https://telecombcn-dl.github.io/2018-dlai/
Deep learning technologies are at the core of the current revolution in artificial intelligence for multimedia data analysis. The convergence of large-scale annotated datasets and affordable GPU hardware has allowed the training of neural networks for data analysis tasks which were previously addressed with hand-crafted features. Architectures such as convolutional neural networks, recurrent neural networks or Q-nets for reinforcement learning have shaped a brand new scenario in signal processing. This course will cover the basic principles of deep learning from both an algorithmic and computational perspectives.
E-Mail Security: Pretty Good Privacy, S/MIME IP Security: IP Security overview, IP Security architecture, Authentication Header, Encapsulating security payload, Combining security associations, Internet Key Exchange Case Studies on Cryptography and security: Secure Multiparty Calculation, Virtual Elections, Single sign On, Secure Inter-branch Payment Transactions, Cross site Scripting Vulnerability.
https://telecombcn-dl.github.io/2018-dlai/
Deep learning technologies are at the core of the current revolution in artificial intelligence for multimedia data analysis. The convergence of large-scale annotated datasets and affordable GPU hardware has allowed the training of neural networks for data analysis tasks which were previously addressed with hand-crafted features. Architectures such as convolutional neural networks, recurrent neural networks or Q-nets for reinforcement learning have shaped a brand new scenario in signal processing. This course will cover the basic principles of deep learning from both an algorithmic and computational perspectives.
E-Mail Security: Pretty Good Privacy, S/MIME IP Security: IP Security overview, IP Security architecture, Authentication Header, Encapsulating security payload, Combining security associations, Internet Key Exchange Case Studies on Cryptography and security: Secure Multiparty Calculation, Virtual Elections, Single sign On, Secure Inter-branch Payment Transactions, Cross site Scripting Vulnerability.
Machine learning on graphs is an important and ubiquitous task with applications ranging from drug design to friendship recommendation in social networks. The primary challenge in this domain is finding a way to represent, or encode, graph structure so that it can be easily exploited by machine learning models. However, traditionally machine learning approaches relied on user-defined heuristics to extract features encoding structural information about a graph. In this talk I will discuss methods that automatically learn to encode graph structure into low-dimensional embeddings, using techniques based on deep learning and nonlinear dimensionality reduction. I will provide a conceptual review of key advancements in this area of representation learning on graphs, including random-walk based algorithms, and graph convolutional networks.
Improving epidemiological research: avoiding the statistical paradoxes and fa...Maarten van Smeden
Keynote at Norwegian Epidemiological Association conference, October 26 2022. Discussing absence of evidence fallacy, Table 2 fallacy, Winner's curse and Stein's paradox.
Explainable AI is not yet Understandable AIepsilon_tud
Keynote of Dr. Nava Tintarev at RCIS'2020. Decision-making at individual, business, and societal levels is influenced by online content. Filtering and ranking algorithms such as those used in recommender systems are used to support these decisions. However, it is often not clear to a user whether the advice given is suitable to be followed, e.g., whether it is correct, whether the right information was taken into account, or if the user’s best interests were taken into consideration. In other words, there is a large mismatch between the representation of the advice by the system versus the representation assumed by its users. This talk addresses why we (might) want to develop advice-giving systems that can explain themselves, and how we can assess whether we are successful in this endeavor. This talk will also describe some of the state-of-the-art in explanations in a number of domains (music, tweets, and news articles) that help link the mental models of systems and people. However, it is not enough to generate rich and complex explanations; more is required in order to understand and be understood. This entails among other factors decisions around which information to select to show to people, and how to present that information, often depending on the target users and contextual factors
Social Media Mining - Chapter 7 (Information Diffusion)SocialMediaMining
R. Zafarani, M. A. Abbasi, and H. Liu, Social Media Mining: An Introduction, Cambridge University Press, 2014.
Free book and slides at http://socialmediamining.info/
Machine learning on graphs is an important and ubiquitous task with applications ranging from drug design to friendship recommendation in social networks. The primary challenge in this domain is finding a way to represent, or encode, graph structure so that it can be easily exploited by machine learning models. However, traditionally machine learning approaches relied on user-defined heuristics to extract features encoding structural information about a graph. In this talk I will discuss methods that automatically learn to encode graph structure into low-dimensional embeddings, using techniques based on deep learning and nonlinear dimensionality reduction. I will provide a conceptual review of key advancements in this area of representation learning on graphs, including random-walk based algorithms, and graph convolutional networks.
Improving epidemiological research: avoiding the statistical paradoxes and fa...Maarten van Smeden
Keynote at Norwegian Epidemiological Association conference, October 26 2022. Discussing absence of evidence fallacy, Table 2 fallacy, Winner's curse and Stein's paradox.
Explainable AI is not yet Understandable AIepsilon_tud
Keynote of Dr. Nava Tintarev at RCIS'2020. Decision-making at individual, business, and societal levels is influenced by online content. Filtering and ranking algorithms such as those used in recommender systems are used to support these decisions. However, it is often not clear to a user whether the advice given is suitable to be followed, e.g., whether it is correct, whether the right information was taken into account, or if the user’s best interests were taken into consideration. In other words, there is a large mismatch between the representation of the advice by the system versus the representation assumed by its users. This talk addresses why we (might) want to develop advice-giving systems that can explain themselves, and how we can assess whether we are successful in this endeavor. This talk will also describe some of the state-of-the-art in explanations in a number of domains (music, tweets, and news articles) that help link the mental models of systems and people. However, it is not enough to generate rich and complex explanations; more is required in order to understand and be understood. This entails among other factors decisions around which information to select to show to people, and how to present that information, often depending on the target users and contextual factors
