How many neurons does it take to ride a bicycle?

📝 Weekly paper summary

It Takes Two Neurons To Ride a Bicycle (Cook, 2004)

Context

This paper is relatively short but tries to understand the interesting challenge of learning to ride a bicycle. The author highlights that previous attempts to get computers to learn how to ride a bike with reinforcement learning (i.e., learning from previous errors) take a very long time to learn. The interesting thing is that humans do not explicitly learn in this statistical way and do not have access to and solve the exact equations of motion while riding a bike. The author walks through creating a two-neuron network that can accurately ride a bike. The purpose of this was to learn more about how systems may learn.

Contributions

This work has no direct applications, but it paves the way for other exciting questions. For example, rather than having a human try to control the bike and provide feedback on how to have the robot ride this particular bike, the next step would be to take the human out of this process and have the robot "learn" to ride the bike without any explicit physical knowledge of the system. Automatically designing some causal network that could accomplish this would provide helpful information to guide future learning theories and the development of artificial intelligence systems.

🧠 Fun fact of the week

Now that Christmas is over, I think it's time for the final holiday-related fun fact. Did you know that Christmas wasn't always on December 25th? Historians believe that the Romans chose this date since it coincided with the Pagan festival of Saturnalia. This festival honoured the god Saturn by offering gifts, decorating with wreaths, lighting candles, and feasting. Christmas absorbed these traditions in ancient Rome to convince the remaining Pagans to accept Christianity as Rome's official religion.

🎙 Podcast recommendation

🗣 Quote of the week

“Be tolerant with others and strict with yourself.”

- Marcus Aurelius