Mastodawn

🔬 Excited to present our latest research at the #MAIN2025 conference today!
🔗 https://laurentperrinet.github.io/talk/2025-12-12-main/

👁️ What if CNNs could see like humans? Our new work shows how foveated vision—concentrating processing at gaze center—makes networks more robust to perturbations & great at localization. Inspired by human vision's architecture (high-resolution foveal center, low-resolution periphery), we embedded this retinotopic transformation into CNN architectures, allowing to actively scan the image. This gives literally a new look to #ConvNets !

📄 Paper: "Foveated Retinotopy Improves Classification and Localization in CNNs"
🔗 https://laurentperrinet.github.io/publication/jeremie-25/

#DeepLearning #ComputerVision #AI #Research #NeuralNetworks #NeuroAI #OpenScience I love #Montreal

George Musser Oct 11, 2024

Convolutional neural networks—the machine-learning systems routinely used for image recognition—are inherently limited, according to a new study by @lamaral and colleagues. #ConvNets fail on data above a certain threshold of complexity. They lean too heavily on shortcuts—spurious correlations that don’t generalize—and on localized features such as texture, to the neglect of the overall scene. h/t @manlius https://doi.org/10.1063/5.0213905 #AI

Computational experiments with cellular-automata generated images reveal intrinsic limitations of convolutional neural networks on pattern recognition tasks

The extraordinary success of convolutional neural networks (CNNs) in various computer vision tasks has revitalized the field of artificial intelligence. The out

AIP Publishing

Show thread

Nikos Tsardakas Renhuldt Apr 23, 2023

Oh, and obviously I've done a ton of work on #oats and #GenomeAnnotation and #GeneMapping, and looked at at bunch of different types of data from #PlantGenomes - this ended up in a Nature publication on the oat genome with me as one of the first authors: https://www.nature.com/articles/s41586-022-04732-y. #GenomicPrediction and #PlantBreeding are things I'm interested in. I've also dabbled in #DeepLearning, mainly #ConvNets for protein contact prediction and image segmentation.

The mosaic oat genome gives insights into a uniquely healthy cereal crop - Nature

Assembly of the hexaploid oat genome and its diploid and tetraploid relatives clarifies the evolutionary history of oat and allows mapping of genes for agronomic traits.

Nature

Brandon Rohrer Feb 20, 2023

If you want to trick an AI built on #ConvNets (like Alpha Go) remember that it can see local patterns and textures way better than global structure. You can misdirect with the one while attacking with the other.

h/t @0xabad1dea

https://infosec.exchange/@0xabad1dea/109892756275113724

abadidea (@[email protected])

the overwhelming moral of the story recently is that AI is terrifyingly dangerous but also hilariously easily tricked https://arstechnica.com/information-technology/2023/02/man-beats-machine-at-go-in-human-victory-over-ai/?utm_brand=arstechnica

Infosec Exchange