Seminar: Robust Data Mining Techniques - Pre-course meeting€¦ · Example: Linear Regression 2 0...
Transcript of Seminar: Robust Data Mining Techniques - Pre-course meeting€¦ · Example: Linear Regression 2 0...
![Page 1: Seminar: Robust Data Mining Techniques - Pre-course meeting€¦ · Example: Linear Regression 2 0 2 4 6 8 10 12 50 0 50 100 150 200 250 Fit without outlier Fit with outlier Robust](https://reader036.fdocuments.net/reader036/viewer/2022063008/5fbbe93b150dce6e7f7139b4/html5/thumbnails/1.jpg)
Seminar: Robust Data Mining TechniquesPre-course meeting
Technische Universitat MunchenDepartment of InformaticsData Mining and Analyticskdd.in.tum.de
January 27, 2017
![Page 2: Seminar: Robust Data Mining Techniques - Pre-course meeting€¦ · Example: Linear Regression 2 0 2 4 6 8 10 12 50 0 50 100 150 200 250 Fit without outlier Fit with outlier Robust](https://reader036.fdocuments.net/reader036/viewer/2022063008/5fbbe93b150dce6e7f7139b4/html5/thumbnails/2.jpg)
Introduction
Why robustness matters?
• Data is inherently noisy
• Guarantees inperformance-critical areas
• Adversarial scenarios
• Crowdsourcing
Robust Data Mining Techniques 1Data Miningand Analytics
![Page 3: Seminar: Robust Data Mining Techniques - Pre-course meeting€¦ · Example: Linear Regression 2 0 2 4 6 8 10 12 50 0 50 100 150 200 250 Fit without outlier Fit with outlier Robust](https://reader036.fdocuments.net/reader036/viewer/2022063008/5fbbe93b150dce6e7f7139b4/html5/thumbnails/3.jpg)
Example: Linear Regression
2 0 2 4 6 8 10 1250
0
50
100
150
200
250Fit without outlierFit with outlier
Robust Data Mining Techniques 2Data Miningand Analytics
![Page 4: Seminar: Robust Data Mining Techniques - Pre-course meeting€¦ · Example: Linear Regression 2 0 2 4 6 8 10 12 50 0 50 100 150 200 250 Fit without outlier Fit with outlier Robust](https://reader036.fdocuments.net/reader036/viewer/2022063008/5fbbe93b150dce6e7f7139b4/html5/thumbnails/4.jpg)
Example: Deep Learning
Robust Data Mining Techniques 3Data Miningand Analytics
![Page 5: Seminar: Robust Data Mining Techniques - Pre-course meeting€¦ · Example: Linear Regression 2 0 2 4 6 8 10 12 50 0 50 100 150 200 250 Fit without outlier Fit with outlier Robust](https://reader036.fdocuments.net/reader036/viewer/2022063008/5fbbe93b150dce6e7f7139b4/html5/thumbnails/5.jpg)
Topics
Robust extensions to data mining algorithms
• Matrix factorization
• Regression
• Classification
• Clustering
• Time series analysis
Further topics in robust machine learning
• Robustness of complex networks
• Differential privacy
• Fooling learning algorithms with adversarial examples
• Designing algorithms for the adversarial setting
Robust Data Mining Techniques 4Data Miningand Analytics
![Page 6: Seminar: Robust Data Mining Techniques - Pre-course meeting€¦ · Example: Linear Regression 2 0 2 4 6 8 10 12 50 0 50 100 150 200 250 Fit without outlier Fit with outlier Robust](https://reader036.fdocuments.net/reader036/viewer/2022063008/5fbbe93b150dce6e7f7139b4/html5/thumbnails/6.jpg)
Learning Outcome
You are going to learn
• about the design of robust data mining algorithms
• to read and understand scientific publications
• to write a scientific report
• how to prepare and give a technical talk
Robust Data Mining Techniques 5Data Miningand Analytics
![Page 7: Seminar: Robust Data Mining Techniques - Pre-course meeting€¦ · Example: Linear Regression 2 0 2 4 6 8 10 12 50 0 50 100 150 200 250 Fit without outlier Fit with outlier Robust](https://reader036.fdocuments.net/reader036/viewer/2022063008/5fbbe93b150dce6e7f7139b4/html5/thumbnails/7.jpg)
Requirements
Paper
• 5 - 8 pages
• Latex template on the course webpage
Presentation
• 30 minutes talk
• 15 minutes discussion
Reviews
• Everyone has to review 2 papers by other students
Robust Data Mining Techniques 6Data Miningand Analytics
![Page 8: Seminar: Robust Data Mining Techniques - Pre-course meeting€¦ · Example: Linear Regression 2 0 2 4 6 8 10 12 50 0 50 100 150 200 250 Fit without outlier Fit with outlier Robust](https://reader036.fdocuments.net/reader036/viewer/2022063008/5fbbe93b150dce6e7f7139b4/html5/thumbnails/8.jpg)
Deadlines
• 1 week before the talk - submission of extended abstractand slides
• Day of the talk - submission of preliminary paper for review
• 1 week after the talk - receiving comments from reviewers
• 2 weeks after the talk - submission of the final paper
Robust Data Mining Techniques 7Data Miningand Analytics
![Page 9: Seminar: Robust Data Mining Techniques - Pre-course meeting€¦ · Example: Linear Regression 2 0 2 4 6 8 10 12 50 0 50 100 150 200 250 Fit without outlier Fit with outlier Robust](https://reader036.fdocuments.net/reader036/viewer/2022063008/5fbbe93b150dce6e7f7139b4/html5/thumbnails/9.jpg)
Grading
The grade is determined based on
• Report
• Presentation (slides and speech)
• Reviews written by you
• Involvement in the class
• Interactions with the supervisor
• Extra bonuses for own contributions (e.g. visualizations,demos, experiments)
Robust Data Mining Techniques 8Data Miningand Analytics
![Page 10: Seminar: Robust Data Mining Techniques - Pre-course meeting€¦ · Example: Linear Regression 2 0 2 4 6 8 10 12 50 0 50 100 150 200 250 Fit without outlier Fit with outlier Robust](https://reader036.fdocuments.net/reader036/viewer/2022063008/5fbbe93b150dce6e7f7139b4/html5/thumbnails/10.jpg)
Schedule
• Before 08.02. - fill out the pre-course surveyhttps://goo.gl/forms/rWxXhsMpZPW6kpcE2
• 03.02. - 08.02. - registration via the matching system
• After 15.02. - notification of the participants and selection oftopics
• April - June - weekly sessions every Monday 14:30 - 16:00
Robust Data Mining Techniques 9Data Miningand Analytics
![Page 11: Seminar: Robust Data Mining Techniques - Pre-course meeting€¦ · Example: Linear Regression 2 0 2 4 6 8 10 12 50 0 50 100 150 200 250 Fit without outlier Fit with outlier Robust](https://reader036.fdocuments.net/reader036/viewer/2022063008/5fbbe93b150dce6e7f7139b4/html5/thumbnails/11.jpg)
Questions?