An autoencoder trained on a homogeneous population of neural networks can predict the accuracy, generalization gap, and activation function of unseen models from their weights alone, without requiring test data.
Borth describes the first proof-of-concept for weight-space learning: an autoencoder compressed weight sequences from thousands of small models into a latent space, and a linear regression head on that space could predict accuracy and other properties of models never seen during training. ✦ AI generated
Damian Borth · The TWIML AI Podcast · 2026-07-27 · original ↗
starts at this moment · 5:44
“You mentioned you started this effort in 2021. Where did you start from and then we'll work our way towards like where we are now with weights based learning.”
we took you know an autoenccoder we have an encoder and decoder we learned the autoenccoder with or train the auto with reconstruction loss a contrastive loss in the middle and then we take we took only the encoder and unknown neural networks that we encoded into the latent space and this embeddings we put into a you know simple you know regression like a linear regression head to predict the accuracy. So you give me a neural network. I never saw this neural network and the idea was can I predict the accuracy of this neural network. Can I test the neural network without the use of test data? Right? Obviously this worked on this really small neural networks only in an homogeneous we call this a model zoo a population of models. So they were all trained on the same data, same architecture, first you know small steps towards this goal. But we could extract this information about the accuracy, the performance, the generalization gap, what kind of activation function was used. And if you would plot this latent spaces, you would see different initializations and how they evolve because we had, you know, 50 epo per model training and 10,000 models. So little trajectories were visible.
verbatim transcript · starts at 5:44
5:442021. Where did you start from and then we'll work our way towards like where we are now with weights based learning. >> Originally this idea came 2020 and we got the first paper published 21 and the idea was very simple. Can we fingerprint a neural network or version a neural network like we can do with software? Right? In software you have you can do a diff right you have a 1 million lines of
6:06code somebody changes something and then you know you do you do a diff you know exactly where the difference is so can we do this with neural networks problem with neural networks is if you do one update of weights during training every weight is a little bit different so there's not much you can extract from this >> right they're fairly unstable locally >> if everything is different nothing is
6:29different right so we were thinking about can we find a space where this neural networks the weights the models of those neural networks are little bit you know compressed and and more understandable and we started to to think about that and in parallel there was this really amazing work the first work I you know from Thomasa and Daniel Kaisers and colleagues from Google Turk and they developed a paper that used
6:58weights as input um you know did some statistical features and crafted features and then predicted the accuracy of those ways and another paper by Is that predicted the generalization gap so people started to use weights to extract information and they were all handcrafted features so I was thinking about a handcrafted features that so traditional machine learning so why not end toend learning so then we developed our idea of you
7:24know autoenccoding sequences of weights into a lower dimension space and then you know reconstructing it. that if we're able to do this from a population of neural networks then you know we can maybe learn a lower dimension manifold that you know populates actually you know where where the neural networks populate that manifold and maybe you know it this manifold encodes information about accuracy I know what training data was
7:51used what training fraction learning rate and all this you know latent generating factors and we started working on this and the first paper was really like well it it works we can compress neuronet networks like very small tiny new toy to toy examples, right? Embarrassingly small like thousands 10 thousands of parameters but it was working and we could predict the accuracies and and it was very nice to
8:16see that and then you know we got published. >> What exactly were you able to predict? So we we took you know an autoenccoder we have an encoder and decoder we learned the autoenccoder with or train the auto with reconstruction loss a contrastive loss in the middle and then we take we took only the encoder and unknown neural networks that we encoded into the latent space and this
8:40embeddings we put into a you know simple you know regression like a linear regression head to predict the accuracy. So you give me a neural network. I never saw this neural network and the idea was can I predict the accuracy of this neural network. Can I test the neural network without the use of test data? Right? Obviously this worked on this really small neural networks only in an