster strikes: Disaster preparedness
• Make technical solutions for network resilience more convenient
• Prepare for post-disaster service delivery
After a disaster: Mitigate the consequences on communications
• Providing (alternative) strategies for information diffusion which
are
• Efficient
• Effective
• Timely
• Fit for the service demand in a post disaster scenario
• Challenges: hostile environment, lack of infrastructure (power),
difficulty in movements, rapidly evolving conditions and needs,
heterogeneity of conditions and needs
27. Impact on population distribution
• Displacements of a large
fracton of the population,
and for several tens of km
• People start moving back
to Abaco only two weeks
after
• Impact on communication
demand distribution in
space and on capacity
provisioning
28. Communications in the aftermath of
Dorian
• On the Abaco Island, cellular communications and landlines have been completely
down for a week, and intermittently working for several days more.
• Communication services were missing
when and where they were needed the most
9/2 – Andrew Schroeder, Director of Research and Analysis at Direct Relief:
“The main takeaway in the Bahamas is that we can’t say much about dynamics in the affected area due to network outage. It’s wiped out in
the affected area.”
29. The main impact of earthquakes on communication
infrastructure is through disruption of power supply
• Disruptions of telecom infrastructure:
• Destruction of antenna masts/BS equipments (seldom), disruptions of cables
(seldom)
• Power failures leading to power outages (very frequent, often as a precaution)
• Scope: ~10 square km and up
• Typically, cellular and fixed communications remain available until telecom
equipment runs out of power (a few hours).
• Restoration activities
• Antenna masts and BSs: deployment of truck mounted devices for temporary
provisioning (a few days after).
• Power distribution network: gradual reactivation and repairing of damages
• Duration: from a few days after the earthquake, to several weeks after.
• Earthquakes tend to change permanently the population distribution in the
area, their requirements and traffic demands.
• Ex. L’Aquila (Italy): “New towns”, fully new districts built in safer areas
30. Rate of population change over different sampling periods in Fukushima, Miyagi, and
Iwate prefectures in Japan (A-D) where the size of circles denotes the number of
observed evacuations (larger circles indicate more observed evacuations) and the color
of circles indicates the rate of population change in a specific area (ranging from -1 to 1
(-100% to 100%)).
Long- term changes in population distribution
modify traffic demand patterns
Tōhoku earthquake and tsunami (March 11, 2011) – cell network call records
Song, X., Zhang, Q., Sekimoto, Y., Horanont, T., Ueyama, S., & Shibasaki, R. (2013, August). Modeling and probabilistic reasoning of population evacuation during large-
scale disaster. In Proceedings of the 19th ACM SIGKDD international conference on Knowledge discovery and data mining (pp. 1231-1239). ACM.
31. • Characterizing how they change in response to an emergency is
key to post-disaster communications design
• Immediate aftermath (first hours/first days) is critical for e.g.
people in entrapment after earthquake
• Barabasi et al.: characterized real time changes in mobility and
communication patterns in the vicinity of eight different
emergencies
• N.W.: only some of them are disasters
• Compared to eight planned, non-emergency events
31
Communication patterns after a disaster may
radically differ from normal conditions
Bagrow, J. P., Wang, D., & Barabasi, A. L. (2011). Collective response of human populations to large-scale emergencies. PloS one, 6(3), e17680.
32. 32
Jet scare refers to a sonic boom interpreted by the local population and initial media reports as
an explosion
34. Patterns of information diffusion in space depend on
the spatial coverage of the emergency
G0: users directly affected by the event
G1: users that receive calls from G0 but are not near the event
G2: users contacted by G1 but not in G1 or G0, etc.
• Great heterogeneity in peak, location and duration
• Some disasters cause a communication cascade, with far-reaching effects
(socially and geographically)
Bagrow, J. P., Wang, D., & Barabasi, A. L. (2011). Collective response of human populations to large-scale emergencies. PloS one, 6(3), e17680.