Social Media Mining - Chapter 7 (Information Diffusion)SocialMediaMining
R. Zafarani, M. A. Abbasi, and H. Liu, Social Media Mining: An Introduction, Cambridge University Press, 2014.
Free book and slides at http://socialmediamining.info/
Important spreaders in networks: exact results on small graphsPetter Holme
To be able to control spreading phenomena (like the spreading of diseases and information) in networks it is important to identify influential spreaders. What "important" means depends on what is spreading and what kind of countermeasures that are available. In this work, we let the susceptible-infected-removed (SIR) model represent the spreading dynamics and contrast three different definitions of importance: Influence maximization (the expected outbreak size given a set of seed nodes), the effect of vaccination (how much deleting nodes would reduce the expected outbreak size) and sentinel surveillance (how early an outbreak could be detected with sensors at a set of nodes). We calculate the exact expressions of these quantities, as functions of the SIR parameters, for all connected graphs of three to seven nodes. We obtain the smallest graphs where the optimal node sets are not overlapping. We find that: node separation is more important than centrality for more than one active node, that vaccination and influence maximization are the most different aspects of importance, and that the three aspects are more similar when the infection rate is low. Furthermore, we discuss similar approaches to study the extinction times in the susceptible-infected- susceptible model.
Exploring the Current Trends and Future Prospects in Terrorist Network Mining cscpconf
In today’s era of hi-tech technologies, criminals are easily fulfilling their inhuman goals against
the mankind. Thus, the security of civilians has significantly become important. In this regard,
the law-enforcement agencies are aiming to prevent future attacks. To do so, the terrorist
networks are being analyzed using data mining techniques. One such technique is Social network analysis which studies terrorist networks for the identification of relationships and associations that may exist between terrorist nodes. Terrorist activities can also be detected by means of analyzing Web traffic content. This paper studies social network analysis, web traffic content and explores various ways for identifying terrorist activities.
Clustering mechanism is the unsupervised classification of patterns observations data items or feature vectors into different clusters. This type of clustering problem has been addressed in many contexts and by researchers in different domains; this makes us to understand its broad appeal and usefulness as one of the steps analysing the whole data. As we all know that there will be huge assumptions in solving the clustering problems which makes it very complex and the clustering process became very slow. Here in this paper we are concentrating on overview of pattern clustering methods from a statistical pattern recognition perspective with a goal of providing useful advice and references to fundamental concepts accessible to the broad community of clustering practitioners. We also present the study of different clustering algorithms as well as the current development in these mechanisms.
ALTERNATIVES TO BETWEENNESS CENTRALITY: A MEASURE OF CORRELATION COEFFICIENTcsandit
In this paper, we measure and analyze the correlation of betweenness centrality (BWC) to five centrality measures, including eigenvector centrality (EVC), degree centrality DEG),
clustering coefficient centrality (CCC), farness centrality (FRC), and closeness centrality(CLC). We simulate the evolution of random networks and small-world networks to test the correlation between BWC and the five measures. Additionally, nine real-world networks are also involved in our present study to further examine the correlation. We find that DEG is
highly correlated to BWC on most cases and can serve as alternative to computationallyexpensive BWC. Moreover, EVC, CLC and FRC are also good candidates to replace BWC on
random networks. Although it is not a perfect correlation for all the real-world networks, there still exists a relatively good correlation between BWC and other three measures (CLC, FRC and EVC) on some networks. Our findings in this paper can help us understand how BWC correlates to other centrality measures and when to decide a good alternative to BWC
Massively Parallel Simulations of Spread of Infectious Diseases over Realisti...Subhajit Sahu
Highlighted notes while preparing for project on Computational Epidemics:
Massively Parallel Simulations of Spread of Infectious Diseases over Realistic Social Networks
Abhinav Bhatele, Jae-Seung Yeom, Nikhil Jain, Chris J. Kuhlman, Yarden Livnat, Keith R. Bisset, Laxmikant V. Kale, Madhav V. Marathe
Controlling the spread of infectious diseases in large populations is an important societal challenge. Mathematically, the problem is best captured as a certain class of reactiondiffusion processes (referred to as contagion processes) over appropriate synthesized interaction networks. Agent-based models have been successfully used in the recent past to study such contagion processes. We describe EpiSimdemics, a highly scalable, parallel code written in Charm++ that uses agent-based modeling to simulate disease spreads over large, realistic, co-evolving interaction networks. We present a new parallel implementation of EpiSimdemics that achieves unprecedented strong and weak scaling on different architectures — Blue Waters, Cori and Mira. EpiSimdemics achieves five times greater speedup than the second fastest parallel code in this field. This unprecedented scaling is an important step to support the long term vision of realtime epidemic science. Finally, we demonstrate the capabilities of EpiSimdemics by simulating the spread of influenza over a realistic synthetic social contact network spanning the continental United States (∼280 million nodes and 5.8 billion social contacts).
Massively Parallel Simulations of Spread of Infectious Diseases over Realisti...Subhajit Sahu
Highlighted notes while studying for project work:
Massively Parallel Simulations of Spread of Infectious Diseases over Realistic Social Networks
Abhinav Bhatele†
Jae-Seung Yeom†
Nikhil Jain†
Chris J. Kuhlman∗
Yarden Livnat‡
Keith R. Bisset∗
Laxmikant V. Kale§
Madhav V. Marathe∗
†Center for Applied Scientific Computing, Lawrence Livermore National Laboratory, Livermore, California 94551 USA
∗Biocomplexity Institute & Department of Computer Science, Virginia Tech, Blacksburg, Virginia 24061 USA
‡Scientific Computing and Imaging Institute, University of Utah, Salt Lake City, Utah 84112 USA
§Department of Computer Science, University of Illinois at Urbana-Champaign, Urbana, Illinois 61801 USA
E-mail: †{bhatele, yeom2, nikhil}@llnl.gov, ∗{ckuhlman, kbisset, mmarathe}@vbi.vt.edu
Abstract—Controlling the spread of infectious diseases in large populations is an important societal challenge. Mathematically, the problem is best captured as a certain class of reactiondiffusion processes (referred to as contagion processes) over appropriate synthesized interaction networks. Agent-based models have been successfully used in the recent past to study such contagion processes. We describe EpiSimdemics, a highly scalable, parallel code written in Charm++ that uses agent-based modeling to simulate disease spreads over large, realistic, co-evolving interaction networks. We present a new parallel implementation of EpiSimdemics that achieves unprecedented strong and weak scaling on different architectures — Blue Waters, Cori and Mira. EpiSimdemics achieves five times greater speedup than the second fastest parallel code in this field. This unprecedented scaling is an important step to support the long term vision of realtime epidemic science. Finally, we demonstrate the capabilities of EpiSimdemics by simulating the spread of influenza over a realistic synthetic social contact network spanning the continental United States (∼280 million nodes and 5.8 billion social contacts).
