Nvidia: researchers train AI to reward dogs for responding to commands

Colorado State University researchers Jason Stock and Tom Cavey published an article about an AI system that rewards dogs for doing tricks.

Computer science graduate students trained image classification networks to determine whether a dog is sitting, standing or lying down. If a dog responds to a command by adopting the correct posture, the machine delivers a treat.

Students used an Nvidia Jetson Edge AI platform to recognize tricks and treats in real time. Stock and Cavey see their prototype system as an aid to the dog trainer – he handles treats – or a way to educate dogs about better behavior at home.

“We demonstrate the potential for a future product to come out of this,” Stock said in a statement.

Fetching dog training data

The researchers needed images of dogs that exhibited the three specified postures. They found the Stanford Dogs data sets, with more than 20,000 images of various sizes depicting dogs in many positions. The images required pre-processing, so they wrote a program to help tag them quickly.

In an email to VentureBeat, Nvidia said: “It still doesn’t work remotely; it is currently for personal use. But that would be an easy setup to make it a remote system. You can think of it as a system, or IP, for licensing devices like Furbo. The researchers see many possible applications, but have not committed to anything yet. “

To refine the model, the researchers applied dog resources from ImageNet to enable transfer learning. Then they applied post-training and optimization techniques to increase speed and reduce the size of the model.

For optimizations, they took advantage of Nvidia’s Jetpack software development kit on Jetson, which is a lightweight AI platform for drones and other systems. It offers an easy way to get things up and running quickly and access the TensorRT and cuDNN libraries, Stock said. The Nvidia TensorRT optimization libraries offered “significant improvements in speed,” he added.

Taking advantage of the university’s computer system, Stock trained the model overnight on two 24 GB Nvidia RTX 6000 graphics processing units (GPUs).

Models implanted in Henry

The researchers tested their models on Henry, Cavey’s Australian shepherd. The model achieved accuracy of up to 92% in tests and demonstrated the ability to make inferences in fractions of a second at almost 40 frames per second.

Using the Jetson Nano, the system makes real-time decisions about the dog’s behavior and reinforces positive actions with a treat, transmitting a signal to release a reward.

“We looked at the Raspberry Pi and Coral, but none were suitable and the choice was obvious to use the Jetson Nano,” said Cavey.

Explanable AI helps provide transparency about the composition of neural networks. It is becoming more common in the financial services industry as a way of understanding fintech models. Stock and Cavey included the model interpretation in their article to provide explainable AI for the pet industry.

They do this with images from the videos that show the postural analysis. A set of images depends on GradCAM – a common technique for displaying where a convolutional neural network model is focused. Another set of images explains the model by touching Integrated Gradients, which helps to analyze pixels.

The researchers said it was important to create a reliable and ethical component of the AI ​​system for coaches and users in general. Otherwise, there is no way to explain the methodology, if it is questioned.

“We can explain what our model is doing, and it can be useful for certain stakeholders – otherwise, how can you support what your model is really learning?” Cavey said.