Upload
others
View
0
Download
0
Embed Size (px)
Citation preview
Fuzzy reinforcement learning (RL) Autonomous Scent DISPENSER: for
creating memorable customer experience of Long-Tail connected Venues 1Chandrasekar Vuppalapati,
2Anitha Ilapakurti,
3Jayashankar Vuppalapati,
4Santosh Kedari
1Hanumayamma Innovations and Technologies, Inc.
Email: [email protected] 2Hanumayamma Innovations and Technologies, Inc.
Email: [email protected] 3Hanumayamma Innovations and Technologies Private Limited.
Email: [email protected] 4Hanumayamma Innovations and Technologies Private Limited.
Email: [email protected]
Abstract In today’s competitive business environment, creating memorable experiences and emotional
connections with consumers is critical to win customer spending and long term brand loyalty. Brands
want their customers to be in pleasing, subliminal scented environments. Even a few micro particles
of scent can do a lot of marketing’s heavy lifting, from improving consumer perceptions of quality to
increasing the number of visits. Hence, high roller venues such as Trump Towers and Caesars Palace
of the World use digital connected intelligent scent disperser systems that deliver seamless olfactory
experiences to connect with consumers on a deeper, emotional and personal level.
The challenges, however, for venues with limited digital connected infrastructure and deficient
intelligent systems are lack of engagement with patrons at a personal and emotional level and thus
miss recurring business opportunities and sustained long-term brand loyalty.
This research paper addresses the challenge by developing fuzzy Reinforcement Learning
autonomous intelligent scent dispensers that bring connected intelligence to non-connected venues.
The fuzzy reinforcement learning enables venue environment feedback loop that improves over
customer experience. The paper presents prototyping solution design, as well as its application and
certain experimental results.
Keywords: Fuzzy Logic; Reinforcement Learning; Machine Learning; Internet Of Things; SARSA;
Artificial Intelligent (AI) Systems; Long Tail; Sensors; iDispenser; Venue Analytics;
1.Introduction
In today’s competitive business environment, creating memorable experiences and
emotional connections with consumers is critical to win consumer spending and long-term brand
loyalty [1]. In order to influence consumers, marketers have used several in-store venue stimuli:
background music, venue lighting and ambient scent. The
ambient scents have direct influence on improving the
mood and increase various types of approach behavior of
consumers [2] and thus have become dominant Brand
promoting actor. With Ambient scent, brands want their
customers to be in pleasing subliminal scented
environments even a few micro particles of scent can do a
lot of marketing’s heavy lifting, from improving consumer
perceptions of quality to increasing the number of visits (Figure 1).
Hence, high roller venues such as Trump Towers, New York, and Caesars Palace, Las Vegas, of the
World use digitalconnected intelligent scent disperser systems that delivers seamless olfactory [4]
experience to connect with consumers on a deeper emotional and personal level – using data to create an unforgettable customer experience and five star Customer Relationship Management
(CRM). [5]
Figure 1: Scents work via physical environments
resulting in emotion [3]
International Journal of Pure and Applied MathematicsVolume 119 No. 12 2018, 12999-13009ISSN: 1314-3395 (on-line version)url: http://www.ijpam.euSpecial Issue ijpam.eu
12999
1.1. The Long Tail1: issue at stake
Internet has changed the dynamics of the venue and hospitality market place [6]. Since the
development of the internet, all the attributes of the hospitality & venue industry has deeply
and continuously evolved: usage of connected networks, usage of consumer digital data &
mining of consumer preferences, exclusive recommendations gleaned from Web 2.0 & 3.0
data sources, Internet Of Things (IoT), wearable data sources and new business models
reshape the industry. This process has even
been sped up recently by the emergence of
disruptive innovations, from the evolutions of
web & mobile technologies to the Artificial
Intelligence (AI) platforms. The long tail is
used in this paper as a multitude of a small,
independent venue operators (Figure 2) with
limited digital connected infrastructure and
deficient intelligent systems can grasp and
synthesize this complexity.
