Twitter | Search | |
Florian Wilhelm 🇪🇺
Data Scientist and passionate mathematician interested in machine learning, deep learning, recommendation systems and data science in general.
2,595
Tweets
891
Following
814
Followers
Tweets
Florian Wilhelm 🇪🇺 retweeted
Danilo Bzdok 23h
fact: Shrinkage-based estimation starts to supersede in Gaussian models with 3 or more parameters
Reply Retweet Like
Florian Wilhelm 🇪🇺 retweeted
Even Oldridge Sep 16
Excited to be at sharing work on how we used and to accelerate our recsys challenge submission by 15.6x! Lots of great deep learning optimizations and tricks in the repo. Check out the blog for details:
Reply Retweet Like
Florian Wilhelm 🇪🇺 retweeted
Sherry Sahebi Sep 17
Very nice best paper on reproducing deep learning recsys algorithms Many deep learning algorithms are not competitive with simple baselines
Reply Retweet Like
Florian Wilhelm 🇪🇺 Sep 17
Reproducibility of algorithms in various conference of the last 3 years from the "Are We Really Making Much Progress? A Worrying Analysis of Recent Neural Recommendation Approaches" talk at .
Reply Retweet Like
Florian Wilhelm 🇪🇺 Sep 16
Erzsébet Frigó uses to explain a new approx. gradient optimization algorithm in her talk "Online Ranking Combination" which is a really neat idea! Makes it visual and fun.
Reply Retweet Like
Florian Wilhelm 🇪🇺 Sep 16
Cool, best long paper at is actually quite self-critical about the progress made with deep neural networks in recommendation systems.
Reply Retweet Like
Florian Wilhelm 🇪🇺 Sep 15
Just arrived at , looking forward to some inspiring talks.
Reply Retweet Like
Florian Wilhelm 🇪🇺 retweeted
Uwe L. Korn Sep 15
Building Linux Python wheels for has not been that easy and we faced some pretty complex toolchain issues. As I'm no longer maintaining them, I wrote up a bit on what makes them complex
Reply Retweet Like
Florian Wilhelm 🇪🇺 retweeted
David Page Sep 11
The paper that introduced Batch Norm combines clear intuition with compelling experiments (14x speedup on ImageNet!!) So why has 'internal covariate shift' remained controversial to this day? Thread 👇
Reply Retweet Like
Florian Wilhelm 🇪🇺 retweeted
Jonas K. Lindeløv Sep 11
I've made a cheat sheet and a bunch of applets to give you an intuitive feel for various reaction time distributions. You can choose datasets, fiddle with parameters, and see working code examples: 1/n
Reply Retweet Like
Florian Wilhelm 🇪🇺 retweeted
Cody Wild Sep 7
Over the years, I've watched conditional renormalization grow from a style transfer hack to a key mechanism in recent results; in this post I chart out how the idea evolved from humble beginnings into a flexible and important technique
Reply Retweet Like
Florian Wilhelm 🇪🇺 retweeted
Python Weekly Sep 4
Causal ML - Uplift modeling and causal inference with machine learning algorithms.
Reply Retweet Like
Florian Wilhelm 🇪🇺 Sep 2
Cool blog post about the evaluation of multimodel sequential by my master student Tim Bicker.
Reply Retweet Like
Florian Wilhelm 🇪🇺 retweeted
PyConDE & PyData Berlin 2019 Aug 30
Are you sure about that?! Uncertainty Quantification in AI helps you to decide if you can trust a prediction or rather not. at
Reply Retweet Like
Florian Wilhelm 🇪🇺 retweeted
Simon Boehm Aug 28
Handed in my Bachelor's thesis yesterday 🙌 - Our paper (Submitted to ICLR 2020): - Blog post (Mainly about Normalizing Flows): Goodbye 👋, Hello 🇨🇭
Reply Retweet Like
Florian Wilhelm 🇪🇺 retweeted
Anna J. Egalite Aug 27
In my intro stats class today, I told students the median is a ”resistant” measure of a distribution’s center & is often preferred to the mean in the case of salary data, etc. I jokingly referenced this meme and in the 15 mins’ break they had, a student created this MASTERPIECE!
Reply Retweet Like
Florian Wilhelm 🇪🇺 retweeted
Jean Feydy Aug 26
Optimal Transport generalizes sorting to spaces of dimension D > 1. GeomLoss implements a smooth Quicksort-like solver, packaged as a PyTorch layer. Keeps a linear memory usage + x10 to x1,000 speedup compared with the Auction and Sinkhorn algorithms.
Reply Retweet Like
Florian Wilhelm 🇪🇺 retweeted
Edouard Grave Aug 24
New blogpost about two recent papers on Transformer networks.
Reply Retweet Like
Florian Wilhelm 🇪🇺 retweeted
Gabriel Peyré Aug 24
Copula remaps the cumulative function to have uniform marginals. It is a way to visualize dependences between variables.
Reply Retweet Like
Florian Wilhelm 🇪🇺 Aug 24
Well, my critics are rather about the Pandas frontend. Haven't tried yet, but will check it out. One thing I need are UD(A)Fs and those only Pandas and PySpark provides so far. too?
Reply Retweet Like