Did a bit of reading tonight. Linky linky:

• What I learned from competing against a ConvNet on ImageNet
• A Large-Scale Car Dataset for Fine-Grained Categorization and Verification
• CNNs Architectures: LeNet, AlexNet, VGG, GoogLeNet, ResNet and more

and finally, a meatier paper that I'm long overdue to read, but have only just started tonight:

• Going deeper with convolutions

I imagine that this one just has to be a seminal paper in deep learning. As described in some of the above links, in 2014 the annual "ImageNet Large Scale Visual Recognition Challenge" (ILSVRC) competition saw a new entry that managed to cut the error rate of the previous year's winner in half. It used a new network architecture called Inception, and the classifier network trained for the competition was called GoogLeNet. The paper Going deeper with convolutions describes that architecture.

More detail once I've actually read (and digested, as much as practical) the whole paper.

Finally, since this post is mostly about linking useful resources, I just want to say that Karoly Zsolani's Two Minute Papers series on YouTube has been reliably fantastic. And since I have a link post in which Inception is the star, clearly I should link to this particular video, thus spawning a whole new round of reading and linking.