Changes in call volume over time since event
35. Onset speed of anomalous call activity
• lower fmid indicates a faster onset
Spatial extent of the events
• non-emergency events are far more centrally
localized
Likelihood of calling an acquaintance on
the onset of a disaster.
Patterns of communications during disasters differ
drastically according to the type of disaster
Bagrow, J. P., Wang, D., & Barabasi, A. L. (2011). Collective response of human populations to large-scale emergencies. PloS one, 6(3), e17680.
36. What can we learn?
36
Technology and preparedness will not free us from disasters
• We need post disaster communication strategies to mitigate consequences
of a disaster
Information exchange after a disaster:
• Increases very quickly and drastically
• Creates waves and cascading effects, even well beyond the
perimeters of the disaster area
• Evolves over time in a disaster-specific manner, and
radically differently than ordinary conditions
Bottom line: Need for disaster-specific
• preparedness strategies, and
• post-disaster communication technologies
37. Voice communications for coordination of first
responders (and not only) are the first service to be
restored
Amatrice earthquake: providing wifi, voice services, and
phone charging station
Dorian Hurricane: satellite phones for voice communications:
first responders and local population
Enable efficient coordination:
• between rescue team members
• between on-field teams and other first-
responder team members
• One of the first services currently
provided after a disaster
• Point-to-point or (small) group
communications
• Slow to be deployed
• Hard to collect and spread information
on local status beyond first responders
38. The smart-everything paradigm is offering
new opportunities for disaster response
and mitigation
• Disaster scenarios are getting information rich, too
• Smart city, smart buildings, IoT paradigms, 5G,…
• Power distribution is still centralized and vulnerable
• Prosumer trend and microgrid, smart grid paradigm are
changing that too
• New disaster response approaches require massive
data, computing and communication resources
• E.g.: Telehealth, post-disaster assessment, etc
39. Situation Awareness in the age of
social media
• When based on organized rescue groups and teams:
• It is slow and partial
• Mainly based on voice communications
• In the last years, social media have been integrated into SA solutions
• As a source of data, and
• As a medium for information diffusion
• Problem: create a live picture of the status of the crisis zone, of
resources and of the operations
• Key for disaster response coordination
• Its effectiveness has a strong impact on quality of life of affected population
40. Impact of social media on Situation Awareness:
The Hurricane Irma case
• Social media (Facebook, Twitter)
Initially: discouraged
• Concerns of robbery, misinformation, and
widespread panic
• Rely on 911
BUT: Official channels were quickly
overwhelmed
• People were routed from 911 to social
media channels (Zello, etc)
• Facebook, Twitter used to monitor people
movements and identify critical issues
Source: https://psmag.com/social-justice/what-harvey-and-irma-taught-about-using-social-media-in-emergency-response
41. How to deliver situation awareness when
the cellular network fails?
Opportunistic communications offer an ideal instrument to
overcame failure of infrastructure-based communications
Exploit pervasive resources and sources of data: UE, IoT devices
Partial failures: Offload cellular network
Coverage holes: Coverage extension through ad hoc networking.