This research paper addresses the challenge by developing fuzzy logic reinforcement-learning
autonomous intelligent scent dispenser that brings connected intelligence2 to non-connected and
limited compute processing venues. The paper presents prototyping solution design, as well as its
application and certain experimental results.
The structure of this paper is as follows. Section 2 discusses the basic concepts and methods about
Fuzzy Logic, Reinforcement Learning, Model Based, Model Free, Q Learning, Bluetooth Low Energy
Framework, Sensors and Fuzzy Control Process and Section 3 shows a case study. Conclusions and
future work are summarized in Section 4.
2. Understanding Fuzzy Logic & Reinforcement Learning for Intelligent Dispenser
2.1. nRF8001 Bluetooth Low Energy (BLE)
BLE is a low powered wireless personal area network that transmits small amount of
data over shorter distance as opposed to the class Bluetooth that consumes more power to transmit
large data over longer distance. The nRF8001 uses v4.0 BLE radio and features simple serial interface
that supports wide range of external application microcontroller. [7].
We have used Bluetooth not only for connectivity purposes but
also to identify number of people ―scanForPeripherals ― in a
venue. BLE scan provides a good view of foot traffic. For small
and medium venues, using Bluetooth to identify number of
people in a venue saves huge infrastructure cost (setting either
video cameras or infra red sensors for entry & exit strip
sensors).
2.2. Congestion and Number of Bluetooth Devices
There is linear relationship between number of people and number of Bluetooth devices in
avenue [8].Use count of Bluetooth Devices to Know the Number of People on New Year's Eve [9].
1 Long Tail in Tourism Industry - https://halshs.archives-ouvertes.fr/halshs-01248302/document
2 How sensory marketing applies to the hotel and restaurant industry in order to influence customer’s behavior in Thailand -
http://www.diva-portal.org/smash/get/diva2:426249/fulltext01
Figure 2: Long Tail
Figure 3: Count Number of People in New
Year's Eve [9]
International Journal of Pure and Applied Mathematics Special Issue
13000
Bluetooth Central manager call for ―scanForDevices‖ is used to count number of devices in a room.
This popular method is being used to count number of people in New Year’s Eve (Figure 3).
2.3. Passive Infra Red Sensor (PIR)
PIR based motion detector is used to sense the
movement humans, animals and other living beings. For
instance, the Texas Instrument (TI) design TIDA –
010693 uses two lenses with different areas of detection
for each PIR Sensor: One lens for middle detection area
for a tall subject (like a human) and another for lower
detection for short subject (like a dog) (Figure 4).
The Crossing point distance (i.e., the certain distance the
beam of the middle and lower bean intersect)
X = 𝛿
tan 0.18 −tan 0.15
(1)
For TI Sensors𝛿𝑖𝑠𝑠𝑒𝑡𝑡𝑜10 cm. Therefore X = 2.96m;In Intelligent Dispenser case the PIR motion
index is a decimal number.
2.4. Microphone
Microphone data used to detect any human presence in a
venue. We have used Micro Electromechanical
Systems (MEMS) based A-weighted filter signals (Figure 5)
as these signals correspond to the human ear frequency
response4
2.5. Fuzzy Logic (FL)
Fuzzy Logic (FL) incorporates a simple rule based If X and Y Then Z approach to solving a control
problem than attempting to build a mathematical model. The FL
model is an empirically based relying on an operator's experience
rather than their technical knowhow of the system. For building
intelligent dispenser model, we have modeled Venue scent diffusion
system using predefined rules governing the behavior of venue
activity and developed activity indicator. Hence, Fuzzy Logic is the
most appropriate ML model [13].
Activity Indicator (Figure 6) is a combination variable that is
composite of # of people in a venue (based on the number of
people to BLE devices present in the venue) [8], the rate of
change of people, motion sensor value, the change or delta of motion sensor value [10] and Noise
Threshold Value, the delta of noise threshold [11]. Activity Indicator ranges from negative value to
positive value (denoting subsiding to active increase of human activity). In other words, negative
activity indicator indicates the number of people in the venue is in descending phase. The positive
value indicates the number of people in the venue is increased (more foot traffic).
