MED-VAE aligns neural activity across subjects without shared stimuli, CCN 2026
somnirons · x · 2026-10-15
Papathanasiou, Jascha Achterberg et al. introduce MED-VAE, a Multi-Encoder-Decoder Variational Autoencoder that achieves cross-subject neural alignment without shared stimuli by anchoring representations to a common scaffold from a pretrained ANN.
Key results on the Natural Scenes Dataset:
- Common latent spaces with superior semantic organization, higher cross-subject alignment than standard methods, and robust generalization to held-out stimuli;
- Reconstruction back to each subject's neural space preserves stimulus-driven signal;
- The alignment directly enables cross-subject neural prediction, demonstrated via cross-subject image decoding.
Adding a new subject is cheap: freeze the shared space and train only a new subject encoder/decoder — 20% of the new subject's data recovers 98% of full-data alignment and 90% of the noise ceiling in self-reconstruction. Published at CCN 2026, code open-sourced.
More from Research
- Random Brainfuck programs on a 2D grid spontaneously evolve self-replicators — spikedoanz · 2026-10-15
- Classifier run on 42,000+ ICLR 2027 submissions finds over half nearly fully AI-assisted — nsaphra · 2026-10-15
- LLM4Code workshop at ICSE 2027 opens call for papers, due Nov 13, 2026 — LingmingZhang · 2026-10-15
- ELSA3D, a unified 3D generation and captioning model, hits SOTA at half the FLOPs, accepted at NeurIPS — CSProfKGD · 2026-10-15
- Moving ViT weights 1.3% without changing predictions scrambles representations, challenging Platonic hypothesis — neuroecology · 2026-10-15
- NeurIPS Spotlight: time-reversed LMs (TRLM) boost best-of-N reranking by up to 5% — LucaAmb · 2026-10-15