An Opportunistic AODV Routing Scheme : A Cognitive Mobile Agents Approachijasuc
In Manet’s Dynamics and Robustness are the key features of the nodes and are governed by several routing protocols such as AODV, DSR and so on. However in the network the growing resource demand leads to resource scarcity. The Node Mobility often leads to the link breakages and high routing overhead
decreasing the stability and reliability of the network connectivity. In this context, the paper proposes a novel opportunistic AODV routing scheme which implements a cognitive agent based intelligent technique to set up a stable connectivity over the Manet. The Scheme computes the routing metric (rf) based on the collaboration sensitivity levels of the nodes obtained based through the knowledge-based decision. This Routing Metric is subsequently used to set up the stable path for network connectivity. Thus minimizes the route overhead and increases the stability of the path. The Performance evaluation is conducted in comparison with the AODV and sleep AODV routing protocol and validated.
An Opportunistic AODV Routing Scheme : A Cognitive Mobile Agents Approachjake henry
In Manet’s Dynamics and Robustness are the key feat
ures of the nodes and are governed by several routi
ng
protocols such as AODV, DSR and so on. However in t
he network the growing resource demand leads to
resource scarcity. The Node Mobility often leads to
the link breakages and high routing overhead
decreasing the stability and reliability of the net
work connectivity. In this context, the paper propo
ses a
novel opportunistic AODV routing scheme which imple
ments a cognitive agent based intelligent technique
to set up a stable connectivity over the Manet. The
Scheme computes the routing metric (rf) based on t
he
collaboration sensitivity levels of the nodes obtai
ned based through the knowledge-based decision. Thi
s
Routing Metric is subsequently used to set up the s
table path for network connectivity. Thus minimizes
the
route overhead and increases the stability of the p
ath. The Performance evaluation is conducted in
comparison with the AODV and sleep AODV routing pro
tocol and validated
An Opportunistic AODV Routing Scheme : A Cognitive Mobile Agents Approachijasuc
In Manet’s Dynamics and Robustness are the key features of the nodes and are governed by several routing protocols such as AODV, DSR and so on. However in the network the growing resource demand leads to resource scarcity. The Node Mobility often leads to the link breakages and high routing overhead
decreasing the stability and reliability of the network Connectivity. In this context, the paper proposes a novel opportunistic AODV routing scheme which implements a cognitive agent based intelligent technique to set up a stable connectivity over the Manet. The Scheme computes the routing metric (rf) based on the collaboration sensitivity levels of the nodes obtained based through the knowledge-based decision. This Routing Metric is subsequently used to set up the stable path for network connectivity. Thus minimizes the route overhead and increases the stability of the path. The Performance evaluation is conducted in comparison with the AODV and sleep AODV routing protocol and validated.
AN OPPORTUNISTIC AODV ROUTING SCHEME: A COGNITIVE MOBILE AGENTS APPROACHijasuc
In Manet’s Dynamics and Robustness are the key features of the nodes and are governed by several routing
protocols such as AODV, DSR and so on. However in the network the growing resource demand leads to
resource scarcity. The Node Mobility often leads to the link breakages and high routing overhead
decreasing the stability and reliability of the network connectivity. In this context, the paper proposes a
novel opportunistic AODV routing scheme which implements a cognitive agent based intelligent technique
to set up a stable connectivity over the Manet. The Scheme computes the routing metric (rf) based on the
collaboration sensitivity levels of the nodes obtained based through the knowledge-based decision. This
Routing Metric is subsequently used to set up the stable path for network connectivity. Thus minimizes the
route overhead and increases the stability of the path. The Performance evaluation is conducted in
comparison with the AODV and sleep AODV routing protocol and validated.
Knowledge discovery in databases (KDD) is a process which consists of multiple contiguous steps, done iteratively to find knowledge from very large datasets. The data-mining step of knowledge discovery is computationally time consuming , if we work with very large databases. Clustering is one form of unsupervised learning in machine learning techniques used for data mining. Here, a Meta learning approach is taken by me that is used to scale supervised learning, to scale unsupervised learning. This approach uses a number of clustering algorithms to be performed for the base classification systems and then it learns relationships among the base classification systems to achieve final predictions. The base classifiers can be executed on independent computers to perform not only processor scaling, but data scaling also.
Data Centers - Striving Within A Narrow Range - Research Report - MCG - May 2...pchutichetpong
M Capital Group (“MCG”) expects to see demand and the changing evolution of supply, facilitated through institutional investment rotation out of offices and into work from home (“WFH”), while the ever-expanding need for data storage as global internet usage expands, with experts predicting 5.3 billion users by 2023. These market factors will be underpinned by technological changes, such as progressing cloud services and edge sites, allowing the industry to see strong expected annual growth of 13% over the next 4 years.
Whilst competitive headwinds remain, represented through the recent second bankruptcy filing of Sungard, which blames “COVID-19 and other macroeconomic trends including delayed customer spending decisions, insourcing and reductions in IT spending, energy inflation and reduction in demand for certain services”, the industry has seen key adjustments, where MCG believes that engineering cost management and technological innovation will be paramount to success.
MCG reports that the more favorable market conditions expected over the next few years, helped by the winding down of pandemic restrictions and a hybrid working environment will be driving market momentum forward. The continuous injection of capital by alternative investment firms, as well as the growing infrastructural investment from cloud service providers and social media companies, whose revenues are expected to grow over 3.6x larger by value in 2026, will likely help propel center provision and innovation. These factors paint a promising picture for the industry players that offset rising input costs and adapt to new technologies.
