Rohit Babbar

Hello, I am an Assistant Professor in the department of Computer Science, Aalto University at Helsinki, Finland. I work on problems in large-scale machine learning, robustness of machine learning models, and Extreme Classification.

I am looking for students interested in doing PhD or Master's thesis in the above or related areas. Please do not hesitate to contact at firstname.lastname_at_aalto.fi.

Research Overview :

Keywords : Large-scale machine learning, robustness of machine learning models, and Extreme Classification.

DiSMEC is a state-of-art classifier for distributed training for datasets with millions of labels. Compared to tree-based methods such as Parabel (which works for vanilla P@k metrics) or PFastreXML (for propensity scored variants), DiSMEC gives much better results on both the metrics with no hyper-parameter tuning. Since 2017, it still remains the most competitive benchmark on Extreme Classification datasets, and works for both multi-label and multi-class datasets. Ideas from this work have changed the course of research in this domain and have adopted in Bing Search engine for related search and advertising. Code can be downloaded here.

Adversarial Extreme Multi-label Classification : This work is an attempt to view Extreme Classification problems from the lens of robustness. We show that by taking this perspective, we get state-of-the-art results on propensity scored variants of P@k and nDCG@k, and hence better predictions on tail-labels. This work also connects to recent works on understanding trade-offs between accuracy and robustness in deep learning models such as Robustness May Be at Odds with Accuracy and Adversarial Risk Bounds via Function Transformation

Extreme Classification Repository - Maintained by Manik Varma, consists of Datasets and (links to) code for most recent methods for extreme classification.

News

Teaching

  • Kernel Methods in Machine Learning (5 ECTS, Jan - April 2019)
  • Supervised Learning with Large Label Sets ( 3 ECTS, Spring 2018)

Publications:

2019

  • Bonsai - Diverse and Shallow Trees for Extreme Multi-label Classification pdf Code

Sujay Khandagale, Han Xiao, Rohit Babbar

  • Data Scarcity, Robustness and Extreme Multi-label Classification pdf

Rohit Babbar and Bernhard Schölkopf

Machine Learning Journal and European Conference on Machine Learning 2019, Wurzburg, Germany

2018

  • Adversarial Extreme Multi-label Classification pdf

Rohit Babbar and Bernhard Schölkopf

Rohit Babbar, Martin Heni, Andreas Peter, Martin Hrabě de Angelis, Hans-Ulrich Häring, Andreas Fritsche, Hubert Preissl, Bernhard Schölkopf, Róbert Wagner

Frontiers in Endocrinology

2017

  • DiSMEC : Distributed Sparse Machines for Extreme Multi-label Classification, pdf Code

Rohit Babbar and Bernhard Schölkopf

ACM International Conference on Web Search and Data Mining (WSDM) 2017, Cambridge;

Also accepted for NIPS 2016 Extreme Classification Workshop

2016

Rohit Babbar, Ioannis Partalas, Eric Gaussier and Massih-reza Amini

Journal of Machine Learning Research , (JMLR 2016)

  • TerseSVM : A scalable approach for learning compact models in Large-scale classification

Rohit Babbar, Krikamol Muandet, and Bernhard Schölkopf

SIAM International Conference on Data Mining (SDM 2016), Miami

2015

  • Efficient Model Selection for Regularized Classification by Exploiting Unlabeled Data

Georgios Balikas, Ioannis Partalas, Eric Gaussier, Rohit Babbar and Massih-Reza Amini

International Symposium on Intelligent Data Analysis, (IDA 2015), Saint-Etienne, France

2014

  • Re-ranking Approach to Classification in Large-scale Power-law Distributed Category Systems, pdf

Rohit Babbar, Ioannis Partalas, Eric Gaussier and Massih-reza Amini

ACM SIGIR Conference , (SIGIR 2014), Gold Coast, Australia,

  • On Power Law Distributions in Large-scale Taxonomies", pdf

Rohit Babbar, Cornelia Metzig, Ioannis Partalas, Eric Gaussier and Massih-reza Amini

SIGKDD Explorations Journal, Special Issue on Big Data

2013

  • On Flat versus Hierarchical Classification in Large-Scale Taxonomies", pdf

Rohit Babbar, Ioannis Partalas, Eric Gaussier and Massih-reza Amini

Neural Information Processing Systems, (NIPS 2013), Lake Tahoe, Neveda, USA

  • Maximum-margin Framework for Training Data Synchronization in Large-scale Hierarchical Classification, pdf,

Rohit Babbar, Ioannis Partalas, Eric Gaussier and Massih-reza Amini

Intl. Conference on Neural Information Processing, (ICONIP 2013), Daegu, Korea

  • Comparative Classifier Evaluation for Web-scale Taxonomies using Power Law,

Rohit Babbar, Ioannis Partalas, Cornelia Metzig, Eric Gaussier and Massih-reza Amini

European Semantic Web Conference (ESWC 2013), Montpiller, France.

2012

  • On Empirical Tradeoffs in Large Scale Hierarchical Classification,

Rohit Babbar, Ioannis Partalas, Eric Gaussier, Cecile Amblard

ACM Intl. Conference on Information and Knowledge Management, (CIKM 2012), Maui, Hawaii.

  • Adaptive Classifier Selection in Large-scale Hierarchical Classification,

Ioannis Partalas, Rohit Babbar, Eric Gaussier, Cecile Amblard

Intl. Conference on Neural Information Processing, (ICONIP 2012), Doha, Qatar.

Academic Trail :

Revewing activities :

  • 2019 - AAAI, ICLR, AISTATS, ICML, NIPS, IJCAI, JMLR, MLJ
  • 2018 - NIPS, ICML, ICLR, AISTATS, AAAI, JMLR, IEEE PAMI
  • 2017 - NIPS, ICML, ICLR, AISTATS

Talks :

  • Magnet Team, INRIA, Lille, January 2017
  • ABC Team, LORIA, Nancy, November, 2016
  • Big Data for Material Sciences Workhsop, Ringberg Castle, July 2016
  • Data Analytics Lab, ETH Zurich, November 2015
  • Bosch Corporate Research Center, Renningen, October 2015
  • Magnet Team, INRIA, Lille, December 2014

Other Talks/Poster Presentations :

  • WSDM 2017, Cambridge
  • NIPS XMC workshop 2016, Barcelona
  • SDM 2016, Miami
  • SIGIR 2014, Gold Coast
  • NIPS 2013, Lake Tahoe
  • ICONIP 2013, Daegu
  • CIKM 2012, Hawaii

Other Stuff

- Long ago, finished Grenoble-Vizille Semi-marathon