Convolutions are an important tool in modern deep neural networks (DNNs). This post is going to discuss some common types of convolutions, specifically regular and depthwise separable convolutions. My focus will be on the implementation of these operation, showing from-scratch Numpy-based code to compute them and diagrams that explain how things work.
Source: thegreenplace.net
The potential harm of A.I.s deliberately designed to kill in warfare is much more pressing. The U.S. and other countries are working hard to develop military A.I., in the form of automated weapons, that enhance battlefield capabilities while exposing fewer soldiers to injury or death. For the U.S., this would be a natural extension of the existing imperfect drone warfare program—failures in military intelligence have led to the mistaken killing of non-combatants in Iraq.
Read More
Motion control problems have become standard benchmarks for reinforcement learning, and deep RL methods have been shown to be effective for a diverse suite of tasks ranging from manipulation to locomotion. However, characters trained with deep RL often exhibit unnatural behaviours, bearing artifacts such as jittering, asymmetric gaits, and excessive movement of limbs. Can we train our characters to produce more natural behaviours?
Read More
Neural networks, which underlie many of Uber’s machine learning systems, have proven highly successful in solving complex problems, including image recognition, language understanding, and game-playing. However, these networks are usually trained to a stopping point through gradient descent, which incrementally adjusts the connections of the network based on its performance over many trials. Once the training is complete, the network is fixed and the connections can no longer change; as a result, barring any later re-training (again requiring many examples), the network in effect stops learning at the moment training ends.
Read More