I'm offering deep learning workshops and trainings.
I spent around four years with different machine learning techniques.
I started with simple models like naive bayesian filtering, then spent quite some time with graphical models, like bayesian networks. Useful if you have little data and want fast inference. Graphical Models can be compiled for fast inference.
Got interested in deep learning with Restricted Boltzmann Machines and auto-encoders.
Applied convolutional networks for Go move prediction (what AlphaGo called supervised learned policy) Used that policy to give hints in my Android App Guess The Move
I've applied convolutional networks to automatically crop fashion images. Data augmentation with elastic deformations was essential, since I started with only 700 tagged images.
I trained a value function that predicted if a certain image was already cropped too much or if there was still the whole fashion model visible. I derived lots of different training images from the few tagged images.
Topics for Workshops and Trainings
- the frameworks Torch and PyTorch
- Reinforcement Learning theory, algorithms and applications.
- generative models, GAN
- data augmentation
All trainings contain many practical exercises.
If you want to use deep learning but don't know how ==> talk to me
If deep learning doesn't work as well as you expected ==> talk to me
- recurrent networks
- semi-supervised learning, ladder networks?
- binarized neural networks, deep learning for low power devices
- tagging masses of images quickly and comfortably