3 TI PIR Sensor - http://www.ti.com/lit/ug/tiducv3b/tiducv3b.pdf 4 AN4426 Application Note:
http://www.st.com/content/ccc/resource/technical/document/application_note/46/0b/3e/74/cf/fb/4b/13/DM00103199.pdf/files/DM00103199.pdf/jcr:content/translations/en.DM00103199.pdf
Figure 4: TI PIR Sensor [10] - Fresnel Lenses View
Figure 5: A-weighted Filter Response [11]
Figure 6: Activity Algorithm
International Journal of Pure and Applied Mathematics Special Issue
13001
Activity Indicator (t) =
𝑁𝑢𝑚𝑏𝑒𝑟 𝑜𝑓 𝑃𝑒𝑜𝑝𝑙𝑒 𝑖𝑛 𝑡𝑒 𝑉𝑒𝑛𝑢𝑒 +
𝑚
𝑡=1
Δ 𝐶𝑎𝑛𝑔𝑒 𝑖𝑛 𝑁𝑢𝑚𝑏𝑒𝑟 𝑜𝑓 𝑃𝑒𝑜𝑝𝑙𝑒 𝑖𝑛 𝑡𝑒 𝑉𝑒𝑛𝑢𝑒
Δ𝑡
+ max𝑡
𝑀𝑜𝑡𝑖𝑜𝑛 𝑆𝑒𝑛𝑠𝑜𝑟 𝑉𝑎𝑙𝑢𝑒 + Δ 𝑀𝑜𝑡𝑖𝑜𝑛 𝑆𝑒𝑛𝑠𝑜𝑟 𝑇𝑟𝑒𝑠𝑜𝑙𝑑
Δ𝑡
+ max𝑡
(𝑁𝑜𝑖𝑠𝑒 𝐿𝑒𝑣𝑒𝑙) + Δ 𝑁𝑜𝑖𝑠𝑒 𝐿𝑒𝑣𝑒𝑙 𝑇𝑟𝑒𝑠𝑜𝑙𝑑
Δ𝑡
(2)
2.5.1. Linguistic Variables
Fuzzy or Linguistic variables are the central to FL and were introduced by Professor LotfiZadeh in
1973. Succinctly put, Linguistic variables are language artifacts not numbers.
The control inputs, namely hardware sensors, in intelligent dispenser model are Fuzzy variables. For
instance, ―BLE Devices‖, ―Noise Level‖, ―Proximity Indicator – received signal strength indicator
(RSSI) Strength‖, ―Microphone5‖ and ―Passive Infra Red (PIR) Sensor
6‖ are modeled as Fuzzy
parameters. In FL error is just the difference, it can be thought of the same way as Fuzzy Variable
[12].
Another salient principle of FL is fuzzy variables (Table 1) are themselves adjectives that modify the
variable (e.g., "large positive" error, "small positive" error "zero" error, ―small negative" error, and
"large negative" error). As a minimum, one could simply have "positive", "zero", and "negative"
variables for each of the parameters. Additional ranges such as "very large burst of devices" and "very
small noise thresholds" are added to handle the BURST response or very nonlinear behavior.
Table 1: Fuzzy Variables
―N‖ = Negative Error or error-dot level
―Z‖ = Zero Error or error-dot level
―P‖ = Positive Error or error-dot level
―D‖ = ―Dispense‖ Output response
―-― = ―No Change‖ or Current Output
―ND‖ = No Dispense
2.5.2. Control Process
The goal of the control process is to model/control dispense
of Scent through the FL by detecting people activity in a
venue (as shown in the Figure 7): number of people in a
room (number of BLE devices), Motion Sensor detection and
Noise detected on the microphones.
