If you look into the mathematics of deep learning AI algorithms, you find that a lot of it is based on calculations using matrices (two dimensional grids of values). That's where the Stanford AI course began, for example. But this limits the complexity AI can comprehend, as Anima Anandkumar explains. "In practice, matrix methods in machine learning can't effectively capture higher-order relationships." Instead, "with their multiple dimensions and flexibility, tensors seem like a natural fit for higher-order problems in AI." But as capacity increases, the questions become more urgent. "It's always important to question how our work is going to impact the world... so much of the way we teach in universities is derived from military school. Engineering came from that background, and some of it lingers. Like thinking that scientists and engineers should focus on the technical stuff and let others take care of the rest. It's wrong. We all need humane thinking." See also: Building Neural Networks with Tensorflow.
Today: 3 Total: 1132 [Direct link] [Share]
