Elmar Haussmann

Tech Lead @NVIDIA | #DeepLearning for #AutonomousVehicles | Previous Co-Founder & CTO @RiseML | PhD in CS | Personal Posts.

[Years Progress @YearsProgress]

▓▓▓▓░░░░░░░░░░░ 25%

[Mohammad Norouzi @mo_norouzi]

SimCLR is a simple and clear way to learn visual representations without class labels!

When SimCLR representations are fine-tuned on 1% of the ImageNet labels, they achieve 85.8% Top-5 accuracy, outperforming AlexNet with 100x fewer labels.

https://t.co/8rsRawjhP0

[Elmar Haussmann @elhaussmann]

@BlackHC Was wondering that too, then assumed likely due to licensing issues?

[Elmar Haussmann @elhaussmann]

@nkoumchatzky @BlackHC @joost_v_amersf Batch acquisition size was effectively ~30k in that experiment. We've looked into BatchBALD (nice work!), but it seemed infeasible at that size. Diversity is indeed something we think is very relevant and are investigating.

[Elmar Haussmann @elhaussmann]

Proud to be part of the team @NvidiaAI working on #ActiveLearning in production #DataDriven #AutonomousVehicles

[Nicolas Koumchatzky @nkoumchatzky]

Very proud of the MagLev team for showing our approach to active learning at scale + amazing results.

Fewer labels, better models: a true technological advantage when building world-class AI perception systems for our Autonomous Vehicles platform.

https://t.co/SujxNZKDbm

[Clément Farabet @clmt]

One of our biggest challenges to build AI for self-driving cars (@nvidiadrive) is to develop datasets to train neural nets to reach extremely high perf targets. We've been exploring a lot to help select better data to reduce our need for it. Our latest: https://t.co/Bq1EPd2ef3

[Clément Farabet @clmt]

Less is More: An Exploration of Data Redundancy with Active Dataset Subsampling. Deep Neural Networks (DNNs) often rely on very large datasets for training. Given the large size of such datasets, it is conceivable... https://t.co/jTSEXQeZVF

[Elmar Haussmann @elhaussmann]

In case you missed #emnlp2018, here are some awesome summaries:
https://t.co/YNRpA01jVk by @CharlotteHase https://t.co/WpseAO00pq by @seb_ruder and https://t.co/JHkXpgMiAn by @PSH_Lewis

[Andy Brock @ajmooch]

This new Schmidhuber&Schmidhuber paper is 🔥🔥🔥

[Alessandro Vespignani @alexvespi]

Are you into soccer, data and machine learning?
This is the right paper to gear up for the world-cup:

Prediction of the FIFA World Cup 2018 - A random forest approach with an emphasis on estimated team ability parameters
https://t.co/D65kyCbt47

[Alec Radford @AlecRad]

What I've been working on for the past year! https://t.co/CAQMYS1rR7

Inspired by CoVE, ELMo, and ULMFiT we show that a single transformer language model can be finetuned to a wide variety of NLP tasks and performs very well with little tuning/tweaking.

[Noah Smith 🐇🇺🇸🇺🇦🇹🇼 @Noahpinion]

I want to make an app where you push a button, and someone, somewhere in the world, randomly gets hit in the head with a fish.

It'll be called "Poisson Distribution".

[RiseML @riseml]

.@riseml adds support for Google Cloud TPUs https://t.co/FdEv22uQx3 @GCPcloud #MachineLearning #DeepLearning

[RiseML @riseml]

Come by booth 1122 @nvidia #GTC18 to learn about using @kubernetesio for #MachineLearning

[RiseML @riseml]

✨ @RiseML now supports distributed deep learning with #Horovod https://t.co/D9PpnZcXVY

[Andrej Karpathy @karpathy]

Google Search trends: Deep Learning vs. Bitcoin. Search traffic is not the best proxy, but I find it interesting that so many of my friends in AI, when asked to guess, guess this relationship consistently waaay off.

[RiseML @riseml]

🚀 We accelerated deep learning training by 377% with this simple I/O trick https://t.co/dbsEHzudGe

[RiseML @riseml]

Reduce your cloud GPU bills with an autoscaling deep learning cluster by @RiseML https://t.co/Lo8A3vGPkY #deeplearning #machinelearning #gpu @awscloud #tensorflow #autoscaling @kubernetesio

[Elmar Haussmann @elhaussmann]

@wheelswordsmith next project after bee detector/tracking @mat_kelcey ? :)

[RiseML @riseml]

Stop by our booth @reworkdl to see how to use #Kubernetes for #DeepLearning #reworkDL #reworkai