2.5.2.1. Definitions:
INPUT# 1 System Status
- Error = Command-Feedback (Activity Indicator Eq.2)
- P = Increased Activity Index (Increase of venue occupancy or Noise or
Motion thresholds),
5Microphone - https://www.embedded.com/electronics-blogs/max-unleashed-and-unfettered/4441291/New-technology-provides-breakthrough-speech-recognition-in-noisy-
environments 6 PIR sensor - https://learn.adafruit.com/pir-passive-infrared-proximity-motion-sensor/how-pirs-work
Figure 7: Activity Indicator Control Process
International Journal of Pure and Applied Mathematics Special Issue
13002
- Z = Muted or no Activity,
- N = Decreased Activity Index (Decrease of venue occupancy or Noise or
Motion thresholds)
INPUT# 2 System Status Error-dot = d (Error)/dt
- P = Getting Increased Activity Index
- Z = Not changing
- N = getting decreased activity index
OUTPUT Conclusion and System Response
Output:
- DISP = Dispense Scent
- = don't change
- NDISP = No Dispense
2.5.3. Matrix Rules
The fuzzy parameters of error (# BLE Devices with RSSI, PIR Sensor threshold and Microphone
Noise Levels) and error-dot (rate-of-change-of-error) were modified by the adjectives "negative",
"zero", and "positive" (Table 2).
In the Figure 9, the linguistic parameter ―BLE Devices‖ indicates number of devices within a venue
(negative value indicates people leaving the proximity of the iDispenser & positive implies entering
into proximity of the iDispenser).
Table 2: Rules
1. IF CMD_AI = N AND d(CMD_AI)/dt = N THEN OUTPUT = NDISP
2. IF CMD_AI = Z AND d(CMD_AI)/dt = N THEN OUTPUT = NDISP
3. IF CMD_AI = P AND d(CMD_AI)/dt = N THEN OUTPUT = DISP
4. IF CMD_AI = N AND d(CMD_AI)/dt = Z THEN OUTPUT = NDISP
5. IF CMD_AI = Z AND d(CMD_AI)/dt = Z THEN OUTPUT = NC
6. IF CMD_AI = P AND d(CMD_AI)/dt = Z THEN OUTPUT = DISP
7. IF CMD_AI = N AND d(CMD_AI)/dt = P THEN OUTPUT = DISP
8. IF CMD_AI = Z AND d(CMD_AI)/dt = P THEN OUTPUT = DISP
9. IF CMD_AI = P AND d(CMD_AI)/dt = P THEN OUTPUT = DISP
2.5.4. Matrix
The fuzzy parameters of error (Activity
Indicator) and error-dot (rate-of-change-of-
activity_Indicator) were modified by the
adjectives "negative", "zero", and
"positive". The following diagram provides
a 3-by-3 matrix (Figure 8) input and output
outcomes. The columns represent "negative
error", "zero error", and "positive error"
inputs from left to right. Please note:
―DISP‖ results into dispensing of the scent. NDISP – no Dispensing required and NC implies no
Change to the condition.
2.5.5. Scent Diffusion Membership Function
Figure 8: Matrix
International Journal of Pure and Applied Mathematics Special Issue
13003
The membership function (Figure 9) is a graphical representation of the magnitude of participation of
each control input (#BLE Devices, Microphone
Thresholds and PIR Sensor Thresholds). It associates a
weighting with # BLE Devices, Microphone and PIR
Thresholds inputs, define functional overlap between
inputs (BLE Devices to Noise Level on Microphone)
(BLE Devices to PIR Thresholds) and (Microphone
Noise Levels to PIR Thresholds), and ultimately
determines an output response of dispenser (Scent
diffusion or not).
2.6. Reinforcement Learning (RL) [14]
RL [14] is learning process of an agent to act in
order to maximize its rewards. In RL learning is
by trial-and-error to map situations to actions
that maximize a numerical reward. The
following figure 10 has standard RL
architecture.
The agent is defined as a scent-dispensing algorithm, the action is the command to dispense or not
dispense, the environment is the target object, in this case activity indicator that is formed in a closed
venue, and the reward is performance improvement after applying the action. The goal of RL is to
choose an action in response to a current stat, based on activity indicator, to maximize the reward,
dispense scent or not. Generally, RL approaches learn estimates of the initialized Q Values Q(s,a),
which maps all system states s to their best action a. We initialize all Q(s,a) and during learning,
choose an action a for state s based on ∈ greedy policy to target platform. Then, we observe the new
state s’ and reward r and update Q-value of the last state-action pair Q(s,a), with respect to the
observed outcome state (s’) and reward r. However, in our case, we have used State-Action-Reward-
State-Action (SARSA).
