Cat Dragged in Another Dead Rodent? AI-powered Door Flap Comes to the Rescue

Letting your precious cat roam free can often be a risky proposition. Every now and then their primal instincts bring out their deepest murderous intents, which inevitably leads to a piece of dead, or worse, half-dead meat package to your doorstep. Now, finding this dreadful crime scene in the middle of the night isn’t particularly enjoyable, luckily, Ben Hamm, a product manager at Amazon, has overcome this issue with machine learning software. For Hamm, it was his cat Metric who lead him to develop a mechanism to prevent this occurrence.

In a presentation he gave at Ignite Seattle, he explains that he first attached an Arduino to a lock for one of those conventional cat flaps. Next, none other than Amazon’s DeepLens camera was used to send images of Metric to Amazon’s machine learning platform, SageMaker. If the cat was entering with, for example, the strangled corpse of a bird, the AI would hopefully detect this and lock the cat flap for fifteen minutes.

But how was his algorithm able to determine when to lock the door and when not to? A technique called Supervised learning was used to train a computer so that it could apply labels to new images on its own. To do that, Hamm had to painstakingly sort 23,000 images of his cat into the distinct categories so that the computer could train itself to recognize patterns in the images which it could then apply to new images of Metric when it was actually put into use.

Hamm used four categories for the computer to be trained on: no cat, cat not on approach, cat on approach, and cat with prey.

Hamm used four categories for the computer to be trained on: no cat, cat not on approach, cat on approach, and cat with prey.

However, there are some limitations to Hamm’s system. For instance, Supervised learning requires data – a lot of it – and it’s a pretty long process. Moreover, the computer was only trained using images of one specific cat, therefore it can only detect patterns for that specific cat.

Nonetheless, over five weeks Metric was only unfairly locked out in a single instance and the AI failed to stop the cat coming in with its prey a mere one out of seven times.

So what do you think? Although this experiment was pretty successful it still has its downs. Though the very fact that AI is continuously becoming an integral part of people’s everyday life is fascinating, to say the least.

Pavan Ayyar