Nvidia: Researchers Train AI to Reward Dogs for Response to Commands

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

The computer science students trained image classification networks to determine whether a dog is sitting, standing or lying down. When a dog responds to a command by taking the right attitude, the machine gives a treat.

The students used an Nvidia Jetson edge AI platform to recognize and treat tricks. Stock and Cavey consider their prototype system as an aid to a dog trainer – it handles the treats – or a way to handle school dogs with better behavior at home.

“We have shown the potential for a future product from this,” Stock said in a statement.

Catch up on dog training

The researchers need dog images that show the three specific attitudes. They found the Stanford Dogs datasets, with more than 20,000 images of different sizes depicting dogs in many positions. The images had to be pre-processed so that they could write a program to name them quickly.

In an email to VentureBeat, Nvidia said: ‘It does not work remotely yet; it is currently for personal. But it would be an easy setup to make it a remote system. You can consider it a system, or IP, to license devices like the Furbo. The researchers see many possible applications, but have not yet committed anything to it. ”

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

For optimization, they used Nvidia’s Jetpack software development package on Jetson, which is a lightweight AI platform for drones and other systems. It provides an easy way to get things up and running quickly and access the TensorRT and cuDNN libraries. Nvidia TensorRT optimization libraries offer a significant improvement in speed, ‘he added.

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

Deploy models on Henry

The researchers tested their models on Henry, Cavey’s Australian Shepherd. The model achieves an accuracy of up to 92% in tests and shows the ability to make a split-second distraction at almost 40 frames per second.

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

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

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

They do this with images from the videos that show the attitude analysis. One set of images depends on GradCAM – a common technique for showing where an evolutionary neural network model is focused. Another set of images explains the model by tapping on Integrated Course, which helps to analyze pixels.

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

“We can explain what our model does, and it can be helpful to certain stakeholders – otherwise you can make a backup of what your model is really learning?” Cavey said.