2.6.1. Model Free vs. Model based
The Model stands for the simulation of the dynamics of the environment (dispense based on activity
indicator Eq. 2). That is, the model learns the transition probabilityT (𝑠1
(𝑠0,𝑎)) from the pair of current
state s0 and action a to the new state s1 (figure
11) [15]. To derive an action, an agent uses a
policy that aims to increase the future rewards
(indefinitely or long run environments). If the
transition probability successfully learned, the
agent will know how likely to enter a specific
state given current state and action. The model-
based approach becomes impractical if state
space and action space grows (S * S * A).
Trial-and-error process enables model-free systems to update knowledge. Model free-based systems
are more practical as the interactions of environment and agent are unknown in many physical
systems [16].
Figure 9: Member Functions [12]
Figure 10:The standard architecture of RL Algorithm
Figure 11: on-policy vs. off-policy agent
International Journal of Pure and Applied Mathematics Special Issue
13004
2.6.2. On-policy and Off-policy
An on-policy agent, our current dispenser in case, learns the value based on an action derived from the
current policy and off-policy agent learns based on the action a* obtained from another policy (a
greedy policy in case of Q-learning).
2.6.3. Fuzzy SARSA Learning (FSL)
By using RL instead of using static matrix rules (Table 2) the performance of target environment can
be captured after applying each state decision. Since, the effectiveness of scenting can be fed back
from Venue managers by providing the input via mobile app. The combination of Fuzzy Logic
controller with SARSA is explained below:
1) Initialize the q-values: Each member of q-value
table assigned to a certain rule that describes some state-
action pairs and is updated during the learning
process. Please see table 2 and Figure 10. It can tell
us the performance of taking the action by taking
into account the reward value.
2) Select an action: to learn from the
system, we use Table 3 to gain the knowledge of the
system. This is called exploration/exploitation
strategy [14],∈ greedy is standard exploration policy.
3) Calculate the control action inferred by fuzzy logic controller: the fuzzy output is a
weighted average of the consequences of the rules:
𝑎 = 𝜇𝑖 𝑥 𝑥 𝑎𝑖𝑁
𝑖−1
Where N is number of rules (Scent dispenser N = 12),
𝜇𝑖 𝑥 𝑖𝑠 𝑡𝑒 𝑓𝑖𝑟𝑖𝑛𝑔 𝑑𝑒𝑔𝑟𝑒𝑒 𝑜𝑓 𝑡𝑒 𝑟𝑢𝑙𝑒 𝑓𝑜𝑟 𝑡𝑒 𝑖𝑛𝑝𝑢𝑡
𝑠𝑖𝑔𝑛𝑎𝑙 𝑥 𝑎𝑛𝑑 𝑎𝑖 𝑖𝑠 𝑡𝑒 𝑐𝑜𝑛𝑠𝑒𝑞𝑢𝑒𝑛𝑡 𝑓𝑢𝑛𝑐𝑡𝑖𝑜𝑛 𝑓𝑜𝑟 𝑡𝑒
𝑠𝑎𝑚𝑒 𝑟𝑢𝑙𝑒
4) Approximate the Q-function from the current q-values and the firing level of the
rules:
Q (s,a) = ( 𝑁𝑖=1 𝜇𝑖 𝑠 𝑥 𝑞[𝑖, 𝑎𝑖])
The action-value function Q (s,a) tells us how desirable it is to reach state s by taking action a by
allowing to take the action a many times and observe the return value.
5) Calculate the reward: the reward is calculated on following criteria: a) proper
dispensing of the scent.
6) Calculate the value of the new state s’:
Table 3: q-values table
International Journal of Pure and Applied Mathematics Special Issue
13005
V (s’) = 𝑢𝑖 𝑠′ . max(𝑞 𝑖, 𝑎𝑘 )𝑁𝑖=1 where max(q[I,ak]) is the maximum of the q-values
applicable in the state s’.
