Research map / Architectures and efficiency / RNNs and normalization
Associative Memories via Predictive Coding
Tommaso Salvatori, Yuhang Song, Yujian Hong, Lei Sha, Simon Frieder, Zhenghua Xu, Rafal Bogacz, Thomas Lukasiewicz
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| Increase depth to 10 layers (ablation) | generative PCN · Tiny ImageNet 500 images, half pixels | images reconstructed: more than 98% | Improved |
| Replace autoencoder with generative PCN trained by predictive coding | associative memory · Tiny ImageNet 500 images, half pixels | images reconstructed (%): 72% -> 98% | Improved |
| Replace autoencoder with generative PCN | associative memory · Tiny ImageNet 500 images, 1/4 pixels | images reconstructed (%): 48% -> 74% | Improved |
| Replace modern Hopfield network with generative PCN | associative memory · CIFAR10 image retrieval | images retrieved: MHN at most 9 corrupted, 5 incomplete | Improved |
| Increase depth of PCN (ablation) | generative PCN · Flickr30k image-from-caption | images recovered | Improved |
| Increase width of PCN layers (ablation) | generative PCN · Flickr30k caption/image recovery | recovered count | No clear effect |
| Increase stored images from 50 to 500 (ablation) | shallow generative PCN · Tiny ImageNet partial reconstruction | images reconstructed | Hurt |
| Reduce provided pixel fraction (ablation) | generative PCN · Tiny ImageNet partial reconstruction | images reconstructed | Hurt |
| Store each image multiple times (3N, 5N) in MHN (ablation) | Modern Hopfield Network · CIFAR10 retrieval | images retrieved: 5 -> 4 (N vs 5N, 50 images, 1/2 pixels) | No clear effect |
Closest papers by meaning
- Sequential Memory with Temporal Predictive Coding (NeurIPS 2023 · similarity 0.89)
- Large Associative Memory Problem in Neurobiology and Machine Learning (ICLR 2021 · similarity 0.88)
- Factual recall in linear associative memories: sharp asymptotics and mechanistic insights (NeurIPS 2026 · similarity 0.88)
- Predictive Attractor Models (NeurIPS 2024 · similarity 0.87)
- Semantically-correlated memories in a dense associative model (ICML 2024 · similarity 0.86)
- Dense Associative Memory Through the Lens of Random Features (NeurIPS 2024 · similarity 0.86)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
