In Drazen talk, you will get a chance to listen to how Data Science Master 4.0 on Belgrade University was created, and what are the benefits of the program.
9654467111 Call Girls In Munirka Hotel And Home Service
Data Science Master 4.0 on Belgrade University - Drazen Draskovic
1. Dražen Drašković, PhD, Ass. Prof.
Belgrade, November 2019
University of Belgrade – School of Electrical Engineering
Education for Digital Transformation
and a new wave of Data Science -
the challenge for Serbia
2. Introduction
• School of Electrical Engineering and DataSci courses
• New master program in Serbia (2019)
• R&D projects and cooperation with other
universities/schools
• Young professionals and a way to become
a pioneer of data science in Serbia
12/26/2019 Data Science Conference 5.0 2
3. Basic facts about our School - ETF
• In 1935 mechanical and electrical engineering
established the common department
at the Technical School of the University of Belgrade
• Founding of the School of Electrical Engineering
in 1948
• In 2019: 22 000 graduated engineers
12/26/2019 Data Science Conference 5.0 3
4. Bachelor and Master program @ETF
BSc academic studies
4 years / 8 semesters
2 study programs:
1) Electrical Eng. and Computing
Power engineering
Telecommunications and information
technologies
Electronics and digital systems
Micro and physical electronics
Signals and systems
Computer engineering
• 2) Software Engineering
• Titles:
– BSc in Electrical Eng. and Computing
– BSc in Software Engineering
MSc academic studies
1 year / 2 semesters
12 modules:
Audio and Video Technologies
Applied Mathematics
Biomedical Engineering
Electronics and digital systems
Energy Efficiency
Power Systems
Nanoelectronics and Photonics
System Engineering and Com.
Microwave Engineering
Signals and systems
Computer Engineering and Informatics
Software Engineering
12/26/2019 Data Science Conference 5.0 4
5. Data Science - new courses
6 years ago: ETF group for Data Science promotion
2 years ago, new curriculum reform
6 new courses (Data Mining, Natural Language
Processing, Introduction to Machine Learning, Data
Analysis in Social Networks, Big Data in SE, Data Analysis
and Visualization)
5 reformed courses (Introduction to AI / Intelligent
Systems / Image Processing / Speech Processing /
Advanced Neural Networks)
12/26/2019 Data Science Conference 5.0 5
6. Master for Digital Transformation 4.0
• Multidisciplinary master program - public call by
the Ministry of Education, Science and Tech. Dev.
• Synergy of two schools = ETF + Faculty Org. Sciences
• Master program "Advanced information technology
in digital transformation"
– 3 semesters
– 90 ECTS (30 per semester)
– from 2019/2020
– 35 students (15+5+15)
– Title: Master engineer in information technologies
• Support of the Digital Serbia Initiative
12/26/2019 Data Science Conference 5.0 6
7. Master 4.0 - Structure
• Four study modules:
– Data analysis
– Distributed computing
– Interactive computing
– Organization and management
• One-semester courses (1 course = 6 ECTS)
• Student has 10 courses in first 2 semesters
• Third semester = professional practice +
team startup project + MSc thesis
12/26/2019 Data Science Conference 5.0 7
8. Teaching staff and infrastructure
• More then 100 professors, TA and researchers
• Guest lecturers from the best world universities
• Industry lecturers
• Providers of practice - the most important domestic
companies
• Providers of support programs for the development of
technological entrepreneurial ventures:
Science Technology Park Belgrade, Innovation Center
ETF, R&D center FON, ICT hub, Impact hub, 30hills, etc.
12/26/2019 Data Science Conference 5.0 8
9. Where are we now?
• September 2019: the program accredited
• October 2019: Entrance exam/23 enrolled students
• November 2019: first courses started
• 70% students choose the "Data analysis" module
• Objectives - short / long term
– new disciplines in our higher education
– increase the number and quality of IT staff
– engineers competitive globally
– improving the entrepreneurial ecosystem
in Serbia
12/26/2019 Data Science Conference 5.0 9
10. R&D projects in DataSci field
• The use of Artificial Intelligence to automate
the Rapid Integrated Assessment mechanism and to
nationalize Sustainable Development Goals in Serbia (UNDP)
• Software tool for sectorization planning with traffic
prediction (SMATSA)
• Image analysis in the manufacture of chemical preparations
• Churn prediction in retail banking
• Recognize movie categories based on movie posters
and photos
• Recognizing perspective in car photos
• Application of convolutional neural networks to improve
photo quality
• Application of reinforcement learning in gaming industry
12/26/2019 Data Science Conference 5.0 10
11. Cooperation with other universities
• Temple University, USA
– Data Science lab. (Prof. Zoran Obradovic, PhD)
• Stanford University, USA
– Machine Learning lab. (Prof. Jure Leskovac, PhD)
• University of Ljubljana, Slovenia
– Lab for Data Technologies, Faculty of Computer and
Information Science (Prof. Marko Bajec, PhD)
• Universitat Politècnica de València, Spain
– Department of Computing Engineering, Data Science lab
(Prof. Ana Pont Sanjuan, PhD)
12/26/2019 Data Science Conference 5.0 11
12. Young professionals
• The best of our school - our students
• A way to become a pioneer of data science
in Serbia - Ms Ljubica Vujovic, MSc EE
12/26/2019 Data Science Conference 5.0 12
13. Pioneering Data Science at Etihad Airways
Ljubica Vujovic
Data Science Conference 2019, Belgrade, Serbia
14. Content
• Data Science Services Team
• Projects Overview
• Building Blocks of Data Science Project
• Local Joiners Estimation
• Let's Get Technical
• Ready To Depart
15. Belgrade Team
Etihad Data Science Team in Belgrade is a part of Enterprise Analytics Transformation.
