Neural net subnetworks are specifiable pure functions — but data entangles the spec
jd_pressman · x · 2026-11-09
Arguing whether a program could be broken into provably correct parts, jdpressman contends many subnetworks in a neural net are pure functions and in principle specifiable. The core obstacle: the spec of e.g. a VQGAN is intimately tied to the content of its training data distribution, resisting formalization.
More from Research
- Physical Intelligence to present π0.7 at CoRL oral with emergent robot capabilities — lucy_x_shi · 2026-11-09
- Fully open MolmoAct2 action reasoning model wins Best Paper at CoRL TAG — jmin__cho · 2026-11-09
- Deep dive coming: score centering for understanding and fixing RL training instability — cwolferesearch · 2026-11-09
- Reflection launches Beam: 501B-param agentic open model trained with AdamC optimizer — andersonbcdefg · 2026-11-09
- Systolic arrays make speculative decoding latency a staircase synced to array width — teortaxesTex · 2026-11-09
- MoRE: Mixture-of-Experts 3D World Model Handles Rigid, Plush and Cloth Objects for Real Robots — anand_bhattad · 2026-11-09