According to M Capital Group: “Specifically, the long-term cost-saving opportunities available from the rise of remote managing will likely aid value growth for the industry. Through margin optimization and further availability of capital for reinvestment, strong players will maintain their competitive foothold, while weaker players exit the market to balance supply and demand.”
Explore our comprehensive data analysis project presentation on predicting product ad campaign performance. Learn how data-driven insights can optimize your marketing strategies and enhance campaign effectiveness. Perfect for professionals and students looking to understand the power of data analysis in advertising. for more details visit: https://bostoninstituteofanalytics.org/data-science-and-artificial-intelligence/
Chatty Kathy - UNC Bootcamp Final Project Presentation - Final Version - 5.23...John Andrews
SlideShare Description for "Chatty Kathy - UNC Bootcamp Final Project Presentation"
Title: Chatty Kathy: Enhancing Physical Activity Among Older Adults
Description:
Discover how Chatty Kathy, an innovative project developed at the UNC Bootcamp, aims to tackle the challenge of low physical activity among older adults. Our AI-driven solution uses peer interaction to boost and sustain exercise levels, significantly improving health outcomes. This presentation covers our problem statement, the rationale behind Chatty Kathy, synthetic data and persona creation, model performance metrics, a visual demonstration of the project, and potential future developments. Join us for an insightful Q&A session to explore the potential of this groundbreaking project.
Project Team: Jay Requarth, Jana Avery, John Andrews, Dr. Dick Davis II, Nee Buntoum, Nam Yeongjin & Mat Nicholas
06-04-2024 - NYC Tech Week - Discussion on Vector Databases, Unstructured Data and AI
Round table discussion of vector databases, unstructured data, ai, big data, real-time, robots and Milvus.
A lively discussion with NJ Gen AI Meetup Lead, Prasad and Procure.FYI's Co-Found
Levelwise PageRank with Loop-Based Dead End Handling Strategy : SHORT REPORT ...Subhajit Sahu
Abstract — Levelwise PageRank is an alternative method of PageRank computation which decomposes the input graph into a directed acyclic block-graph of strongly connected components, and processes them in topological order, one level at a time. This enables calculation for ranks in a distributed fashion without per-iteration communication, unlike the standard method where all vertices are processed in each iteration. It however comes with a precondition of the absence of dead ends in the input graph. Here, the native non-distributed performance of Levelwise PageRank was compared against Monolithic PageRank on a CPU as well as a GPU. To ensure a fair comparison, Monolithic PageRank was also performed on a graph where vertices were split by components. Results indicate that Levelwise PageRank is about as fast as Monolithic PageRank on the CPU, but quite a bit slower on the GPU. Slowdown on the GPU is likely caused by a large submission of small workloads, and expected to be non-issue when the computation is performed on massive graphs.
Levelwise PageRank with Loop-Based Dead End Handling Strategy : SHORT REPORT ...
Spreading Phenomena in Social Networks
1. Spreading Phenomena in Complex Networks
Shubhabrata Ghosh Manojit Chakraborty Souvik Das Pallavi Mazumder
Heritage Institute of Technology, Kolkata
Dept. of Computer Science and Engineering
April 3, 2017
c Manojit Chakraborty (HIT-K) Spreading Phenomena April 3, 2017 1 / 62
3. A Story to Start
On the night of February 21, 2003, a physician from southern China checked into the
Metropole Hotel in Hong Kong.He previously treated patients suering from a disease that
was called atypical pneumonia.
Next day, after leaving the hotel, he went to the local hospital, this time as a patient. He
died there several days later of atypical pneumonia
That night sixteen other guests of the Metropole Hotel also contracted the disease that
was named Severe Acute Respiratory Syndrome, or SARS.
c Manojit Chakraborty (HIT-K) Spreading Phenomena April 3, 2017 3 / 62
4. A Story to Start
These guests carried the SARS virus with them
to Hanoi, Singapore, and Toronto, sparking
outbreaks in each of those cities.
Super Spreader
The physician became an example of a Super
Spreader, an individual who is responsible for a
disproportionate number of infections during an
epidemic.
Hubs
A network theorist will recognize Super
Spreaders as Hubs, nodes with an exceptional
number of links in the contact network on
which a disease spreads
.
Figure: Super Spreaders
c Manojit Chakraborty (HIT-K) Spreading Phenomena April 3, 2017 4 / 62
5. Introduction
Complex Networks are everywhere.They crop up wherever there are interactions between actors.
Phenomena Agent Network
Venereal disease Pathogens Sexual network
Research Paper Scientists Citation network
Rumor spreading Information, memes Communication network
Computer viruses Digital viruses Internet network
Bedbugs Parasitic insects Hotel-traveler network
Malaria Plasmodium Mosquito-human Network
Table: Dierent agents and corresponding networks
c Manojit Chakraborty (HIT-K) Spreading Phenomena April 3, 2017 5 / 62
6. Network Representation
Networks portray the interactions between dierent actors.
Actors or individuals are nodes/vertices in
the graph
If there's interaction between two nodes,
there's an edge/link between them
The links can have weights or intensities
signifying the strength of connections
The links can be directed, like in the web
graph. There's a directed link between two
nodes (pages) A and B if there's a
hyperlink to B from A
Figure: Networks
c Manojit Chakraborty (HIT-K) Spreading Phenomena April 3, 2017 6 / 62
7. Population Representation
As time progresses, human population has been demonstrated using dierent representations by
various scientists. Each representation gave a better analogy than its previous one.
Homogeneous Mixing
Random Network by Erdos-Renyi (1959) [1]
Scale Free Network by Albert-Barabasi (1999) [2]
Figure: Random Network Figure: Scale Free Network
c Manojit Chakraborty (HIT-K) Spreading Phenomena April 3, 2017 7 / 62
10. SUSCEPTIBLE-INFECTED (SI) MODEL
At rst we will simply demonstrate dierent epidemic models on Homogeneous Mixing
representation of a population.
