A collaboration between Ford University Research Program and University of Minnesota University PI Shashi Shekhar Ford PI Shounak Athavale Eric Marsman Outline Proposal Summary Tasks and Progress ID: 593496
Download Presentation The PPT/PDF document "Ephemeral Network Broker to Facilitate F..." is the property of its rightful owner. Permission is granted to download and print the materials on this web site for personal, non-commercial use only, and to display it on your personal computer provided you do not modify the materials and that you retain all copyright notices contained in the materials. By downloading content from our website, you accept the terms of this agreement.
Slide1
Ephemeral Network Broker to Facilitate Future Mobility Models/Transactions
A collaboration between Ford University Research Program and University of Minnesota
University PI: Shashi Shekhar
Ford PI:
Shounak Athavale
Eric MarsmanSlide2
Outline
Proposal Summary, Tasks and ProgressNext stepsJournal Paper UpdatesSlide3
Proposal Summary
Ephemeral Networks: Groups of people, good, and services that encounter each other in the physical world or are in close geographic proximity during routine activities (e.g. commute, shopping, entertainment)
Objective: To
build an ephemeral network broker
to connect
people, goods and services
based on
mobility profiles
(e.g. GPS trajectories),
geographic proximity, and known intents
(e.g. calendars, wish lists, gift registries, shopping lists).
The broker will assist in identifying
real-time and near future
mobile commerce
transaction opportunities
in ephemeral networks
(MCEN) from inception to conclusion in support of new Future Mobility Business models.Slide4
Proposal Summary: Related Work and Limitations
Related work:Social Network AnalysisEphemeral social networks (temporal proximity e.g. at a conference)
Limitations of Related Work:Do not model socio-economic MCEN semantics (e.g. need, desire, readiness for transaction, trust)
Does not provide quantitative measures to distinguish most promising MCEN opportunities (e.g. recurring commute, travel, other trips) and other ESNs (e.g. rare meeting of academic in research conferences).
Do not scale up to megacities.Slide5
Proposal Summary: Research Tasks
Conceptual Modeling:Leverage novel commerce opportunities facilitated by ephemeral networks and routine activities generating trips to define core concepts and taxonomies to facilitate the design of interest measures and scalable algorithms for identifying MCEN opportunities.
Interest Measures:
Develop quantitative interest measures to separate commercially promising MCEN opportunities from other ephemeral social networks.
Scalable Algorithms:
Design scalable techniques for MCEN opportunities leveraging measure from Task 2
Validation:
Do proposed algorithms scale-up to megacities?
How accurately can we predict MCEN opportunities in real-time or near future?
Synthetic datasets can be generated using city simulator or traffic simulator. Real datasets may come from FORD.Slide6
What has been done so far?
Conceptual Modeling:
Assumed that mobile consumers explicitly specify their service type needs and readiness based on spatial proximity In future: May incorporate ratings to model trust (more interesting with mobile providers)
Learning of user preferences from historical trajectories
Interest Measures:
Focused on maximizing
unexpected real-time demand
under different supply-demand ratio scenarios
In future:
Formally modeling providers fairness
Develop measures for a transaction’s promise using historical transactions and routines to
predict
near-future commerce opportunitiesSlide7
What has been done so far?
Scalable Algorithms:
Designed a scalable greedy algorithm with novel provider-centric heuristics In future: Design algorithms with better matching size and load balancing
Design scalable algorithms for other interest measures (e.g. predicting future opportunities from historical trajectories)
Validation:
Designed a city simulated for fixed supply and variable demand proportional to population
Considered generating whole trajectories for incoming requests
Evaluated greedy approach for several supply-demand rations in Minneapolis for 120,00 requests and 120 restaurants
In future:
Validate algorithms for predicting promising transactions
Validate proposed algorithms using real-datasets and larger synthetic datasetsSlide8
Next Steps: Short Term Plan
Short term:Formally modeling provider fairness/load balancing
(Journal Extension)Propose method for improve matching size and provider balance over greedy algorithm (Journal Extension)
Modeling mobile service providers
(SSTD 2017 -Mar12
th
, 2
nd
Intl. Conf. on Smart Data and Smart Cities - Mar 31
st
)
With rendezvous points
May also model group matching ( grouping requests for optimizing provider route)Slide9
Next Steps: Long Term Plan
Long term: Design interest measures and scalable algorithms:
for predicting near-future opportunities from historical trajectories and current intents based on routine activities (pattern mining for online prediction)
Validation:
Incorporate GPS trajectory generation into City simulator (+ real datasets?)
Enriching the model:
Incorporating ratings to model trust
Learning of user preferences from historical trajectoriesSlide10
Journal Paper Updates:
Formally Defining a Provider Fairness objective Function
Assuming matching size is maximized:Possible Alternatives:Minimize max. utilization of all providers: (communication networks)
Min
u
max
where
u
max
=
Minimize total utilization cost of all providers: (traffic engineering)
Min
Maximize Shannon’s diversity index (entropy measure):
Max
Slide11
Journal Paper Updates:
Optimization Problem FormulationWe propose a mixed-integer programming formulation to optimize the matching size for the set of requests available at time t
Methods to combine both objective functions: maximizing matched consumers and maximizing provider fairness:
Maximize matching size (most favorable) where provider fairness is added as a constraint
pros:
Easiest to implement
cons:
difficult to set fairness threshold
Aggregate functions using a weighted linear sum of objectives
Pros:
can set smaller weight for provider fairness
cons:
difficult to find most appropriate weights
Optimize for matching size first (most favorable), then find max fairness function that achieves the max matching.
