NeurIPS 2017

Filtering Variational Objectives

NeurIPS 2017 tensorflow/models

When used as a surrogate objective for maximum likelihood estimation in latent variable models, the evidence lower bound (ELBO) produces state-of-the-art results.

REBAR: Low-variance, unbiased gradient estimates for discrete latent variable models

NeurIPS 2017 tensorflow/models

Learning in models with discrete latent variables is challenging due to high variance gradient estimators.

LightGBM: A Highly Efficient Gradient Boosting Decision Tree

NeurIPS 2017 Microsoft/LightGBM

We prove that, since the data instances with larger gradients play a more important role in the computation of information gain, GOSS can obtain quite accurate estimation of the information gain with a much smaller data size.

Attention Is All You Need

NeurIPS 2017 tensorflow/tensor2tensor

The dominant sequence transduction models are based on complex recurrent or convolutional neural networks in an encoder-decoder configuration.


GANs Trained by a Two Time-Scale Update Rule Converge to a Local Nash Equilibrium

NeurIPS 2017 jantic/DeOldify

Generative Adversarial Networks (GANs) excel at creating realistic images with complex models for which maximum likelihood is infeasible.


A Unified Approach to Interpreting Model Predictions

NeurIPS 2017 slundberg/shap

Understanding why a model makes a certain prediction can be as crucial as the prediction's accuracy in many applications.


Convolutional Gaussian Processes

NeurIPS 2017 pyro-ppl/pyro

We present a practical way of introducing convolutional structure into Gaussian processes, making them more suited to high-dimensional inputs like images.

Doubly Stochastic Variational Inference for Deep Gaussian Processes

NeurIPS 2017 pyro-ppl/pyro

Existing approaches to inference in DGP models assume approximate posteriors that force independence between the layers, and do not work well in practice.

Improved Training of Wasserstein GANs

NeurIPS 2017 tensorpack/tensorpack

Generative Adversarial Networks (GANs) are powerful generative models, but suffer from training instability.


Dynamic Routing Between Capsules

NeurIPS 2017 naturomics/CapsNet-Tensorflow

We use the length of the activity vector to represent the probability that the entity exists and its orientation to represent the instantiation parameters.