S: Susceptible individuals.
I: Infected individuals, when infected they
can infect others continuously
N: Total population.
β: Likelihood of transmission of disease
from Infected to Susceptible
k: average number of contacts a typical
individual has
Susceptible contacts per unit of time
βkS
N
Overall rate of infection
dI(t)
dt
=
I(t)βkS(t)
N
Figure: SI Model
c Manojit Chakraborty (HIT-K) Spreading Phenomena April 3, 2017 10 / 62
11. SI Model
di
dt
= iβks, By solving this equation,
i =
i0e βkt
1 − i0 + i0e βkt
At the beginning the fraction of infected individuals
increases exponentially
With time an infected individual encounters fewer
and fewer susceptible individuals. Hence the growth
of i slows for large t
Figure: SI Model Graph
c Manojit Chakraborty (HIT-K) Spreading Phenomena April 3, 2017 11 / 62
12. Susceptible Infected Susceptible(SIS) Model
It has the same two states as the SI
Model, susceptible and infected.
The dierence is now infected individuals
recover at a xed rate µ, becoming
susceptible again
The equation describing the dynamics of this
model :-
di
dt
= βki(1 − i) − µi
Figure: SIS Model
c Manojit Chakraborty (HIT-K) Spreading Phenomena April 3, 2017 12 / 62
13. SIS Model
Figure: SIS Model Graph
c Manojit Chakraborty (HIT-K) Spreading Phenomena April 3, 2017 13 / 62
14. Susceptible Infected Recovered(SIR) Model
In the SIR model recovered individuals
enter a recovered state, meaning that they
develop immunity rather than becoming
susceptible again.
The dierential equations for the
susceptible s, infected i and the removed r
state.
ds
dt
= −βki(1 − r − i)
di
dt
= −µi + βki(1 − r − i)
dr
dt
= µi
Figure: SIR Model
c Manojit Chakraborty (HIT-K) Spreading Phenomena April 3, 2017 14 / 62
15. SIR Model
Figure: SIR Model
c Manojit Chakraborty (HIT-K) Spreading Phenomena April 3, 2017 15 / 62
16. Comparison between Models
Figure: Comparing SI, SIS, SIR Models
c Manojit Chakraborty (HIT-K) Spreading Phenomena April 3, 2017 16 / 62
17. Network Epidemics
Network Epidemics
Individual can transmit a pathogen only to those
they come into contact with, hence pathogens
spread on a complex contact network. .
These contact networks are often scale-free, hence
k is not sucient to characterize their
topology.
The failure of the basic hypotheses prompted a
fundamental revision of the epidemic modeling
framework by Romualdo Pastor-Satorras and
Alessandro Vespignani in 2001[3] Figure: The Great Plague
c Manojit Chakraborty (HIT-K) Spreading Phenomena April 3, 2017 17 / 62
18. SI Model on a Network
Degree Block Approximation
A mathematical formalism that is used to
distinguish nodes based on their degree.
This assumes that nodes with the same
degree are statistically equivalent.
Thus, the fraction of nodes with degree k
that are infected among all Nk degree-k
nodes in the network is denoted by:
ik =
Ik
Nk
Figure: Degree Block Approximation
c Manojit Chakraborty (HIT-K) Spreading Phenomena April 3, 2017 18 / 62
19. SI Model on a Network
The total fraction of infected nodes is the sum of all infected degree-k nodes:
i = k pkik
Given the dierent node degrees, we write the SI model for each degree k separately:
dik
dt
=β(1 − ik)kθk
The infection rate is proportional to β and the fraction of degree-k nodes that are not yet
infected, is (1 − ik).
c Manojit Chakraborty (HIT-K) Spreading Phenomena April 3, 2017 19 / 62
20. SI Model(Homogeneous) vs SI Model(Network)
The average degree k in case of homogenous mixing is replaced with each node's
actual degree k.
The density function θk represents the fraction of infected neighbors of a susceptible node
k. But in case of homogenous mixing assumption θk is simply the fraction of the infected
nodes, i.
While, in case of homogenous mixing, there's just a single equation which explains the time
dependent behavior of the whole system. But in a network,
dik
dt
=β(1 − ik)kθk represents a
system of kmax coupled equations, one equation for each degree present in the network.
c Manojit Chakraborty (HIT-K) Spreading Phenomena April 3, 2017 20 / 62
21. SI Model on a Network
On solving the equation:
dik
dt
= β(1 − ik)kθk, we get:
ik = io(1 +
k( k −1)
k2
− k
(e
t
τSI
-1))
where (τSI
) is the characteristic time for the
spread of the pathogen.
τSI
=
k
β( k2
− k )
Figure: SI Model
c Manojit Chakraborty (HIT-K) Spreading Phenomena April 3, 2017 21 / 62
22. SI Model on a Network
The higher the degree of a node, the more
likely that it becomes infected. For any time
t we can write ik = g(t) + kf(t), indicating
that the group of nodes with higher degree
has a higher fraction of infected nodes
(Figure alongside).
Since i = k pkik, the total fraction of
infected nodes grows with time as:
i =
kmax
0
ik pk dk
= i0(1 +
( k 2 − k )
k2 − k
(e
t
τSI − 1)
Figure: Fraction of Infected Nodes in SI Model
c Manojit Chakraborty (HIT-K) Spreading Phenomena April 3, 2017 22 / 62
23. Network Epidemics
Now we will derive τSI
for dierent networks. But before that we need to know what are the
two types of networks we are concerned with.
Random Networks
A random network consists of N nodes where each node pair is connected with probability p.
For a large N, it's degree distribution follows Poisson's distribution.
Scale Free Network
This is a network whose degree distribution follows a power law. That is, the fraction P(k) of
nodes in the network having k connections to other nodes goes for large values of k as: P(k)
∼ k−γ where 2γ3.
c Manojit Chakraborty (HIT-K) Spreading Phenomena April 3, 2017 23 / 62
24. τSI
FOR DIFFERENT NETWORKS
Random Networks
Here, k2 = k ( k +1), obtaining τ ER
SI
=
1
β k
which is the same for homogenous networks.