Slide12
Problem Definition: On-Demand Spatial Service Propositions
Input:A set P of service providers
A set R of consumer requests arriving dynamicallyA number of required propositions K
Output:
K service provider(s) propositions for each request
Objective:
Maximize number of matched requests
Secondary Objective:
Maximize provider-balance
Constraints:
Each returned proposition satisfies the consumer’s max. travel time and waiting time constraints and does not violate the provider’s service rate.Slide13
Mixed Integer Programming Formulation (1/2)
Decision Variables:
1
Notations
Variable Description
N
# consumers
M
# providers
K
Number of require propositions
per consumer
a
i
Arrival time of
C
i
d
i
Max travel time constraint of
C
i
w
i
Max waiting time constraint
of
C
i
d
ij
Shortest travel time between locations of
C
i
and
P
jcjService rate (capacity) per hour for PjejEarliest available service time at PjSlide14
Mixed Integer Programming Formulation (2/2)
Objective Function:Constraints:
(1)(2)(3)
(4)
(5)
(6)
Limitations:
Integer programming may not scale for a large number of consumers and providers
O(
MxN
) variables and constraints
However, solution can also serve as an upper bound for max # matches at time t
Maximizes matching and balance based on current requests, but does not adapt to requests that may appear in the future
1Slide15
Reducing Problem Size using Spatial Partitioning
The complexity of the optimization problem can be reduced by dividing it into smaller matching sub-problems:
Using a min-cut graph partitioning approach e.g. O(E), O(VE2)Continue until problem size is below a threshold
Consumer
Provider
Possible MatchSlide16
Greedy Approach:
Novel Supply-Demand Ratio Aware HeuristicsLeast Accepted First (LAF) and Least Appearance as Candidate First (LCF) heuristics only considered total number of previous matches/occurrences of a provider, but do not its service capacity, resulting in the following limitations:
Do not balance utilizationCurrent decisions may not maximize future matching since
Providers with low demand are favored even if they have near full utilization (i.e. low capacity providers)
Does not consider temporal heterogeneity of demand (e.g. changes in day/night population) and thus providers with historically low demand will be favored even during periods of their high demand
Propose New heuristics:
Prioritize least utilized providers
(i.e. least total utilization)
to balance providers utilization
Prioritize providers with highest recent supply-demand ratio over
(i.e. least recent utilization)
using a moving time horizon to leave capacity for future requests
Note: Replacing the
“exactly K propositions”
constraint with “
Up to K propositions
” also requires
non-straightforward
handling of the LAF and LCF heuristicsSlide17
“Up to K” Formulation: An Iterative Max-Flow Approach
Maximum Matching in bi-partite graphs can be solved by finding the max flow in a flow network
However, we cannot guarantee that each consumer is matched to “
exactly K”
providers
For up to K propositions, we may apply an iterative max flow approach (iterations ≤ K). In each iteration:
Add consumer nodes assigned to fewer than K providers
Add provider nodes with remaining service capacity
Connect feasible matches which are not already selected in previous iterations
S
D
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1Slide18
IWCTS 2016
Journal Extension
Problem Definition
Objective:
max. matched requests
Consumers
had 2
constraints
:
Max. travel distance
Max waiting time
Proof of NP-hardness
Add a “Basic Concepts” subsection
Modify
consumer constraints
Formally define
a measure of provider fairness
Proposed Approach
Proposed 2 service-provider
centric heuristics:
Least Accepted First
Least appearance as Candidate First
Provide integer programming-based approach
Propose a spatial partitioning approach for enhancing the scalability
Propose a new provider capacity-aware/supply-demand ratio aware heuristic
(Up to K): Propose an iterative max-flow approach
Add example for each different heuristic
Experimental Evaluation
Four
experiments:
% matched requests
% matched providers
avg. requests/provider
Stdev
of requests/providerExpand experimental evaluation section by adding
more experiments
.
from original 10-page paper
tie-breaker heuristics
Discussion
None
Add discussion section (other related work, choice of simulation framework)Slide19
Thank you.Slide20
Basic Concepts
A Service Provider: A provider registered in the system is defined using its location and service rate per hour over the day (e.g. 15 requests per hour)
A Consumer Request: A request for from a mobile consumer, including the consumer’s current location, max. acceptable travel distance and max. acceptable waiting time before service.
A Service Provider Proposition
: A quadruple (r, p, d, w) where:
r
ϵ
set of available consumer requests
p
ϵ
set of registered service providers
d: distance between r and p
w: waiting time before r is served by p
Example:
(C
1
, P
1
, 2 miles, 5 min)Slide21
IWCTS Paper Contributions
Formally defined the problem of On-demand Spatial Service Propositions.
Proposed new category of service-provider centric heuristics for increasing the number of engaged service providers while meeting the conflicting requirements of the broker, consumers and service providers.
Our experimental results show that our proposed heuristics (Least Accepted First) result in:
a larger number of matched requests for balanced supply and demand scenarios
a larger number of matched service providers with a
more balanced provider assignment
particularly when the available supply exceeds the available demand.