Ferenc Huszar - inFERENCe (Page 3)

inFERENCe

Sign in Subscribe

Ferenc Huszar

ICML Highlight: Contrastive Divergence for Combining Variational Inference and MCMC

ICML Highlight: Contrastive Divergence for Combining Variational Inference and MCMC

Welcome to my ICML 2019 jetlag special - because what else do you do when you wake up earlier than anyone than write a blog post. Here's a paper that was presented yesterday which I really liked. * Ruiz and Titsias (2019) A Contrastive Divergence for Combining Variational Inference

Notes on the Limitations of the Empirical Fisher Approximation

Notes on the Limitations of the Empirical Fisher Approximation

This post is a short not on an excellent recent paper on empirical Fisher information matrices: * Kunstner, Balles and Hennig (2019) Limitations of the Empirical Fisher Approximation [https://arxiv.org/abs/1905.12558] I was debating with myself whether I should write a post about this because it's

Perceptual Straightening of Natural Videos

Perceptual Straightening of Natural Videos

Video is an interesting domain for unsupervised, or self-supervised, representation learning. But we still don't know what type of inductive biases will enable us to best exploit the information encoded in the temporal sequence of video frames. Slow Feature Analysis [http://scholarpedia.org/article/Slow_feature_analysis] (SFA)

Causal Inference 3: Counterfactuals

Causal Inference 3: Counterfactuals

Counterfactuals are weird. I wasn't going to talk about them in my MLSS lectures on Causal Inference, mainly because wasn't sure I fully understood what they were all about, let alone knowing how to explain it to others. But during the Causality Panel [https://www.youtube.

Causal Inference 2: Illustrating Interventions via a Toy Example

Causal Inference 2: Illustrating Interventions via a Toy Example

Last week I had the honor to lecture at the Machine Learning Summer School in Stellenbosch, South Africa [https://mlssafrica.com/]. I chose to talk about Causal Inference, despite being a newcomer to this whole area. In fact, I chose it exactly because I'm a newcomer: causal inference

Online Bayesian Deep Learning in Production at Tencent

Bayesian deep learning methods often look like a theoretical curiosity, rather than a practically useful tool, and I'm personally a bit skeptical about the practical usefulness of some of the work. However, there are some situations a decent method of handling and representing residual uncertainties about model parameters

👻Halloween Special: Critical reviews of the worst NIPS 2018 papers.

👻Halloween Special: Critical reviews of the worst NIPS 2018 papers.

Boo. I scared you, didn't I?

The Blessings of Multiple Causes: Causal Inference when you Can't Measure Confounders

The Blessings of Multiple Causes: Causal Inference when you Can't Measure Confounders

Happy back-to-school time everyone! After a long vacation (in blogging terms), I'm returning with a brief paper review which conveniently allows me to continue to explain a few ideas from causal reasoning. Since I wrote this intro to causality [https://www.inference.vc/untitled/], I have read a

ML beyond Curve Fitting: An Intro to Causal Inference and do-Calculus

ML beyond Curve Fitting: An Intro to Causal Inference and do-Calculus

Since writing this post back in 2018, I have extended this to a 4-part series on causal inference: * ➡️️ Part 1: Intro to causal inference and do-calculus [https://www.inference.vc/untitled] * Part 2: Illustrating Interventions with a Toy Example [https://www.inference.vc/causal-inference-2-illustrating-interventions-in-a-toy-example/] * Part 3: Counterfactuals [https://www.inference.

The Lottery Ticket Hypothesis - Paper Recommendation

The Lottery Ticket Hypothesis - Paper Recommendation

I wanted to highlight a recent paper I came across, which is also a nice follow-up to my earlier post [https://www.inference.vc/pruning-neural-networks-two-recent-papers/] on pruning neural networks: * J Frankle and M Carbin (2018) The Lottery Ticket Hypothesis: Training Pruned Neural Networks [https://arxiv.org/abs/1803.03635] Many