Skip to main content
The TWIML AI Podcast

Why Models Are AI’s Next Training Dataset with Damian Borth - #772

47 min episode · 2 min read
·
Damian Borth

Episode

47 min

Read time

2 min

Topics

Startups, Artificial Intelligence, Software Development

AI-Generated Summary

Key Takeaways

  • Weight-Space Autoencoding: Train an encoder-decoder on populations of neural networks to learn a compressed latent manifold of weights. From this manifold, you can predict model accuracy without test data and generate new model weights. Early results on small homogeneous model zoos achieved R-squared accuracy prediction outperforming handcrafted feature baselines from Google researchers.
  • Scaling via Windowed Tokenization: Rather than encoding entire model weights at once, process weights as windowed sequences with position encodings indicating layer, block, and absolute position. This approach, developed with Michael Mahoney at UC Berkeley, enables the method to scale beyond toy networks to ResNet-scale architectures and beyond, unlocking training on diverse open-weight repositories.
  • GPU Hour Reduction via Knowledge Transfer: By using an ImageNet-trained ViT as an encoder prompt to generate remote sensing foundation models, Borth's lab matches the performance of TelIFM — trained for 12,000 GPU hours — using only 350 GPU hours, a 30x reduction. This demonstrates cross-domain knowledge transfer through weight generation rather than data-based pretraining.
  • Dataset Prompting for Privacy-Preserving Model Generation: An in-review paper proposes replacing model prompts with dataset embeddings — a single aggregated vector representing a private dataset — to generate task-specific weights without exposing raw data. This mirrors CLIP's aligned text-image space, enabling regulated institutions like central banks to receive custom model weights without sharing proprietary training data.
  • Hugging Face as Training Corpus: Download open-weight models from Hugging Face, strip weights into flat sequences, apply normalization and diversity-based filtering, and train a backbone on roughly 2,000 quality-filtered models out of 20,000 candidates. The resulting model generates weights across architectures including ViTs, ResNets, and EfficientNets, and can cross domains — producing a GPT-2-scale language model initialized from computer vision weights.

What It Covers

Damian Borth, professor at University of St. Gallen, presents weight-space learning — a framework treating trained neural network weights as training data for new models. Starting in 2021, his lab built autoencoder architectures that analyze, compress, and generate neural network weights, potentially replacing pretraining by sampling models on demand from a foundation model of neural networks.

Key Questions Answered

  • Weight-Space Autoencoding: Train an encoder-decoder on populations of neural networks to learn a compressed latent manifold of weights. From this manifold, you can predict model accuracy without test data and generate new model weights. Early results on small homogeneous model zoos achieved R-squared accuracy prediction outperforming handcrafted feature baselines from Google researchers.
  • Scaling via Windowed Tokenization: Rather than encoding entire model weights at once, process weights as windowed sequences with position encodings indicating layer, block, and absolute position. This approach, developed with Michael Mahoney at UC Berkeley, enables the method to scale beyond toy networks to ResNet-scale architectures and beyond, unlocking training on diverse open-weight repositories.
  • GPU Hour Reduction via Knowledge Transfer: By using an ImageNet-trained ViT as an encoder prompt to generate remote sensing foundation models, Borth's lab matches the performance of TelIFM — trained for 12,000 GPU hours — using only 350 GPU hours, a 30x reduction. This demonstrates cross-domain knowledge transfer through weight generation rather than data-based pretraining.
  • Dataset Prompting for Privacy-Preserving Model Generation: An in-review paper proposes replacing model prompts with dataset embeddings — a single aggregated vector representing a private dataset — to generate task-specific weights without exposing raw data. This mirrors CLIP's aligned text-image space, enabling regulated institutions like central banks to receive custom model weights without sharing proprietary training data.
  • Hugging Face as Training Corpus: Download open-weight models from Hugging Face, strip weights into flat sequences, apply normalization and diversity-based filtering, and train a backbone on roughly 2,000 quality-filtered models out of 20,000 candidates. The resulting model generates weights across architectures including ViTs, ResNets, and EfficientNets, and can cross domains — producing a GPT-2-scale language model initialized from computer vision weights.

