Gradients delivered to your weights on time.
Let's leave delayed reward for those RL researchers. You want your gradients on time and where you expect them. We will outsource the calculation of hand crafted gradients specifically designed for each weight.
Don't let vanishing gradients impact your company.
Sometimes gradients just disappear and there is nothing you can do about it. Not! It is just because some pipes have leaks inside your GPU. We can fix these by simply creating more gradient from your existing gradients with the so called "Normalization Trick".
We can connect anything.
For over 4 years we have connected different data sources with perfect combinations of autoencoders (AEs), recursive neural networks (RNNers), and classic Convolutional Neural Networks (Yanners). We can take anything as input and produce anything as output!
Forever vigilant
We use a combination of Jupyter notebooks, Tensorboard, Visdom, and thousands of carefully crafted printf statements to make sure training converges 100% of 30% of the time.
We don't believe neural networks are people yet.
No problem. No need to worry about that in case you were thinking that could be an issue. It isn't. Besides that one time, there is no reason for us to suspect that we are creating sentient life.
Contact
We are here to help with with your problems. Connect with us at
