Brain-Inspired AI Learns Timing, Saves Energy

Most AI works like a giant calculator, processing numbers in steps. But brain-inspired AI, called spiking neural networks, sends pulses between artificial neurons — much like real brain cells. These pulses happen at exact times, and that timing matters. Until now, these networks could learn how strongly to connect neurons, but not the delays between them. DelRec fixes that by teaching the network both at once.

Being able to learn delays is a big deal. Think of it as learning not just who to talk to, but how long to wait before you reply. This extra skill helps the AI handle sequences — like understanding speech, predicting stock prices, or tracking motion in video — with better accuracy. In tests, DelRec beat other methods at chaotic time-series prediction and matched top performers on four other challenging tasks.

Even better, learning delays made the networks more efficient. They used less memory and consumed less energy, which is crucial for tiny chips in phones, smartwatches, and autonomous drones. The networks also became more robust, meaning they didn't get confused when the timing of their inputs was slightly off. That's like being able to understand someone who speaks a little irregularly.

The catch? This is research from a lab, not a product you can use today. Also, spiking networks run best on special neuromorphic hardware, which isn't in your phone yet. Still, this brings us closer to AI that thinks faster, costs less, and works well in the real world — and it gives scientists new insight into how our own brains coordinate timing.