Ralf Herbrich

Ralf Herbrich

Managing Director at Amazon Development Center Germany GmbH and Director of Machine Learning

Location
Berlin Area, Germany
Industry
Information Technology and Services

As a LinkedIn member, you'll join 300 million other professionals who are sharing connections, ideas, and opportunities.

  • See who you and Ralf Herbrich know in common
  • Get introduced to Ralf Herbrich
  • Contact Ralf Herbrich directly

View Ralf's full profile

Ralf Herbrich's Overview

Current
  • Managing Director at Amazon Development Center Germany GmbH
  • Director of Machine Learning at Amazon
Past
  • Engineering Manager at Facebook
  • Development Manager at Microsoft
  • Director of Future Social Experiences (FUSE) Labs UK at Microsoft
  • Senior Research Leader (Online Services and Advertising Group) at Microsoft Research
  • Research Leader (Applied Games Group) at Microsoft Research
  • Researcher at Microsoft Research
  • PostDoc Researcher at Microsoft Research
Education
  • High School Einstein Gymnasium, Angermünde
Connections

500+ connections

Websites

Ralf Herbrich's Summary

In general, I want my work to have a positive impact to many people's lives. I particularly enjoy the application of principled research to large-scale real-world problems. I enjoy learning new things every day and getting to work with (new) people.

I do not accept existing scientific boundaries and think that the largest breakthroughs will be made at the intersection of existing disciplines. My research interests include Bayesian inference and decision making, game theory, information retrieval, learning theory and knowledge representation and reasoning. I have published over 50 peer reviewed conference and journal papers in these fields.

I am one of the inventors of the Drivatars™ system in the Forza Motorsport series as well as the TrueSkill™ ranking and matchmaking system in Xbox 360 Live (especially in Halo and Halo Reach). I also co-invented the click-prediction technology used in bing's online advertising system (adPredictor). I am co-inventor for several patents relating to these inventions.

Specialties

soft skills

Ralf Herbrich's Experience

Public Company; 10,001+ employees; AMZN; Internet industry

March 2013Present (1 year 7 months) Berlin Area, Germany

Director of Machine Learning

Amazon

Public Company; 10,001+ employees; AMZN; Internet industry

November 2012Present (1 year 11 months) Greater Seattle Area

Engineering Manager

Facebook

Public Company; 5001-10,000 employees; FB; Internet industry

October 2011November 2012 (1 year 2 months) Palo Alto

I am heading the Unified Ranking and Allocation team which focuses on large-scale, distributed probabilistic inference as well as efficient-to-evaluate value models to allow optimizing for long-term economic value. For example, the team works on ad click-through-rate models.

Development Manager

Microsoft

Public Company; 10,001+ employees; MSFT; Computer Software industry

February 2011September 2011 (8 months)

The Bing Personalization team focuses on prototyping and enabling personalized experiences across Microsoft's Online Services Division, including Bing Mobile, Bing News, Bing Web and AdCenter, through agile development and fast deployment of computational intelligence and social computing technologies.

Director of Future Social Experiences (FUSE) Labs UK

Microsoft

Public Company; 10,001+ employees; MSFT; Computer Software industry

November 2009February 2011 (1 year 4 months)

Setup and manages an agile applied research and development team that builds new social experiences arcross Microsoft’s Online Services; we usually releases new demos and prototypes every 3 months.

FUSE UK makes heavy use of computational intelligence technologies (e.g., large scale Bayesian ranking & recommendation systems) and builds new experiences through exploitation, combination and analysis of available data sources within and outside of Microsoft (web search queries, RTW firehoses, etc.)

Sample projects include Project Emporia (http://www.projectemporia.com), a new personalized news reader (available on Web and Windows Mobile 7), influence analysis techniques in distribution networks such as Twitter (used within Project Emporia and published at NIPS) as well as social advertising projects.

Senior Research Leader (Online Services and Advertising Group)

Microsoft Research

Public Company; 10,001+ employees; MSFT; Computer Software industry

June 2008October 2009 (1 year 5 months)

Jointly building and managing a cross-divisional research group at Microsoft (MSR, Online Services Division, Microsoft's Business Division) focused on applied research in online services (together with Thore Graepel).

Supported the launch of adPredictor, the new click-through rate prediction system used in today's bing keyword auctions (2008 - 2009).

Research Leader (Applied Games Group)

Microsoft Research

Public Company; 10,001+ employees; MSFT; Computer Software industry

January 2006May 2008 (2 years 5 months)

Jointly heading a small team of PostDocs, Researcher and Research Developers (together with Thore Graepel).

Developing and shipping the ranking system for Halo 3 together with Bungie (2006 - 2007).

Developing and inventing the new click-through prediction system used today in bing's keyword auction system (adPredictor).

Researcher

Microsoft Research

Public Company; 10,001+ employees; MSFT; Computer Software industry

November 2001December 2005 (4 years 2 months)

Inventing and developing large-scale Bayesian algorithms for classification and ranking learning, (invariant) pattern recognition, (structural) fault networking learning and reinforcement learning.