Notable Moment

When Borth's team first generated neural network weights using their autoencoder, the mean squared error appeared low — suggesting success — but loading those weights into an actual network completely destroyed its performance. The culprit was MSE averaging away the high-frequency weight details that determine whether a network functions at all.

Know someone who'd find this useful?

Episode Transcript

I wanna send a big thanks to Blitsy for supporting the podcast and sponsoring this episode. Want to accelerate software development velocity by five x? You need Blitsy, which brings autonomous software engineering to your enterprise. Blitzy starts by reverse engineering your code base, building a dynamic understanding of your entire application ecosystem. Your engineers simply declare intent. And once approved, Blitzy autonomously executes entire software epics, delivering validated end to end tested code. More than 80% of the work completed in a single run. Blitsy is not just generating code. It's developing software at the speed of compute. Experience Blitsy firsthand at blitsy.com/twimal. That's blitzy.com/twiml. One of the biggest questions facing AI today is how foundation models keep improving as high quality training data becomes harder to find. Some researchers are betting on synthetic data, others on inference time reasoning. Today's guest has his chips on something very different. Every trained model represents thousands or even millions of GPU hours spent discovering what works. Instead of treating those weights just as the end of the training process, what if they're also the beginning of the next one? Damien Borth, professor of AI and machine learning at the University of Saint Gallen, sees train models themselves as data. Data that can be learned from, analyzed, and even used to generate entirely new models. When I asked him to explain the idea behind weights based learning, here's where he started. So we basically thought about this very simple idea. What happens actually if we take the weights of trained neural networks as the input to train a neural network to understand these weights that we have out there much, much better. Thinking about that that you can treat the weights as a input modality gives you suddenly this opportunity of of can we be much, much faster in creating new weights for particular task, or can we be much more precise in analyzing weights when somebody gives me a new electrode that I'm not knowledgeable about and I never saw before? I'm Sam Charrington, and this is the TwiML AI podcast. For over a decade, I've been exploring the ideas and innovation shaping the future of AI through conversations like this one that help you understand what's real, what's next, and what matters. Let's jump in. We started, like, in 02/2021, the work on what we call weightspace learning. And weightspace learning is a quite interesting way at looking at machine learning in general. That's currently the major topic. We also do a little bit of work in remote sensing and then, representation learning on tabular data. We're now walking and combining everything together to, you know, focus more on weight based learning, which I think is a really interesting way forward, solves a couple of problems that the community currently encounters, and, started from a very very esoteric idea to something that works surprisingly well. You know, we think about ways as the product of training a model, and, you …

Get the full transcript (8,099 words) + summary by email — free

One-time email with the complete transcript and AI summary of this episode. No account needed.

One email, no spam. We’ll also show you what SignalCast does.

Browse all The TWIML AI Podcast transcripts →

You just read a 3-minute summary of a 44-minute episode.

Get The TWIML AI Podcast summarized like this every Monday — plus up to 2 more podcasts, free.

Pick Your Podcasts — Free
Part of this week's recap (Jul 27 – Aug 2)

Keep Reading

Books, tools, and gear mentioned in this episode

SignalCast may earn commission on purchases via these links.

company

  • Download open-weight models from Hugging Face, strip weights into flat sequences, apply normalization and diversity-based filtering, and train a backbone on roughly 2,000 quality-filtered models out of 20,000 candidates.

More from The TWIML AI Podcast

We summarize every new episode. Want them in your inbox?

Similar Episodes

Related episodes from other podcasts

Explore Related Topics

This podcast is featured in Best AI Podcasts (2026) — ranked and reviewed with AI summaries.

Read this week's Startups & Product Podcast Insights — cross-podcast analysis updated weekly.

You're clearly into The TWIML AI Podcast.

Every Monday, we deliver AI summaries of the latest episodes from The TWIML AI Podcast and 192+ other podcasts. Free for one show.

Start My Monday Digest

No credit card · Unsubscribe anytime