Restoring p2p communications might not be always feasible
Not in the immediate aftermath of a disaster, at least
They take a large toll on network resources
We focus on non-real time diffusion of information of interest to a
(sub)set of the on site actors
Priority to information of common interest
Key idea: Empowering both first responders and people on site to take
informed decisions
Collaborative approach, combining infrastructure, IoT devices, and UE
42. 1) A node generates content
Floating Content: A scheme for collaborative, probabilistic
information storage based on opportunistic communications
43. 2) Content is replicated opportunistically
Floating Content: A scheme for collaborative, probabilistic
information storage based on opportunistic communications
44. III: Content starts to float
3) Content “floats” within the Anchor Zone (AZ)
Nodes delete content
upon going outside AZ
Floating Content: A scheme for collaborative, probabilistic
information storage based on opportunistic communications
45. III: Content starts to float
Zone of Interest (ZOI)
Success Ratio: Fraction of users entering the ZOI with the content
Floating Content: A scheme for collaborative, probabilistic
information storage based on opportunistic communications
46. Floating Content implements a probabilistic,
infrastructure-less spatial information storage
• Typical applications:
• Proactive caching at the user
• Events notifications (e.g. road accidents)
• Proximity marketing
• Social networking (chats, situated introductions)
• Infrastructure may take a coordinating role
• Orchestrates content diffusion and delivery
• Initiates it
• Monitors and optimizes resources utilization (both MNO’s and
user’s)
• Takes over when distributed implementation fails (and/or when
MNO resources are available)
• Nowadays, (partially) available even in disaster locations
47. Situation awareness with FC: Creating a shared
vision of a disaster area and of rescue operations
• Goal: Build collaboratively a common view,
accessible by everyone in the region
• Without infrastructure
• Many-to-many: no bottleneck
• Information sources: Smartphones, WSNs, and
any “smart thing”
50. 50
Design of appropriate incentive schemes
for coopeation
• Common interest in contents (due to
context)
• Thus, willing to help its diffusion
• Storage: offer spatial redundancy
• Delivery: (selective) replication
• How to (optimally) manage redundancy?
• What about churn?
• Main limiting factor: Power supply
Open issues
51. 51
Ensuring only trustable information is
floated
• Solution taken in app:
• Confirmation
• «Rating» of contributors
Open issues
52. Challenges in implementing FC-
based content sharing
• FC requires a critical mass in order to work
• Typically, people/cars accumulate in the vicinity of an event
• Smart objects can be made part of the replication scheme, if allowed by
TX range
• Drones could be used to collect contributions from static IoT devices
• Seeding: A FC based app must be preloaded in some
nodes in the area, and spread opportunistically when
needed
• Current implementation: Smartphones/IoT devices as wifi APs + captive
portal
• How to configure IoT devices/gateways to join a FC scheme?
• Engineering a FC application may be a challenging task
• Balance QoS and resource efficiency
• Manage incentives to cooperation
52
53. FC Integration with IoT
• Floating content finds a natural implementation in WSN
• Absence of a sink due to disaster: Information may be
aggregated and floated
• Duty cycling/intermittent failures
• Moving (IoT) nodes can relay aggregate information between
WSN and surviving access points
• Floating Content can boost proximity-based IoT
• Extend the interaction range of smart objects AND of users –
enriching situation awareness
• How to discover IoT resources when infrastructure fails? Need
for (some) infrastructure support
• How to enable P2P exchanges between devices and UE when
they are not designed for it?
54. Performance challenges tackled
• Derivation of application-level performance guarantees
• In general, a function of mobility patterns and specific app
• Appropriate mix of:
• Non-spatial modeling approach: Mean Field
• Spatial models (geometry and mobility dependent)
• Extension of the above to realistic settings
• Vehicular
• Office and campus
• Derived simple analytical models providing robust estimates
of success ratio
• Key step: Use of a (simple) abstraction for some key features
of mobility
54
55. Performance challenges tackled
• Derivation of resource optimal configuration in heterogeneous
settings
• user density, speed, communication technology, storage, resource
costs, (among others)
• a discrete version of reaction-diffusion models, coupled with Mean
Field
• valid for stable settings and long lasting contents
• Extension to dynamic settings and short-lived contents
• No steady state solutions, no Mean Field
• Based on infrastructure support in:
• Collection of data about user mobility and request patterns
• Coordination of content replication and storage
• Content seeding
• Determination of the storage capacity of a probabilistic
caching system implemented via FC
56. • Each node plays two roles
(client and server)
• A single global
model or N
personalized models
• Merge-Update-Send cycle
Bringing computing and AI in post disaster scenarios:
Gossip Learning in dynamic settings
GL: decentralized learning without a central
server
Client3
Client2
Client1
Server
Federated Learning
Problem: enable users/devices to make sense of data in a disaster scenario
57. Conclusions
• FC for Situation Awareness is an active area of research
• Many applications under different names
• Infrastructure ubiquity in 5G/B5G and context-awareness
applications and services are key drivers
• Our work provides some tools for bringing FC closer to
adoption in disaster scenarios
• Application requirements, and QoS constraints
• Deal with (and take advantage of) node heterogeneity and
dynamicity
• Several issues are still open
• Support for distributed learning architectures
• Integration of IoT
• Adaptation to resource (energy) constrained devices
57
3 elements.