Scale-free network with γ ≥ 3
If the contract network is scale-free with degree exponent γ ≥ 3, both k and k2 are
nite. Consequently τSI
is also nite and the spreading dynamics is similar to a random
network but with an altered τSI
.
Scale-free network with γ ≤ 3
For γ ≤ 3 in the N → ∞ limit k2 → ∞ hence
τSI
=
k
β( k2
− k )
predicts τSI
→ 0 In other words, the spread of a pathogen on a
scale-free network is instantaneous.
c Manojit Chakraborty (HIT-K) Spreading Phenomena April 3, 2017 24 / 62
25. SIS Model on a Network
In case of Epidemic Modeling, the equation for SI model was:
dik
dt
= β(1 − ik)kθk
The continuum equation describing the dynamics of the SIS model on a network is a
straightforward extension of the SI model
dik
dt
= β(1 − ik)kθk − µik
The dierence between the two equations is the presence of the recovery term -µik.
This changes the characteristic time of the epidemic to:
τSIS
=
k
β( k2
−µ k )
c Manojit Chakraborty (HIT-K) Spreading Phenomena April 3, 2017 25 / 62
26. Spreading Rate for Different Networks
Spreading Rate
Spreading rate(λ) of a pathogen is dened as the ratio of transmission probability β and the
recovery rate µ. λ =
β
µ
The higher is λ, the more likely that the disease will spread. Yet, the number of infected
individuals does not increase gradually with λ. Rather, the pathogen can spread only if its
spreading rate exceeds an epidemic threshold λc.
Random Network
For a random network the epidemic threshold, λc =
1
k +1
If λ λc, the pathogen will spread until it reaches an endemic state, where a nite
fraction i(λ) of the population is infected at any time.
If λ λc, the pathogen dies out, i.e. i(λ)=0.
c Manojit Chakraborty (HIT-K) Spreading Phenomena April 3, 2017 26 / 62
27. Vanishing Epidemic Threshold
Scale-Free Network
For a scale-free network the epidemic
threshold, λc =
k
k2
As for a scale-free network k2 diverges
in the N→ ∞ limit, for large networks the
epidemic threshold is expected to vanish.
Figure: Epidemic Threshold
c Manojit Chakraborty (HIT-K) Spreading Phenomena April 3, 2017 27 / 62
28. Epidemic Models on Networks
CONCLUSION :
In a large scale-free
network τ=0
In large scale-free
network λc=0
Figure: Epidemic Models On Networks
c Manojit Chakraborty (HIT-K) Spreading Phenomena April 3, 2017 28 / 62
29. Contact Networks
Network epidemics predicts that the speed with
which a pathogen spreads depends on the degree
distribution of the relevant contact network.
We found that k2 aects both the characteristic
time τ and the epidemic threshold λc.
None of the precious ndings are consequential if
the network on which a pathogen spreads is
random- in that case the predictions of network
epidemics are indistinguishable from the predictions
of the traditional epidemic models encountered in
the previous slides. Figure: Face-to-face Contact Network
c Manojit Chakraborty (HIT-K) Spreading Phenomena April 3, 2017 29 / 62
30. Sexually Transmitted diseases
HIV, the pathogen responsible for AIDS, spreads
mainly through sexual intercourse.
The scale-free nature of the sexual network indicates
that most individuals have relatively few sexual
partners. A few individuals, however, had a high
number of sexual partners during their lifetime.
Consequently the sexual network has a high k2,
which lowers both τ and λc.
Figure: The Sex Web
c Manojit Chakraborty (HIT-K) Spreading Phenomena April 3, 2017 30 / 62
31. Airborne Diseases
To predict the spread of pathogens, we
must know how far infected individuals
travel.
In the context of epidemic phenomena, the
most studied mobility data comes from air
travel, the mode of transportation that
determines the speed with which a
pathogen moves around the globe.
Consequently the air transportation
network, that connects airports with direct
ights, plays a key role in modeling and
predicting the spread of pathogens Figure: Air Transportation Network
c Manojit Chakraborty (HIT-K) Spreading Phenomena April 3, 2017 31 / 62
32. Network Science
Networks are powerful models of complex
systems in various domains.
Due to limitations of data collection
techniques,static network representation of
a given system was studied earlier.
Many real-world systems are not static but
change over time.
Today it has become possible to record
temporal changes in network structure (or
topology).
Figure: Activity of Police (blue) and Fascists
(black) obtained from time slices.(by months)
c Manojit Chakraborty (HIT-K) Spreading Phenomena April 3, 2017 32 / 62
33. Temporal Network
Many studies performed under the assumption
of static network structures can now be
extended to take into account the network's
dynamics.
Data on time-varying networks are becoming
accessible across a variety of contexts.
This avalanche of data is prompting a surge of
activity in the eld of temporal networks
Figure: Temporal Network showing social
interactions
c Manojit Chakraborty (HIT-K) Spreading Phenomena April 3, 2017 33 / 62
34. Aggregated Network
To accurately predict an epidemic process we
must consider the fact that pathogens spread
on temporal networks, a topic of increasing
interest in network science
By ignoring the temporality of these contact
patterns, we typically overestimate the speed
and the extent of an outbreak.
Figure: Aggregated Network showing social
interactions
c Manojit Chakraborty (HIT-K) Spreading Phenomena April 3, 2017 34 / 62
35. Social Interactions
The timing of the interactions between two
connected nodes is random.
This means that the interevent times between
consecutive contacts follow an exponential
distribution, resulting in a random but uniform
sequence of events
Therefore the contact patterns have an
uneven,'bursty' character in time Figure: Social Media websites are popular among
all age groups
c Manojit Chakraborty (HIT-K) Spreading Phenomena April 3, 2017 35 / 62
36. Degree Correlation
Many social networks are assortative,
implying that high degree nodes tend to
connect to other high degree nodes. Do
they aect the spread of a pathogen?
