Repository of Team TensorFlow
Switch branches/tags
Nothing to show
Clone or download
Fetching latest commit…
Cannot retrieve the latest commit at this time.
Permalink
Type Name Latest commit message Commit time
Failed to load latest commit information.
sprint1_tf_tutorial
sprint2_stanford_tf_cs20SI
sprint3_tf_models/src
sprint4_tf_pt_gan/src
.gitignore
.gitmodules
README.md
ressourcensammlung.md

README.md

team-tensorflow

Repository of the ML-KA Subgroup Team TensorFlow.

Abstract

Team TensorFlow was founded by enthusiasts that already had some theoretical knowledge in neural networks and wanted to transition themselves into NN-practitioners, expecting a challenging learning curve. We are aware that there are several frameworks for neural networks out there, backed by some major companies, and all of them have assets and drawbacks. We’ve decided to go for TensorFlow, because it’s backed by Google, provides an awesome documentation, a strong community and some freely available third-party resources (like online courses). The framework requires medium-level implementations with a non-negligible level of granularity, but we are expecting this to enforce our skills even better than an implementation with a higher-level abstraction (like Keras).

Our main objective is to gain experience in implementing and training neural networks with only marginal references to conventional ML algorithms. Some theoretical background in neural networks is recommended. Feel free to join us, we are open for everyone.

Schedule

Weekly meetings are intended. They are currently scheduled Tuesday, 17:30-19:00. For up-to-date and location info, please refer to the calendar @ http://ml-ka.de/.

What we do

We learn to

  • read TensorFlow code (Python API)
  • build own models
  • apply pretrained models
  • write well-structured code

by code inspection and discussion of already existing models. Moreover, we document our approaches and resources to accelerate everyone's learning curve.

Things to Come

  • We are currently focussing on CNNs and follow their success story. Our outline is based on this blog post. A snapshot of models / problems:
    • MNIST (Multilayer Perceptron and CNN)
    • AlexNet and improvements
    • GoogleLeNet (with the Inception Module)
    • ResNet
    • Generative Adversarial Network
  • Future tracks deal with Recurrent Nets and Reinforcement Learning