3.Case Study
We have tested and deployed Fuzzy Logic (FL)
based
iDispenser(http://www.hanuinnotech.com/venueanalytics.html)
We have several Venue Users that are using the system and providing valuable venue and usage data
& improved CRM (Figure 12).
4. Conclusion
This paper presented a novel approach to creatingFuzzy Logic based reinforcement learning
intelligent scent dispenser that bridges the chasm between limited connected network and availability
of compute power to perform intelligent venue operations. As witnessed through our market reach, we
strongly conclude that there are still parts of the world that are not touched and blessed through the
data insights based intelligent devices – chiefly not have a connected network. With this research
paper and with the development of the device, we brought connected intelligence to non-connected
venues. We strongly believe that by implementing Reinforcement Learning FL based dispensing has a
huge positive impact in healthcare, travel, entertainment and other captive services business verticals.
Acknowledgements
We sincerely thank you the engineering teams, the marketing teams and the product management of
Hanumayamma Innovations and Technologies, Inc., and Hanumayamma Innovations and
Technologies Private Limited for their valuable contribution to product deployments in India and
USA and their active involvement in collecting customer data
References
Journals, Books and Conferences
[1] Martín-Ruiz, David, Carmen Barroso-Castro, and Isabel Mª Rosa-Díaz. "Creating customer value
through service experiences: An empirical study in the hotel industry." Tourism and Hospitality
Management 18, no. 1 (2012): 37-53.
[2] HuaMeng, ―THE EFFECTS OF AMBIENT SCENT ON CONSUMER BEHAVIOR: A REVIEW OF
THE LITERATURE‖, Aug. 2016, Accessed on: Feb 25, 2018,
https://etd.ohiolink.edu/!etd.send_file?accession=kent1468275876&disposition=inline
[3] Constant Berkhout, Retail Marketing Strategy: Delivering Shopper Delight, Publisher: Kogan Page
(November 28, 2015), ISBN-13: 978-0749476915
Figure 12: iDispenser Case Study
International Journal of Pure and Applied Mathematics Special Issue
13006
[4] Making Sense of Scents: Smell and the Brain, Society for Neuroscience,
http://www.brainfacts.org/Sensing-Thinking-Behaving/Senses-and-Perception/Articles/2015/Making-
Sense-of-Scents-Smell-and-the-Brain, Creation Date: 27 Jan 2015 | Review Date: 27 Jan 2015
[5] David Norton, The High Roller Experience: How Caesars and Other World-Class Companies Are
Using Data to Create an Unforgettable Customer Experience, Publisher: McGraw-Hill Education; 1
edition (November 7, 2017), ISBN-13: 978-1259862953
[6] Christian Longhi and Sylvie Rochhia, Long Tails in the Tourism Industry: Towards Knowledge
Intensive Service Suppliers, https://halshs.archives-ouvertes.fr/halshs-01248302/document Access
Date: February 25, 2018
[7] Ilapakurti, J. S. Vuppalapati, S. Kedari, S. Kedari, C. Chauhan and C. Vuppalapati, "iDispenser — Big
Data Enabled Intelligent Dispenser," 2017 IEEE Third International Conference on Big Data
Computing Service and Applications (BigDataService), San Francisco, CA, 2017, pp. 124-130. doi:
10.1109/BigDataService.2017.53
[8] Streitz, Norbert, Markopoulos, Panos, Distributed, Ambient and Pervasive Interactions, Publisher:
Springer International Publishing, Softcover ISBN: 978-3-319-58696-0
[9] Cooking Hacks, Counting Bluetooth Devices to Know the Number of People on New Year's Eve,
https://www.cooking-hacks.com/blog/counting-bluetooth-devices-to-know-the-number-of-people-on-
new-years-eve/, Access date: February 28, 2018
[10] Texas Instrument, Advanced Motion Detector Using PIR Sensor Reference for False Trigger
Avoidance, Source: http://www.ti.com/lit/ug/tiducv3b/tiducv3b.pdf, Access date: February 28, 2018
[11] STMicroelectronics, AN4426 Application Note,