Our mission is to develop automated machine learning solutions and data driven business rules which will
help in cost savings, provide new revenue streams and increase customer satisfaction.
Who are we?
• Five members
•BS, BEng, MEng, MSc, PhD
• Who am I?
16. Projects Overview
Our work is mainly divided between Operational and Commercial projects.
Commercial Projects
• Customer Lifecycle Value
• Timing and Cadence of
Email Campagne
• Types of Offers
• Content of Offers
Operational Projects
• Statistical Taxi-out Fuel Estimation
• Holding Fuel and Holdings
Estimation
• Local Joiners Passengers
Volume Estimation
• Catering Estimation
18. Local Joiners Estimation
• Forecast of locally joining passengers are based on business
logic and prone to inaccuracy. Use of predictive analytics
and automation to forecast is opportunity for improved stuff
productivity, reduced overtime hours and better guest's
experience.
• Check-in Data, Reservation Data, Flight Schedule Data.
• EDA insights.
• 6 machine learning models.
19. Let's Get
Technical
Number of expected passengers checking
in for a given 30-minute interval for
a future calendar month
XGBoost was used for training
Dataiku Platform is used
for automation and
scheduling
Two scenarios
Data set checks
Model perfomance
checks
23. Ready To Depart
• Airline industry can seize opportunities
both in commercial and operational sector
with growing data science talent in-house
• Entering in DS field from quantitive field
• Data Science project is not only about
training models. It is complex process
which
includes understanding business requirem
ents, generating actionable insights
and interpretable machine learning
models, clearly communicating results and
benefits and working closely with data
engineers with purpose of putting project
in production
Education for Digital Transformation and a new wave of Data Science - the challenge for Serbia
Division for data processing was formed within our school in 1967. We have been engaged in this field,for over 50 years. The Computer Engineering Chair later emerged from that division
in accordance with ACM/IEEE
All courses have the mandatory practical project, which has all requirements of a real industrial project.
Also, students can work in the latest programming languages and software tools for Data Scientist.
In February 2019, MINISTRY published the call for new multidisciplinary master program,
based on the collaboration of academia and IT companies.
The cooperation between 2 higher education institutions was the minimum requirements, and
we prepared new program in synergy with the Faculty of Organizational Sciences.
The Master thesis will be with a research character and the work on the thesis was followed by 2 mentors, from academia and from industry.
Like any responsible institution, we have set several objectives when we developed this program.
As the saying goes: The world remains for the younger.
Team: Different level od education and experience, but mostly from quantitive fields
Me: Engineering background, mostly with research experience, SC first contact with clients and creating DS producti from scratch to production
Etihad: working in DS team, devoloping models but closely cooperating with Data Engineers and business stakeholders
How does it look like?
PS: Understand existing methods, detect it's weaknesses or space for improvement, define the problem with clearity and conciseness -> ask questions (1-2 days)
Data: Once you understand the problem, describe kind of data you need -> communicate data prerequasities to stakeholder of domain expert
It is good in this phase to understand if you can obtain data in real time, to detect the latency of data extraction -> these can affect your solution design later (1 week)
EDA: This is the place where questions should be asked -> set of features which pottentialy can be beneficial for model (1 week)
Model Development: Trying out different models, understanding the best way to train and validate, estimate the quality of model ( are errors uniformly distributed, is model robust enough/adaptive enough) - > model and training/testing method (1 week)
Benefit Estimation: transfer the predictions to $$ - > numbers (1 week)
Communication: presenting to stakeholders solution design, communicating the benefits and risk, estimating the cost of solution ( additional costs for vendors) -> stakeholders feedback
Refining Output: modyfing the solution design according to the feedback (1-2 days up to week)
Production: Creating production code at Dataiku (with automated scenarios), generating instructions for data enginners, testing (2 weeks)
Problem Statement : More local joiners require more rostered staff and/or more staff allocated to check-in counters, in order to ensure a good customer experience and that queue lengths are not overly long. Ideally, one could know the expected volume of passengers arriving at the airport in advance. Appropriate staffing volumes would also reduce our reliance on staff to perform overtime to cover staffing shortages. Less overtime means less manpower costs.
Data: Daily updated
EDA: arrival curves, route region based (ppl arriving earlier for USA flights than GCC flights)
Solution Design: Models for economy, business, bagdrop counters for 1 day and 1 month ahead.
Jesam li ovo stavila sto je lepo ili sto cu stvarno nesto pametno da kazem ne znam, razmisli jos o ovome