Artificial Intelligence

More flexible models with TensorFlow eager execution and Keras

If you have used Keras to create neural networks you are no doubt familiar with the Sequential API, which represents models as a...

It is a wonderful life

Orientation was a wake-up call. “All the first-year students were assembled beneath the big dome,” he recalls. “And a professor told each...

A look at activations and cost functions

You’re building a Keras model. If you haven’t been doing deep learning for so long, getting the output activations and cost function right...

“Every project is a new adventure,” says architectural shape-shifter

Samyn grew up in the Belgian countryside, where his father was a mechanical engineer and his mother was a painter. With a...

Posit AI Blog: Representation learning with MMD-VAE

Recently, we showed how to generate images using generative adversarial networks (GANs). GANs may yield amazing results, but the contract there basically is:...

Calculating the costs of war

To fuel such change, Crawford works to make clear the full cost of military activity, providing data on dollars spent, lives lost,...

Naming and locating objects in images

We’ve all become used to deep learning’s success in image classification. Greater Swiss Mountain dog or Bernese mountain dog? Red panda or giant...

The race to produce rare earth materials

Rivalia prefers to work with existing waste products as opposed to coal that has not yet been burned. This approach is risky;...

You sure? A Bayesian approach to obtaining uncertainty estimates from neural networks

If there were a set of survival rules for data scientists, among them would have to be this: Always report uncertainty estimates with...

Latest articles