Assortative correlations decrease λc and
dissasortative correlations increase it
Despite the changes in λc, for the SIS
model the epidemic threshold vanishes for
a scale-free network with diverging second
moment, whether the network is
assortative, neutral or disassortative Figure: Graph showing appearance of certain
minerals in certain foods
c Manojit Chakraborty (HIT-K) Spreading Phenomena April 3, 2017 36 / 62
37. Social Interactions
The mobile phone network allows us to explore
the role of tie strengths and communities on
spreading phenomena.
The spread of information on a weighted
mobile call graph, where the probability that a
node passes information to one of its neighbors
is proportional to the strength of the tie
between them. Figure: Temporal Network showing social
interactions
c Manojit Chakraborty (HIT-K) Spreading Phenomena April 3, 2017 37 / 62
38. Spreading conversation
Spreading in a small network
neighborhood, following the real link
weights.The information is released from
the red node, the arrow weight indicating
the tie strength.
The simulation was repeated 1,000 times.
The size of the arrowheads is proportional
to the number of times the information
was passed along the corresponding
direction, and the color indicates the total
number of transmissions along that link.
Figure: Link weight and communities
c Manojit Chakraborty (HIT-K) Spreading Phenomena April 3, 2017 38 / 62
39. Spreading conversation
Same in as previous case,but we assume
that each link has the same weight
w = wij
In the control simulation the information
tends to follow the shortest path. When
the link weights are taken into account,
information ows along a longer backbone
with strong ties.
Figure: Link weight and communities
c Manojit Chakraborty (HIT-K) Spreading Phenomena April 3, 2017 39 / 62
40. Contagion
Simple contagion is the process we
explored so far: It is sucient to come
into contact with an infected individual to
be infected. The spread of memes,
products and behavior is often described
by Complex contagion
The dierence between simple and
complex contagion is well captured by
Twitter data.
Figure: two types of contagion
c Manojit Chakraborty (HIT-K) Spreading Phenomena April 3, 2017 40 / 62
41. Immunization
Immunization strategies specify how vaccines,
treatments or drugs are distributed in the
population.
Yet, often cost considerations, the diculty of
reaching all individuals at risk, and real or
perceived side eects of the treatment prohibit
full coverage.
Figure: Patient being injected
c Manojit Chakraborty (HIT-K) Spreading Phenomena April 3, 2017 41 / 62
42. Immunization
The main purpose of immunization is to
protect the immunized individual from an
infection.
Secondary purpose is to reduce the speed
with which the pathogen spreads in a
population.
Eective degree of each susceptible node
changes from k to k (1 − g), which
decreases the spreading rate of the pathogen
from λ =
β
µ
to λ = λ(1 − g)
Figure: Vaccines are indispensable to stop spread
of diseases
c Manojit Chakraborty (HIT-K) Spreading Phenomena April 3, 2017 42 / 62
43. Random Network
If the pathogen spreads on a random network, for a high g the spreading rate λ could fall
below the epidemic threshold.The immunization rate gc is calculated as
gc = 1 −
µ
β k +1
if vaccination increases the fraction of immunized individuals above
gc, it pushes the spreading rate under the epidemic threshold λc.
In this case τ becomes negative and the pathogen dies out naturally. This explains why
health ocial encourage a high fraction of the population take the inuenza vaccine.
Similarly, a condom not only protects the individual who uses it from contacting the HIV,
but also decrease the rate at which AIDS spreads in the sexual network.
c Manojit Chakraborty (HIT-K) Spreading Phenomena April 3, 2017 43 / 62
44. Heterogeneous Network
In heterogeneous networks a virus can be
eradicated by increasing the epidemic threshold
through hub immunization. The gure shows
that, more hubs are immunized (i.e. the smaller
is k'max), the larger is λc, increasing the
chance that the disease dies out. Immunizing
the hubs changes the network on which the
disease spreads
Figure: immunization in heterogeneous networks
c Manojit Chakraborty (HIT-K) Spreading Phenomena April 3, 2017 44 / 62
45. Heterogeneous Network
For heterogeneous network the equation
becomes gc = 1 −
µ k
β k2
let us consider a digital virus spreading on the
email network. If we make the email network
random and undirected, we have k2 =3.26.
Using λ=1 in we obtain gc=0.76.
Yet, the email network is scale free with
k2 =1,271 (undirected version).In this case
predicts gc=0.997 for λ=1. Figure: immunization in heterogeneous networks
c Manojit Chakraborty (HIT-K) Spreading Phenomena April 3, 2017 45 / 62
46. Epidemic Eradication
Figure: Rahima Banu,the last smallpox infected patient in Bangladesh in 1976
c Manojit Chakraborty (HIT-K) Spreading Phenomena April 3, 2017 46 / 62
47. A Light in Darkness
During much of its history, humanity has been
helpless when faced with a pandemic. Lacking
drugs and vaccines, infectious diseases
repeatedly swept through continents,
decimating the world's population.
Despite the spectacular medical advances, we
have eective vaccines only against a small
number of pathogens. Consequently
transmission- reducing and
quarantine-based measures remain the main
tools of health professionals in combatting new
pathogens.
c Manojit Chakraborty (HIT-K) Spreading Phenomena April 3, 2017 47 / 62
48. Realtime Prediction
The real-time prediction of an epidemic
outbreak is a very recent development.
The 2009 H1N1 outbreak was the rst
beneciary of these developments, becoming
the rst pandemic whose spread was predicted
in real time.
The emergence of any new pathogen raises
several key questions.these questions are
addressed using powerful epidemic simulators.
Figure: Ebola virus
c Manojit Chakraborty (HIT-K) Spreading Phenomena April 3, 2017 48 / 62
49. REAL-TIME FORECAST
Epidemic forecast aims to foresee the real time spread of a pathogen, predicting the number
of infected individuals expected each week in each major city.
GLEAM( Global Epidemic and Mobility computational model )
GLEAM maps each geographic location into the nodes of a network.
