Extreme Classification: A New Paradigm for Ranking & Recommendation

Subscribers:
343,000
Published on ● Video Link: https://www.youtube.com/watch?v=1X71fTx1LKA



Duration: 58:30
3,670 views
53


The objective in extreme multi-label classification is to learn a classifier that can automatically tag a data point with the most relevant subset of labels from a large label set. Extreme multi-label classification is an important research problem since not only does it enable the tackling of applications with many labels but it also allows the reformulation of ranking and recommendation problems with certain advantages over existing formulations. Our objective, in this talk, is to develop an extreme multi-label classifier that is faster to train and more accurate at prediction than the state-of-the-art Multi-label Random Forest (MLRF) algorithm [Agrawal et al. WWW 13] and the Label Partitioning for Sub-linear Ranking (LPSR) algorithm [Weston et al. ICML 13]. MLRF and LPSR learn a hierarchy to deal with the large number of labels but optimize task independent measures, such as the Gini index or clustering error, in order to learn the hierarchy. Our proposed FastXML algorithm achieves significantly higher accuracies by directly optimizing an nDCG based ranking loss function. We also develop an alternating minimization algorithm for efficiently optimizing the proposed formulation. Experiments reveal that FastXML can be trained on problems with more than a million labels on a standard desktop in eight hours using a single core and in an hour using multiple cores.




Other Videos By Microsoft Research


2016-06-22The First Order World of Galton-Watson Trees
2016-06-22Verasco, a formally verified C static analyzer
2016-06-22Symposium: Deep Learning - Max Jaderberg
2016-06-22Satisfiability of Ordering CSPs Above Average Is Fixed-Parameter Tractable
2016-06-22Symposium: Deep Learning - Harri Valpola
2016-06-22Machine Learning as Creative Tool for Designing Real-Time Expressive Interactions
2016-06-22Symposium: Deep Learning - Sergey Ioffe
2016-06-22Multi-rate neural networks for efficient acoustic modeling
2016-06-22Robust Spectral Inference for Joint Stochastic Matrix Factorization and Topic Modeling
2016-06-22Computational Limits in Statistical Inference: Hidden Cliques and Sum of Squares
2016-06-22Extreme Classification: A New Paradigm for Ranking & Recommendation
2016-06-22A Lasserre-Based (1+epsilon)-Approximation for Makespan Scheduling with Precedence Constraints
2016-06-22Oral Session: Learning Theory and Algorithms for Forecasting Non-stationary Time Series
2016-06-22Recent Developments in Combinatorial Optimization
2016-06-22Invited Talks: Computational Principles for Deep Neuronal Architectures
2016-06-22Coalescence in Branching Trees and Branching Random Walks
2016-06-22Oral Session: Randomized Block Krylov Methods
2016-06-22Oral Session: Fast Convergence of Regularized Learning in Games
2016-06-22Ito Processes, Correlated Sampling and Applications
2016-06-22Statistical and computational trade-offs in estimation of sparse principal components.
2016-06-22How to Write a Great Research Paper



Tags:
microsoft research
algorithms