Multimodal Neurons in Artificial Neural Networks

 There is a fascinating new paper out in distill by some folks at openAI titled 'MultiModal neurons in Artificial Neural Networks'.  Anyone familiar with research into visual perception has heard of 'grandmother neurons', or the more updated 'Halle Berry neuron'.

This distill paper analyzes an equivalent kind of phenomena taking place in the feature representation constructed internally int he recent CLIP mode.  We covered CLIP in recent HTC posts here and here.

You can read the distill publication online here.

You can also read all about this work on the openai blog here.

Our old friend Yannic Kilcher has put together a really great analysis of the paper you can watch below.


The CLIP model consists of 2 different components, a ResNet vision model, and a Transformer language model

Yannic points out that CLIP is probably using the text input and associated language model more than the visual input for how it is grouping things internally. A really good example of this is show below.

Now if you dig into the appendix, they talk about their new 'faceted feature visualization'.  We are going to dive into this more in another post, since i believe this is the key to getting the cool visualization images they are showing.

Here we propose a new feature visualization objective, faceted feature visualization, that allows us to steer the feature visualization towards a particular theme (e.g. text, logos, facial features, etc), defined by a collection of images. The procedure works as follows: first we collect examples of images in this theme, and train a linear probe on the lower layers of the model to discriminate between those images and generic natural images. 

We then do feature visualization by maximizing the penalized objective, f(g(x)) + w^T(g(x)\odot \nabla f(g(x))), where w are the weights of that linear probe, and f \circ g is the original feature visualization objective, composed of two functions, g, which takes the input into an intermediate activations and f which takes those intermediate activations into the final objective.



Comments

Post a Comment

Popular posts from this blog

CycleGAN: a GAN architecture for learning unpaired image to image transformations

Image
 Like the Doublemint Twins touting the joys of Doublemint gum, 2 GANs are surely better than 1 GAN.  Especially if we package them together inside of one meta GAN module.  And this is exactly what the CycleGAN architecture does. Have you ever harbored dark secrets of turning a horse into a zebra?  The CycleGAN was developed to do just that. Learn how to turn a horse into a zebra.  And more. Now right away you can notice a difference between the image to image transformation GAN architectures we've been discussing over the last few posts.  Those last few posts described systems that learn from a database of matched input-output image pairs. And if your goal is to turn an edge representation into a nicely filled in continuous tone image, it's easy to build your database of matched input-output image pairs that your GAN system can then learn off of.  Take a continuous tone image (which will be the output of the database pair entry), then run it through an edge detector algorithm. 
Read more

Pix2Pix: a GAN architecture for image to image transformation

Image
 I thought following up yesterday's TraVelGAN post with a Pix2Pix GAN post would be useful to compare what is going on in the 2 architectures.  Two different approaches to the same problem. I stole this Pix2Pix Overview slide below from an excellent deeplearning.ai GAN course (note that they borrowed it from the original paper) because it gives you a good feel for what is going on inside of the Pix2Pix architecture.    Note how the Generator part is very much like an auto-encoder architecture, but rebuilt using the U-Net architecture features (based on skip-connections) that fastai has been discussing in their courses for several years before it became more widely known to the deep learning community at large  (and which originally came from an obscure medical image segmentation paper) . So the Generator in this Pix2Pix GAN is really pretty sophisticated, consisting of a whole image to image auto-encoder network with U-Net skip connections to generate better image quality at highe
Read more

Smart Fabrics

Image
"What makes smart fabrics revolutionary is that they have the ability to do many things that traditional fabrics cannot, including communicate, transform, conduct energy and even grow."  This is a quote from Rebeccah Pailes-Friedman, who is a professor at Pratt with a research focus on developing wearable technology and smart textiles. Enhancing textiles can be viewed through different lens.  Enhancements can be aesthetic, or performance boosting, or durability boosting. 
Read more