Scientists make AI brains faster at learning robot movements

A team of scientists has discovered a new technique to make a particular kind of artificial brain work faster and use less power. This artificial brain, called a "spiking neural network," is designed to mimic how real brains process information by sending quick electrical pulses, similar to how neurons in your head communicate.

The challenge researchers were solving is called "inference latency"—basically, the delay between when a system sees something and when it responds. Think of it like the time it takes you to catch a ball after you see it coming. The longer that delay, the slower your reaction feels. For robots learning to copy human movements (a process called "behavior cloning"), this delay matters a lot.

The scientists developed a clever trick called "input adaptive leakage." Instead of processing every bit of information in exactly the same way, the system now adjusts how much information it holds onto based on what it's actually seeing. Imagine a student who focuses hard on important parts of a lesson but lets minor details slip away—that's the basic idea. This selective approach means the artificial brain can make decisions faster while using less energy.

Why should you care? Faster, more efficient AI brains could power more responsive robots and systems in real life, from factory automation to eventual everyday helpers. They'd also use less electricity, which is better for the environment. This research, published in the journal Nature, suggests we're getting better at building artificial brains that work more like real ones.