Nvidia: researchers train AI to reward dogs for taking commands

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

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

The students used an Nvidia Jetson edge AI platform for real-time recognition and treats. Stock and Cavey see their prototype system as an aid to the dog trainer – he manages the treats – or as a way to teach dogs better behavior at home.

“We have demonstrated the potential for a future product to come out,” Stock said in a statement.

Retrieval of dog training data

The researchers needed images of dogs exhibiting the three specified postures. They found the Stanford Dogs datasets, with more than 20,000 images of varying sizes showing dogs in many positions. The images needed preprocessing, so they wrote a program to label them quickly.

In an email to VentureBeat, Nvidia said, “It doesn’t work remotely yet; it is currently for in person. But that would be an easy setup to make it a remote system. You could think of it as a system, or an IP address, for licensing devices like the Furbo. Researchers see many possible applications but have yet to commit to anything. “

To refine the model, the researchers applied characteristics of ImageNet’s dogs to enable transfer learning. Then they applied post-training and optimization techniques to increase the speed and reduce the size of the model.

For optimizations, they leveraged Nvidia’s Jetpack SDK on Jetson, which is a lightweight artificial intelligence platform for drones and other systems. It offers a simple way to get things up and running quickly and access the TensorRT and cuDNN libraries, Stock said. Nvidia TensorRT optimization libraries offered “significant improvements in speed,” he added.

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

Models deployed on Henry

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

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

“We looked at Raspberry Pi and Coral, but neither was adequate, and the choice was obvious to us to use Jetson Nano,” Cavey said.

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

They do this with images from the videos that show the analysis of posture. A set of images relies on GradCAM – a common technique for showing where a convolutional neural network model is focused. Another set of images explains the model by tapping Built-in gradients, which helps analyze pixels.

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

“We can explain what our model does, and that might be useful to some stakeholders. How else can you save what your model is actually learning? ” Cavey said.