About the ICML 2018 Reading Challenge


ICML is one of the top conferences in machine learning. Their 35th conference is due to happen in Stockholm in July.

Over 600 papers have been accepted to the conference. Nurture.ai will be organising a challenge to read these papers. Join us anytime, anywhere from 26 May to 15 June!

How to join:

  1. Pick a paper accepted by ICML 2018 (See the full list of accepted papers here). Or start with top picks from our research fellows.

  2. Read the paper on the nurture.ai platform, annotate the paper to form a discussion thread. We encourage:

  • Questions and elaborations about terms and methods used in a research preprint, why a particular method was used, and whether the results could be improved by using another method
  • Ideas for extending the research further
  • Flagging results reported and choice of datasets
  • Contrasting methods and approaches in other papers


Here are some papers to start with:

Dissecting Adam: The Sign, Magnitude and Variance of Stochastic Gradients
Attention-based Deep Multiple Instance Learning
The Mechanics of n-Player Differentiable Games
Yes, but Did It Work?: Evaluating Variational Inference
A Progressive Batching L-BFGS Method for Machine Learning
Analyzing the Robustness of Nearest Neighbors to Adversarial Examples
Neural relational inference for interacting systems
Not to Cry Wolf: Distantly Supervised Multitask Learning in Critical Care
Learning Longer-term Dependencies in RNNs with Auxiliary Losses
Overcoming Catastrophic Forgetting with Hard Attention to the Task
NetGAN: Generating Graphs via Random Walks
DiCE: The Infinitely Differentiable Monte Carlo Estimator
QMIX: Monotonic Value Function Factorisation for Deep Multi-Agent Reinforcement Learning
Learning Adversarially Fair and Transferable Representations
TACO: Learning Task Decomposition via Temporal Alignment for Control