DSReg: multi-lab preprint disentangles predictive representations without decoders
jiqizhixin · x · 2026-11-07
Researchers from UIUC, Cold Spring Harbor Laboratory, Brown, Max Planck and ELLIS Tübingen released DSReg, a preprint on separating physical factors entangled in predictive representations without a decoder or reconstruction loss. Key points:
- Search over rotations: starting from a whitened, linearly identifiable representation, local regressions estimate how observations depend on coordinates; DSReg favors rotations making those dependencies sparse.
- Theoretical conditions: recovery up to sign and permutation requires distinct dependency footprints plus a functional no-cancellation assumption.
- Scalability: validated within the paper's setting.
More from Research
- Unverified claim: frontier labs skewing ICLR reviews with malicious reviews, scandal brewing — examachine · 2026-11-07
- Persona Atlas visualizes Qwen's experts and heads, exposing bizarre inner personas — Sauers_ · 2026-11-07
- Steering Qwen's layer-15 attention output makes it claim to be a group of people — Sauers_ · 2026-11-07
- Daily agent #19: an overnight sequencing QC agent for molecular biology workflows — nlarusstone · 2026-11-07
- Can we CRISPR herpes viruses out of our DNA? An open question — moultano · 2026-11-07
- Inductive Bio Wins 4th Straight OpenADMET Blind Challenge, Beating 270 Entrants — graceisford · 2026-11-07