Eight publications on machine learning presented at world’s best conferences

3 July 2018

Eight publications on machine learning, by researchers from the Institute of Informatics, are accepted by the upcoming leading International Conference on Machine Learning (ICML) and Uncertainty in Artificial Intelligence (UAI). Three articles will be presented at the ICML and five at the UAI.

ICML and UAI are among the best machine learning and AI conferences with a long tradition (the 35th edition of the ICML and the 34th edition of the UAI this year). Acceptance rate at both conferences is low (ICML: ~25% and UAI: ~30%) that allows to keep highest quality of all accepted papers. These numbers show that the Informatics Institute succeeds in its mission of being leading AI institute not only in Europe but also in the world. It is also worth to stress out that some co-authors of the articles (Tim Davidson, Luca Falorsi and Nicola De Cao) are MSc students that makes the success especially valuable.

A discrepancy between the variational lower bound (blue line) and the true log-likelihood function (red). © UvA- AMLAB

ICML 2018

The accepted papers are:

  • Neural Relational Inference for Interacting Systems. Thomas Kipf (UvA), Ethan Fetaya, Kuan-Chieh Wang, Max Welling (UvA), Richard Zemel
  • BOCK: Bayesian Optimization with Cylindrical Kernels. ChangYong Oh (UvA), Efstratios Gavves (UvA), Max Welling (UvA)
  • Attention-based Deep Multiple Instance Learning. Maximilian Ilse (UvA), Jakub Tomczak (UvA), Max Welling (UvA)

Low-dimensional data manifold given by the Variational Auto-Encoder. © UvA- AMLAB

UAI 2018

The accepted papers  are:

  • Sylvester Normalizing Flows for Variational Inference. Rianne van den Berg (UvA), Leonard Hasenclever, Jakub Tomczak (UvA), Max Welling (UvA).
  • Hyperspherical Variational Auto-Encoders.Tim Davidson (UvA), Luca Falorsi (UvA), Nicola De Cao (UvA), Thomas Kipf (UvA), Jakub M. Tomczak (UvA).
  • From Deterministic ODEs to Dynamic Structural Causal Models. Paul K. Rubenstein, Stephan Bongers (UvA), Joris M. Mooij (UvA), Bernhard Schoelkopf.
  • Constraint-based Causal Discovery for Non-Linear Structural Causal Models with Cycles and Latent Confounders. Patrick Forré (UvA), Joris M. Mooij (UvA).
  • Causal Discovery in the Presence of Measurement Error. Tineke Blom (UvA), Sara Magliacane, Anna Klimovskaia, Joris M. Mooij (UvA).

The papers will be presented during the ICML, from 10-15 July 2018 in Stockholm, Sweden, and the UAI, from 6-10 August in Monterey, California, the USA.

Latent representation for a Variational Auto-Encoder with Gaussian prior and posterior (baseline) and a Hyperspherical Variational Auto-Encoder with von-Mises-Fisher posterior and uniform prior over a hypersphere. © UvA- AMLAB

The papers will be presented during the ICML, from 10-15 July 2018 in Stockholm, Sweden, and the UAI, from 6-10 August in Monterey, California, the USA.

Published by  IVI