Niki Kilbertus

My mission is to build socially beneficial, robust, and theoretically substantiated machine learning systems.

I am a PhD student in the Cambridge-Tübingen program (prospective graduation in 2020), where I am co-supervised by Bernhard Schölkopf and Carl Rasmussen. My advisor is Adrian Weller. During my PhD I interned at Deepmind and Amazon.

I am a member of Pembroke College, funded by the Cambridge-Tübingen PhD fellowship with generous donations from Microsoft.

My background is in Physics and Mathematics. I was fortunate to spend time at Harvard, working with Paul Chesler and Wilke van der Schee, as well as at Stanford, working with William East and Tom Abel.

News

06/2019: Improving consequential decisions under imperfect predictions @ KDD 2019 Workshop (DCCL)
06/2019: Convolutional neural networks: a magic bullet for gravitational-wave detection? @ Physical Review D
05/2019: The sensitivity of counterfactual fairness to unmeasured confounding @ UAI 2019
02/2019: 2nd edition of our book Quod erat knobelandum is now available at Springer [German]
12/2018: Organized the NeurIPS18 Workshop on Privacy Preserving Machine Learning
11/2018: Generalization in anti-causal learning @ NeurIPS18 Workshop (CRACT)
05/2018: Two papers @ ICML 2018

Selected Publications & Projects

The sensitivity of counterfactual fairness to unmeasured confounding

NK, Philip Ball, Matt Kusner, Adrian Weller, Ricardo Silva

UAI 2019

[paper] [code]

Convolutional neural networks: a magic bullet for gravitational-wave detection?

Timothy Gebhard*, NK*, Ian Harry, Bernhard Schölkopf (* equal contribution)

Physical Review D (accepted, coming soon), 2019

[paper] [code] [data generation]

Fair Decisions Despite Imperfect Predictions

NK, Manuel Gomez-Rodriguez, Bernhard Schölkopf, Krikamol Muandet, Isabel Valera

[paper] [code]

shorter version: Improving consequential decision making under imperfect predictions

KDD 2019 Workshop on Data Collection, Curation, and Labeling for Mining and Learning (DCCL)

Generalization in anti-causal learning

NK*, Giambattista Parascandolo*, Bernhard Schölkopf (* equal contribution)

NeurIPS 2018 Workshop on Critiquing and correcting trends in machine learning

[paper]

Blind Justice: Fairness with Encrypted Sensitive Attributes

NK, Adrià Gascón, Matt J. Kusner, Michael Veale, Krishna P. Gummadi, Adrian Weller

ICML 2018

also at: FATML 2018 [talk] and PIMLAI 2018

[paper] [bibtex] [poster] [code]

Learning Independent Causal Mechanisms

Giambattista Parascandolo, NK, Mateo Rojas-Carulla, Bernhard Schölkopf

ICML 2018

also at: NIPS 2017 Workshop on Learning Disentangled Representations

[paper] [bibtex]

Avoiding Discrimination Through Causal Reasoning

NK, Mateo Rojas-Carulla, Giambattista Parascandolo, Moritz Hardt, Dominik Janzing, Bernhard Schölkopf

NeurIPS 2017

[paper] [bibtex] [poster] [related grant: Digital Impact Grant by Stanford PACS]

ConvWave: Searching for Gravitational Waves with Fully Convolutional Neural Nets

Timothy Gebhard*, NK*, Giambattista Parascandolo, Ian Harry, Bernhard Schölkopf (* equal contribution)

NeurIPS 2017 Workshop on Deep Learning for Physical Sciences

[paper] [bibtex] [code] [poster]

Universal Hydrodynamic Flow in Holographic Planar Shock Collisions

Paul Chesler, NK, Wilke van der Schee

Journal for High Energy Physics, 2015

[paper, arxiv version] [detailed project report (pdf, ~1MB)]

Quod erat knobelandum

Clara Löh, Stefan Krauss, NK

Springer Spektrum, (1st edition: 2016, 2nd edition: 2019)

[springer] [amazon]

Master Thesis Physics: Numerical Analysis of Gravitational Wave Generation during Metric Preheating

[thesis (pdf, ~12MB)] [code]

Master Thesis Mathematics: Numerical Analysis of Causal Fermion Systems on R x S^3

[thesis (pdf, ~3.8MB)]

Physics Project: Sky-MoCa: The Skyrmion Phase in 3D Lattice Simulations

[report] [code]

News mentions and science communication

MPI news: The Question is Why -- Algorithms learn a Sense of Fairness (2017)
Financial Times: Finding a fair way to tame the bigoted bots (2018)
New Scientist: How to stop artificial intelligence being so racist and sexist (2018)
Second Nexus: Niki Kilbertus of Max Planck Institute for Intelligent Systems Has a Plan to Remove Bias From AIs (2018)
The Alan Turing Institute: Can justice be blind when it comes to machine learning? (2018)
MPI news: Blind Justice -- Researchers take new approach to machine learning fairness by applying privacy methods (2018)
Max Planck Forschung: Auf Fairness programmiert (2019)
Matt, Adrià, and Adrian talked about our work in three different episodes of The Talking Machines podcast (2018, 2019)

Talks

Albert Einstein Institute (Potsdam-Golm, Germany): Machine Learning powered CBC Search
Alan Turing Institute (London, UK): Fairness in Machine Learning
Max Planck Institute for Software Systems (Saarbrücken, Germany): Fairness in Machine Learning
Stanford University (CA, USA): Searching for Gravitational Waves with Machine Learning
University of Regensburg (Regensburg, Germany): Fully Convolutional Networks for Gravitational Wave Searches
Microsoft Research (Cambridge, UK): Learning Independent Causal Mechanisms
Amazon Research (Cambridge, UK): Blind Justice: Fairness with Encrypted Sensitive Attributes

Miscellaneous

Organizer of the NeurIPS 2018 workshop on Privacy Preserving Machine Learning
Main organizer of the CamTue workshops: Mallorca 2017, Tenerife 2018
I received a research grant from the Digital Impact Grant by Stanford PACS
I thoroughly enjoy teaching, was active in the Schülerzirkel Mathematik in Regensburg, a TA for many courses in Math, Physics, and CS, lectured a semi-annual course on Computer- and Microcontroller-Technology, and co-lectured the course Green-IT at the summer academy 2016 in Leysin, organized by the German Academic Scholarship Foundation.
I like building things, for example: Babyzen - A flexible sensor BoosterPack [codeproject article][short video][report (pdf)] or some things here.