-A Disruption of ordinary operativity of a system, of community life, of a society
-Damages
-Requiring more resources than those available
Are all of these three éléments required?
1+2 without 3: still a form of ordinary administration, if already planned for it.
Without 2: no problem, trivial
Without 1: permamnet disaster is not a disaster: disasetrs are limited in time, for a long or short period.
UNISDR : UN Office for Disaster Risk Reduction
UNISDR : UN Office for Disaster Risk Reduction
So it is a problem which involve everyone, rich and poor regions alike
Disasters are bound to increase with climate changes!!
True especially for those disasters potentially affecting ICT infrastructure
Too conservative means too expensive in the local perception of risk
Someone must have used these éléments to allocate resources, and thus directly impact the onset of disasters
Condition a is often implicit
Condition a is often implicit
Fallacies in estimating risk
Have you heard of the Procrustean bed? Procrustes, a character in Greek mythology, kept a house by the side of the road where he offered hospitality to passing strangers, who were invited in for a pleasant meal and a night’s rest in his very special bed. He described it as having the unique property that its length exactly matched whosoever lay down upon it. If they were too long he would cut off their legs in order to fit the bed. If they were too short he would place them on a rack and stretch them until they would fit the dimensions of his bed.
Does the future always look like the past?
Bottom line: our cognitive biases, and intrinsic complexity due to network effects (among other things) are some of the factors «facilitating» disasters
Teher is alarge amount of network reslience technologies available
Technology is not a problem
provide a number for complaints.
After 2 weeks, still missing electricity and telecommunications
(plus fuel, water, etc)
«»pleas edo not blame us, we will do our best
Anticipating users dissatisfaction, and trying to do something to manipulate their perceptio and shed away responsibility
We started to see how disasters are born by looking at the process of disaster minimization, and at its key steps.
We nowask ourselves. Given these pathologies, and such a process, ….
One cannot be naive: in 20201, tehre has been a lot of papers oan dsolutions for post disaster and emergency comms.
Many solutionas, but also many weaknesses which mad ethem highly uneffective during a disaster.
So what happens?
Should we make better solutions? Orrather fight for adoption?
The answer (for me) is: both.
Facilitate adoption by proposing new technologies which are cheaper, easier to deploy and easier/cheaper to prepare (typically, empowering people: FC)
Easy to observe how disasetrs strike more poor countries, which cannot allocate enough respourecs (see definition)
Anyway, heer we are engineers, we cannot act on that
Btw, some disasters are «planned»: resources are there, but not allocated due to poor policy and political decisions
What we can do is proose technological solutiosn which are easier to adopt and thus cheaper, lowering the cost (in its various dimensions) of disaster preparedness and of post disaster interventions.
In this, information diffusion plays a key rôle…
Information diffusion play sa key rôle in all this
Conservative dimensioning and overprovisioning
Protection against conditions which are unforeseeable or hard to characterize beforehand is very expensive
Cost/benefit analysis would require estimates of likelihood of disaster, which is often very hard
We focus on natural disasters, though many things here can be extended to disasters caused by other kind of events
Damage to economy, disruption of people’s lifes and change in their outlook (place not safe anymore)
So not only restoring and make ii work despite of the disatser, but cater for the specific needs of a disaster
Of cousre, this snot only seen in L’aquila…
4 months later, still significant differences in pop density,a s inffreed by call records!
Disasters generate spikes, several time the amount of ordinary traffic
Events do it too, but less abrupt
And in planned fashion
(when unlanned… same problems!)
Events which are not scares but which hold consequences, leave permament change son patterns
More localized events cause waves
Larger ones less (or they are less visible, just out of scale, marginale ffect)
What did we learn?
