Training efficiency has become a significant factor for deep learning as the neural network models, and training data size grows. GPT-3 is an excellent example to show how critical training efficiency factor could be as it takes weeks of training with thousands of GPUs to demonstrate remarkable capabilities in few-shot learning.

To address this problem, the Google AI team introduce two families of neural networks for image recognition. First is EfficientNetV2, consisting of CNN (Convolutional neural networks) with a small-scale dataset for faster training efficiency such as ImageNet1k (with 1.28 million images). Second is a hybrid model called CoAtNet, which combines convolution and self-attention to achieve higher accuracy on large-scale datasets such as ImageNet21 (with 13 million images) and JFT (with billions of images). As per the research report by Google, EfficientNetV2 and CoAtNet both are 4 to 10 times faster while achieving state-of-the-art and 90.88% top-1 accuracy on the well-established ImageNet dataset.

We talk to Anil Mankar, Co-founder and Chief Development Officer at BrainChip, about artificial intelligence and neural-networking ICs.

[Editorial Note] This article has been planned and published as part of a series of articles related to open-source AI models and insight sharing by Sung Chang-yeop, who is a Developer Relations Engineer at Common Computer. This is the 11th article so far and its mainly about BlenderBot 2.0, a chatbot released by Facebook AI.

Improved autonomy

One of the main features of Spot is Autowalk, a system that enables the robot to record and repeat paths. An operator takes the robot through the path using the remote controller interface. The robot memorizes the path and can repeat when commanded to do so. Autowalk can be used for inspection missions in industrial facilities, mines, factories, and construction sites.

The new update improves Autowalk, reducing the need for human guidance and intervention. Robot operators can now edit Autowalk missions and add actions such as capturing images, reading indicators, or run third-party code. Spot also has been given better planning capabilities and can find the best path to perform target actions. Its pathfinding capacity has also been improved to adapt to changes in its inspection paths such as new obstacles. And it can be scheduled to carry out scheduled inspections without human supervision during off-hours.