Transport between these nodes, representing the links, are provided by global
transportation data, like airline schedules.
GLEAM estimates the epidemic parameters, like the transmission rate or reproduction
number, using a network-based approach.
It relies on chronological data that captures the worldwide spread of the pandemic, rather
than medical reports.
c Manojit Chakraborty (HIT-K) Spreading Phenomena April 3, 2017 49 / 62
50. Modeling the 2009 H1N1 Pandemic
c Manojit Chakraborty (HIT-K) Spreading Phenomena April 3, 2017 50 / 62
51. Modeling the 2009 H1N1 Pandemic
c Manojit Chakraborty (HIT-K) Spreading Phenomena April 3, 2017 51 / 62
52. Modelling the 2009 H1N1 Pandemic
For H1N1, the predictions were compared with data
collected from surveillance and virologic sources in 48
countries during the full course of the pandemic.
Peak Time
Peak time corresponds to the week when most
individuals are infected in a particular country.
Early Peak
GLEAM predicted that the H1N1 epidemic will peak
out in November, rather than in January or February,
the typical peak time of inuenza- like viruses.
The Impact of Vaccination
c Manojit Chakraborty (HIT-K) Spreading Phenomena April 3, 2017 52 / 62
53. What if Analysis
By incorporating the time and nature of each containment and mitigation procedure,
simulations can estimate the eciency of specic contingency plans.
Travel Restrictions
Given the important role air travel plays in the
spread of a pathogen, faced with a dangerous
pandemic, like an Ebola outbreak the rst
instinct is to restrict travel.
For example, there was a 40% decline in travel
to and from Mexico in May 2009, during the
H1N1 outbreak.
c Manojit Chakraborty (HIT-K) Spreading Phenomena April 3, 2017 53 / 62
54. What If Analysis
Antiviral Treatment
During the 2009 H1N1 pandemic Canada,
Germany, Hong Kong, Japan, the UK, and the
USA distributed antiviral drugs to mitigate the
impact of the disease. This prompted modelers
to ask what would have been the impact if all
countries that had drug stockpiles would have
distributed it to their population.
c Manojit Chakraborty (HIT-K) Spreading Phenomena April 3, 2017 54 / 62
55. Spread of a Pandemic
Effective Distance
Before there was a strong correlation
between the time of the outbreak and the
physical distance from the origin of the
outbreak.
Today, with airline travel, physical distance
has lost its relevance for epidemic
phenomena.
Thus, we replace the conventional
geographic distance with an eective
distance derived from the mobility
network [4].
c Manojit Chakraborty (HIT-K) Spreading Phenomena April 3, 2017 55 / 62
56. Spread of a Pandemic
Mobility Network
Each link is directed and
weighted, characterized by a ux
fraction 0 ≤ pij ≤ 1,fraction of
travelers that leave node i and
arrive at node j
The spread of a pathogen is
dominated by the most probable
trajectories predicted by the
mobility matrix pij. So, the
eective distance dij between
two connected locations i and j
dij = (1 − lnpij) ≥ 0
Note that dij = dji Figure: Mobility Network
c Manojit Chakraborty (HIT-K) Spreading Phenomena April 3, 2017 56 / 62
57. Spread of a pandemic
Figure: The spread of a pandemic with an initial outbreak in Hong Kong.
c Manojit Chakraborty (HIT-K) Spreading Phenomena April 3, 2017 57 / 62
58. Mobility Network
A surprising but welcome aspect of epidemic forecast is that the predictions of dierent models
are rather similar, despite the fact that they use dierent mobility data.
The eective distance helps us understand why the various model predictions converge. We can
write the arrival time of a pathogen to location a as
Ta =
deff (P)
Veff (β, R0, γ, )
We see that the relative arrival times are independent of the epidemiological parameters. For
example, for an outbreak that starts at node i, the ratio of the arrival times to nodes j and l is
Ta(j/i)
Ta(l/i)
=
deff (j/i)
deff (l/i)
i.e. the ratio depends only on the eective distances. Therefore, the relative arrival times of the
disease depend only on the topology of the mobility network.
c Manojit Chakraborty (HIT-K) Spreading Phenomena April 3, 2017 58 / 62
59. Effective Distance and Arrival Time
1 Geographic Distance
Arrival times vs. geographic distance from
its source (Mexico) for the 2009 H1N1
pandemic.
2 Eective Distance
Epidemic arrival time Ta vs. eective
distance Deff for H1N1, demonstrating
the strong correlations between the
eective distance
c Manojit Chakraborty (HIT-K) Spreading Phenomena April 3, 2017 59 / 62
60. Summary
The joint advances in data collection and network epidemics have oered the capability to
predict the real-time spread of a pathogen. The developed models can help design
response and mitigation scenarios.
Interestingly, the recent success of epidemic forecast is not due to the improved
understanding of the underlying biology of infectious pathogens.
When it comes to the spreading of a pathogen, the epidemic parameters are of secondary
importance. The most important factor is the structure of the mobility network.
c Manojit Chakraborty (HIT-K) Spreading Phenomena April 3, 2017 60 / 62
61. References
[1] P. Erd®s and A. Rényi. On random graphs, I. Publicationes Mathematicae (Debrecen),
6:290-297 (1959).
[2] A.L. Barabási and R.Albert. Emergence of scaling in random networks. Science,
286:509-512 (1999).
[3] R. Pastor-Satorras and A. Vespignani. Epidemic spreading in scalefree networks. Physical
Review Letters, 86:3200-3203 (2001).
[4] D. Brockmann and D. Helbing. The Hidden Geometry of Complex, Network-Driven
Contagion Phenomena. Science, 342:1337-1342 (2014).
Picture Courtesy : A.L. Barabási. Spreading phenomena. Network Science, 1:379-436 (2016)
c Manojit Chakraborty (HIT-K) Spreading Phenomena April 3, 2017 61 / 62
62. Thank you
c Manojit Chakraborty (HIT-K) Spreading Phenomena April 3, 2017 62 / 62