Developing and shipping the Drivatar system - the adaptive game AI in the Xbox racing game (series) Forza Motorsport (2002 - 2004).

Developing and shipping TrueSkill, the ranking and matchmaking system used in all of Xbox Live's competetive online gaming (2004 - 2005).

PostDoc Researcher

Microsoft Research

Public Company; 10,001+ employees; MSFT; Computer Software industry

September 2000October 2001 (1 year 2 months)

Working on machine learning theory, optimization theory, and applied machine learning (support vector machines, Bayes Point Machines, Hidden Markov Models, Neural Networks, Graphical Models).

Written and published "Learning Kernel Classifiers" with The MIT Press (http://learning-kernel-classifiers.org/) based on my PhD thesis (sold more than 2000 copies to date)

Published 10 conference papers and 2 journal papers on kernel methods and a new learning theory framework ("Algorithmic Luckiness") at NIPS, ICML and JMLR.

Ralf Herbrich's Patents

  • Ranking search results using feature score distributions

    • United States Patent Application US20130024448
    • Filed July 21, 2011

    Document features or document ranking values can be associated with a distribution of values. Feature values, feature value coefficients, and/or document ranking values can be generated based on sampled values from the distribution of values. This can allow the relative ranking of a document to vary. As additional information is obtained regarding the document, leading to greater certainty about the appropriate ranking of the document, the width or variation generated by the distribution can be reduced to provide more stable ranking values.

  • Ranking search results using feature score distributions

    • United States Patent Application US20130024448
    • Filed July 21, 2011

    Document features or document ranking values can be associated with a distribution of values. Feature values, feature value coefficients, and/or document ranking values can be generated based on sampled values from the distribution of values. This can allow the relative ranking of a document to vary. As additional information is obtained regarding the document, leading to greater certainty about the appropriate ranking of the document, the width or variation generated by the distribution can be reduced to provide more stable ranking values.

  • Ranking search results using feature score distributions

    • United States Patent Application US20130024448
    • Filed July 21, 2011

    Document features or document ranking values can be associated with a distribution of values. Feature values, feature value coefficients, and/or document ranking values can be generated based on sampled values from the distribution of values. This can allow the relative ranking of a document to vary. As additional information is obtained regarding the document, leading to greater certainty about the appropriate ranking of the document, the width or variation generated by the distribution can be reduced to provide more stable ranking values.

Ralf Herbrich's Publications

Ralf Herbrich's Skills & Expertise

  1. Project Management
  2. Research
  3. Group leader
  4. C#
  5. C++
  6. SQL
  7. F#
  8. Strategy development
  9. People Management & Development
  10. Machine Learning
  11. Algorithms
  12. Information Retrieval
  13. Strategy
  14. Computer Science
  15. Data Mining
  16. Distributed Systems
  17. Neural Networks
  18. Agile Methodologies
  19. Software Engineering
  20. Software Development
  21. JavaScript
  22. Scalability
  23. Image Processing
  24. Pattern Recognition
  25. Software Design
  26. Hadoop
  27. Optimization
  28. Artificial Intelligence
  29. Recommender Systems
  30. Python
  31. Natural Language Processing
  32. MapReduce
  33. Big Data
  34. Object Oriented Design
  35. User Experience
  36. Mobile Devices
  37. High Performance Computing

View All (37) Skills View Fewer Skills

Ralf Herbrich's Languages

  • English

    (Full professional proficiency)
  • German

    (Native or bilingual proficiency)
  • Russian

    (Elementary proficiency)

Ralf Herbrich's Education

Research Fellow at Darwin College Cambridge

PhD, Computer Science

20002003

Microsoft Research Fellow, interested in the philosophy of Science.

Technische Universität Berlin

PhD, Theoretical Statistics

19972000

Focus on computational and statistical learning theory.

Jointly writing my thesis with Thore Graepel.

Technische Universität Berlin

Diploma, Computer Science & Business

19921997

Majors in AI (symbolic and rule-based systems) and computer graphics.

Additional courses in "The Theory of Science" and computer-assisted radiology.

15 months civil service as a paramedic in local hospital. Worked in the theatre and assisted (minor) surgeries (1993 - 1994).

High School Einstein Gymnasium, Angermünde

College, Abitur

19901992

Ralf Herbrich's Additional Information

Websites:
Interests:

cooking, running, fishing, (online) gaming, my kids, reading (innovation styles and management)

Groups and Associations:

Contact Ralf for:

  • career opportunities
  • new ventures
  • job inquiries
  • expertise requests
  • business deals
  • reference requests
  • getting back in touch

View Ralf Herbrich’s full profile to...

  • See who you and Ralf Herbrich know in common
  • Get introduced to Ralf Herbrich
  • Contact Ralf Herbrich directly

View Ralf's full profile

Viewers of this profile also viewed...