We chose the classic game Pong for our first learning test. The Atari framework tests the ability of a machine-learning technology to learn to play Atari games with no pre-programmed information about how the games work.

GoFish learns better, faster and with less data than the current state-of-the-art neural network

Even after the neural-networks learner has gotten ten times as much training, the GoFish learner still beats it soundly. In fact, the GoFish learner most often wins its sets by about 21 to 6, while the neural-networks learner generally loses by the same margin.

This is important for any task, research or process for which time is of the essence or data is sparse. When GoFish is applied to robotics, robots can learn faster. In medicine, new drugs are innovated faster.