Problem is larger, involving politics, economy, culture
Great heterogeneity,
Fast onset, geographicalyl extended,
Generate waves, with implicatiosn even out of the immediate disaster area
Design has to take into account the peculiariuties of tehs pecific disaster:
inadequate situation awareness in disasters has been identified as one of the primary factors in human errors with grave consequences such as loss of lives and destruction of critical infrastructure
As water levels began rising across Florida in Hurricane Irma's early days last week, many Florida residents were urged to download Zello, an app that served as a digital walkie-talkie of sorts. When Hurricane Harvey devastated Houston and its surrounding areas just one week prior, Zello had emerged as a literal lifeline, allowing Good Samaritans to coordinate and offer aid to first responder rescue efforts.
While Houston-area emergency services used social media to broadcast updates, Texans affected by Harvey's torrential waters were generally discouraged by official sources from using Twitter, Facebook, and other forms of social sharing, citing concerns about robbery, misinformation, and widespread panic. But the immediate impact of Zello and other forms of social media for Harvey evacuees can't be denied—Zello has now topped the charts of the iTunes store, and emergency responders are beginning to cite social media as a tool to utilize.
Voice communications are key, but definitely not enough
Exploiting the information richness of the local environment is key
Empowering affected population with data services (e.g. for social media support) allows addressing a large amount of the need for information
of the population
of coordination, monitoring and first responder services
Idea comes from seeing how social networks impacted (by facilitating) teh life and cateerd for the needs of people in a disaster setting
No full connectivity, not possible anyway, unrealistic via p2p and ah hoc (too much resources, bandwidth and energy)
Illustrate the basics
Small rounds: wireless nodes in the plane, moving (or not) according to arbitrary mm
Dashed line: transmission range
We assume that:
At some point in tuime, one por more nodes possess the content
This co0ntent is of interest, and should be delivered, to all those ndoes entering a given region,
Here circular (AZ)
Primary performnec goal is persistence
However, not enough for pplications: majke example. Attdcation site, or cinema
Thus, Zoi and success ratio
Those who do not get it, get it from the infrastructire 8possibly). Or failéure acceptable.
To illustrate how FC can be applied in proximity based, we will make use of two applications which we have proposed and which we are currently investigating
Everyone possesses a piece of the puzzle, a piece of info
Alone, each sensor is useless (takes time to collect info and make up a common view)
Organize interaction in the form of a map
The app starts with fluctuating in the region affected a
map of the region itself. Each participant then enriches the map with geographically
contextualized information, and floats the resulting, enriched map. Whenever a user
receives different versions of the same enriched map (with different tags and information),
the user consolidates the information, possibly eliminating duplicated data
and outdated information.
How to use a non real time, slow comm service to speed up and make more efficient rescue services
However, nodes move….
How to guarantee performance of storage and delivery in such a setting?
How to engineer such a system for reliable (stochastic) performance guarantees?
However, nodes move….
How to guarantee performance of storage and delivery in such a setting?
How to engineer such a system for reliable (stochastic) performance guarantees?
The app starts with fluctuating in the region affected a
map of the region itself. Each participant then enriches the map with geographically
contextualized information, and floats the resulting, enriched map. Whenever a user
receives different versions of the same enriched map (with different tags and information),
the user consolidates the information, possibly eliminating duplicated data
and outdated information.
Decentralized ML algorithms have been proposed to tackle scalability issues. In these algorithms, learning is implemented collaboratively amongst all nodes with no central server. One of the state-of-the-art approaches in this field is Gossip Learning, which implements a decentralized version of FL
. Each node in this distributed algorithm acts as a client for other nodes. At the same time, it can act as a coordinating server that merges received models.
This protocol uses asynchronous gossip communications to generate a global model,
each node executes a merge- update- send cycle, in which when a node receives models from neighbors, it produces an aggregate model (merging), and updates the model by training the aggregated model on its local dataset and then sends it to some random peer or its one-hop neighborhood. This process is repeated continuously until the maximum number of iterations is reached or the global model accuracy is greater than a threshold.