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

Colorado State University researchers Jason Stock and Tom Cavey have published a paper on the AI ​​system that rewards dogs for doing tricks.

Computer science grade students have trained an image classification network to determine if a dog is sitting, standing, or sleeping. If a dog responds to commands by adopting a proper posture, the machine treats.

Students used the Nvidia Jetson Edge AI platform for real-time trick recognition and behavior. Stock and Cave see their prototype system as the help of a dog trainer – he takes care of eating things – or a way for school dogs to behave well at home.

“We have shown the potential for future production to come out of this,” Stoke said in a statement.

Obtaining dog training data

Researchers needed images of a dog that exhibited three clear postures. They found Stanford Dogs datasets, which contained more than 20,000 images of various sizes, depicting dogs in many situations. Images needed preprocessing so they wrote a program to help label them quickly.

In an email to VentureBit, Nvidia said, “It doesn’t work remotely yet; He is currently in person. But it will be a simple setup to create a remote system. You might think of it as a system or IP for licensing devices like Farbo. Researchers have looked at many potential applications, but have not yet committed to anything. “

To improve the model, the researchers applied the features of dogs from Imagenet to enable transfer learning. Next, they applied post-training and optimization techniques to increase speed and reduce model size.

For Iz optimization, they tapped on Jetson in Nvidia’s JetPacks software software development kit, a lightweight AI platform for drones and other systems. It provides an easy way to quickly get things up and running and to access TenserRT and QDNN libraries, Stock said. He added that Nvidia TensorRT optimization libraries offered “significant improvements in speed”.

Tapping into the university’s computing system, Stoke trained the model overnight on two 24GB Nvidia RTX6000 graphics processing units (GPUs).

Model on Henry

Researchers tested their models on Henry, a Straw Shepherd from Cavani. Tests by the model achieved up to 92% accuracy and demonstrated the ability to make split-second predictions at about 40 frames per second.

Using the Jetson Nano, the system makes real-time decisions about dog behavior and reinforces positive actions with treatment, transmitting the signal to release the prize.

“We looked at raspberry pie and coral, but both weren’t enough, and it was obvious for us to use the Jetson Nano,” Cave said.

Detailed A.I. Helps to provide transparency about the creation of neural networks. It has become more common in the financial services industry as a way to understand finatech models. Stock and Kave included model interpretations in their paper to provide an explanatory explanation for the pet industry.

They do this with video images that show currency analysis. A set of images relies on a gradcam – a common technique for displaying where the convoluted neural network model is centered. Another set of images illustrates the model by tapping into integrated gradients, helping to analyze pixels.

The researchers said it was important for trainers and general users to create a reliable and ethical component of the AI ​​system. Otherwise, there is no way to explain the method, whether it comes into question.

“We can explain what our model does, and it can be helpful to certain stakeholders – otherwise how can you back up what your model is learning?” Kave said.