http://www.st.com/content/ccc/resource/technical/document/application_note/46/0b/3e/74/cf/fb/4b/13/
DM00103199.pdf/files/DM00103199.pdf/jcr:content/translations/en.DM00103199.pdf , Access date:
February 28, 2018
[12] Rajesh Kumar, Fundamental of Artificial Neural Network and Fuzzy Logic, Publisher: Laxmi
Publications (January 1, 2010) ISBN-13: 978-8131807101
[13] AnupamShukla, RituTiwari, Rahul Kala, Real Life Applications of Soft Computing, Publisher: CRC
Press; 1 edition (May 21, 2010), ISBN-13: 978-1439822876
[14] R.S. Sutton and A.G. Barto, ―Reinforcement Learning: An Introduction (Adaptive Computation and
Machine Learning)‖, A Bradford Book; 1st Edition edition (March 1, 1998), ISBN-13: 978-
0262193986
[15] Steeve Huang, ―Introduction to Various Reinforcement Learning Algorithms. Part I (Q-Learning,
SARSA, DQN, DDPG)‖, Source: https://towardsdatascience.com/introduction-to-various-
reinforcement-learning-algorithms-i-q-learning-sarsa-dqn-ddpg-72a5e0cb6287 Access Date: March 25,
2018
[16] P. Jamshidi, A. M. Sharifloo, C. Pahl, A. Metzger and G. Estrada, "Self-Learning Cloud Controllers:
Fuzzy Q-Learning for Knowledge Evolution," 2015 International Conference on Cloud and Autonomic
Computing, Boston, MA, 2015, pp. 208-211.doi: 10.1109/ICCAC.2015.35,
URL: http://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=7312157&isnumber=7312127
Authors Biography
Chandrasekar Vuppalapati: Chandra is a seasoned Software IT Executive with diverse experience in
Software Technologies, Enterprise Software Architectures, Cloud Computing, Big Data
Business Analytics, Internet Of Things (IoT), and Software Product & Program Management.
Chandra held engineering and Product leadership roles at GE Healthcare, Cisco Systems, St.
Jude Medical, and Lucent Technologies, a Bell Laboratories Company. Chandra teaches
Software Engineering, Mobile Computing, Cloud Technologies, and Web & Data Mining for
Masters program in San Jose State University.
Anitha Ilapakurtihas wide variety of Software development project management experience in HealthCare,
Banking, Data Modeling, and Real Time Application systems. She has pioneered in software
development processes for the Business Agile systems and interacted with Fortune 500
business leaders in communicating and sharing her visions and product development.From her
International Journal of Pure and Applied Mathematics Special Issue
13007
humble beginning as a programmer and analyst at a software development center in the Silicon Valley, Anitha
has handled several development and project management roles in the software application development life
cycles. Her most notable are: Application Architect and Business Manager before becoming a software
entrepreneur. She holds a B.SC, M.SC in Computer Sciences from Nagrjuna University, India, and Master of
Sciences in Software Engineering from San Jose State University, USA.
Santosh Kedarileads Indian operation in the capacity of Director of Software Services. Santosh co-founded
Sanjeevani Electronic Health Records (EHR) and passionate about democratizing EHR for
greater good. He conducts health camps in Southern India and plays a pivotal role in Sales
operation of Dairy Analytics. Apart from work, Santosh enjoys movie watching and wants to
produce healthcare documentaries and knowledge sharing small movies.
Jayashakar Vuppalapatiis a seasoned Executive with diverse experience in Software Technologies, Enterprise
Software Architectures, Big Data Business Analytics, Internet Of Things (IoT), Engineering
Management and new and innovative product designs. Jay held engineering and Product
leadership roles at FedEx, Anthem, Intel, HCL technologies, Health Net and other major
corporations. Currently, Jay holds CTO role and leads South Asian business markets
International Journal of Pure and Applied Mathematics Special Issue
13008
13009
13010