davidstutz Goto Github PK

followers: 958.0 following: 298.0 repos: 118.0 gists: 1.0

Name: David Stutz

Type: User

Bio: Research scientist at DeepMind, davidstutz.de, [email protected]

Twitter: davidstutz92

Location: London, UK

I'm a research scientist at DeepMind doing research on robust and reliable deep learning. Before, I did a PhD at the Max Planck Institute for Informatics. Here on GitHub you can find research projects as well as open source software developed in my spare time.

Support my open source projects by donating:

David Stutz's Projects

pytorch-bit-error-training-articles

PyTorch code corresponding to my blog series on bit error robustness and bit error training.

pytorch-custom-c-cuda-operations

Example of easily implementing custom tensor operations in C and CUDA.

random-bit-error-robustness

Implementation of robust quantization, weight clipping and random bit error training to improve robustness against bit errors in quantized weights.

robust-generalization-flatness

Implementation of average- and worst-case robust flatness measures for adversarial training.

seeds-revised

Implementation of the superpixel algorithm called SEEDS [1].

seminar-cnn-image-retrieval

Seminar paper "Neural Codes for Image Retreival".

seminar-convolutional-neural-networks

Seminar paper "Understanding Convolutional Neural Networks".

seminar-ipiano

Seminar paper "iPiano: Inertial Proximal Algorithm for Non-Convex Optimization".

seminar-neural-networks

Seminar paper "Introduction to Neural Networks".

sphinx-example

Example and cheat sheet for the Sphinx Python documentation generator.

superpixel-benchmark

An extensive evaluation and comparison of 28 state-of-the-art superpixel algorithms on 5 datasets.

superpixel-benchmark-data

Converted datasets for davidstutz/superpixel-benchmark.

superpixel-benchmark-nvd3

Interactive plots of results from the superpixel benchmark davidstutz/superpixel-benchmark using NVD3.

superpixels-revisited

Library containing 7 state-of-the-art superpixel algorithms with a total of 9 implementations used for evaluation purposes in [1] utilizing an extended version of the Berkeley Segmentation Benchmark.