Deep Learning for Symbolic Mathematics

Typically when you think of applications of deep learning neural networks, they include the types of things we've been discussing here, like image recognition, audio analysis and synthesis, language completion or translation, etc.  problems usually thought of from the standpoint of statistical modeling or manifold learning. You don't typically think of them for use when dealing with symbolic math problems.

I recently came up on an interesting paper that addresses this problem. Guillaume Lample and Francois Charton in Facebook's AI research group proposed an interesting way to restructure symbolic math expressions so that they could then be used to train RNN seq2seq neural nets. 

They restructured the symbolic expression problem so that they could use the same deep neural nets used for language translation.

What they did was restructure mathematical expressions as trees. Operators and functions are internal nodes, operands are children, and numbers, constant and variable are leaves of the tree. This clever restructuring of the representation is what allows the problem set to then be used to train a neural net.



There's a nice easy to read write up of this work in Quanta Magazine.

You can read the original paper on Deep Learning for Symbolic Mathematics here.

 So the take away message is that how you go about representing your data can be just as important as what kind of neural network you use to solve a particular problem.

The neural net for symbolic learning actually out performed Mathematica on some integration problems. Of course the usual criticisms were made of the neural net system, 'the neural net doesn't really understand the math', etc.

Some of you might remember John Koza's work in the 90's on evolving software using genetic programming. Software expressions were also modeled as tree structures and the tree structures were then evolved in this work.

 There were a number of papers around that time that also used tree structures to represent and manipulate mathematical expressions.  Karl Sims work on artificial evolution for computer graphics comes to mind.



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