Week of October 12, 2026
2026-W42
Week 2026-W42: 20 curated papers with 20 free plain-English explainers.
Direct Quantification of Molecular Helicity-Dependent Effective Magnetic Fields in Chiral Polypeptide/Ferromagnet Heterostructures
Helical peptides can still nudge a nearby magnet through an insulating spacer, and the push flips with molecular handedness. The authors measure that enantiome…
Read free explainer →
A Balanced Data Diet: Addressing the Exploration Bottleneck in Mega-Scale RL for Robot Control
Million-environment robot RL wastes experience on tasks the policy already owns or cannot yet try. Success Guided Sampling keeps training on the capability fro…
Read free explainer →
Measure Now, Mitigate Later: Virtual Error Cancellation for Logical Quantum Circuits
Residual logical errors still bias early fault-tolerant quantum algorithms. Virtual error cancellation uses only syndrome records in classical post-processing…
Read free explainer →
Generative Neural Retargeting for Human-to-Robot Dexterous Manipulation
Human videos are abundant; robot-feasible motions are not. Generative Neural Retargeting learns a flow-matching model of dynamically feasible trajectories so e…
Read free explainer →
Dex-One2Many: Learning Dexterous Manipulation from a Single Human Demonstration
One human video is a poor robot policy if you only clone the motion. Dex-One2Many turns that video into sequential scene graphs that guide simulated RL — diver…
Read free explainer →
VersaCamVLA: Camera-Configurable VLA Policies for Robotic Manipulation
Vision-language-action robot policies usually assume the cameras they were trained with. VersaCamVLA maps any number of posed RGB views into fixed-size scene t…
Read free explainer →
VioLA: Learning Generalist Humanoid Control Policies from Human Data
Humanoid generalists usually predict joints and then need teleop fine-tuning per task. VioLA predicts body and hand motion latents that pretrained controllers…
Read free explainer →
SciExam for ENSO: Can AI Agents Build Climate Models?
Most AI-science tests grade against a known answer. SciExam for ENSO hides the grade: agents get real observations and six hours to build a low-order ENSO mode…
Read free explainer →
Entanglement entropy and magic of ZX-diagrams
Entanglement and magic usually need expensive tensor work to read off a big circuit. This paper estimates both directly from ZX-diagrams: a flow-based normal f…
Read free explainer →
Quantum Co-Design of Inhomogeneous Many-Body Neutrino Fast Flavor Transformation
Dense-neutrino flavor evolution is a many-body problem that explodes with particle number and spatial structure. QCNO maps an inhomogeneous forward-scattering…
Read free explainer →
WorldGuide: Goal-Directed Video World Model for Procedural Task Execution
Pretty video generators do not finish a recipe: they must choose the next action from what they just made, do it, and stop. WorldGuide closes that loop with a…
Read free explainer →
Three-dimensional Lagrangian ecosystems: carbon dynamics and potential for artificial fertilization
A fertilized ocean patch is stretched and diluted while it blooms, so the carbon budget is easy to miscount. This idealized 3D Lagrangian model reproduces an a…
Read free explainer →
Emergent gold-induced interfacial magnetism in monolayer FePS$_3$
Gold is not a passive sticky note for a 2D antiferromagnet. On FePS3, an ultrathin gold film induces a sizable interfacial moment: monolayer XMCD triples versu…
Read free explainer →
Toward Reliable Patient-Specific Aortic Strain Mapping from 4D CTA: Validation, Spectral Structure, and Clinical Potential
A patient’s ascending-aorta stretch can be mapped from everyday 4D CT angiography — and the map stays stable across human and neural-net segmentations when you…
Read free explainer →
Unlocking the Regulatory Genome by ARGUS: An Evidence-Constrained Agentic Framework for Interpreting Single Nucleotide Variants
Most disease-linked DNA spelling changes sit in regulatory DNA, and unconstrained language models happily hallucinate transcription-factor stories. ARGUS keeps…
Read free explainer →
Data Representation Matters: Optimizing Machine Learning for X-ray Source Classification by Leveraging Spatio-Spectral Information
Faint extended X-ray sources are hard to tell apart when you flatten the photons first. A 3D CNN that eats XMM-Newton event cubes — space plus energy together…
Read free explainer →
Variational Derivation of the Weak Form for Viscous Flows Involving Free Surfaces with Implications for Geodynamic Simulations
Free-surface “drunken sailor” instabilities force tiny time steps in geodynamic Stokes solvers, and the usual gravity-stabilization terms were mostly heuristic…
Read free explainer →
A neural characteristic mapping method: Lagrangian PINNs based on flow maps for transport-dominated problems
When a solution grows sharp fronts, a PINN that tries to fit the solution itself smears them. These Lagrangian PINNs fit the flow map in short time chunks inst…
Read free explainer →
Disk Structure May Determine AGN Variability and Explain the Accretion Disk Size Problem: No Broad Line Region Required
A single 3D radiation-MHD AGN disk that is thin on top and thick below emits two different skies: the thin side looks like classic X-ray-driven variability, wh…
Read free explainer →
Kinetic wave activity and proton heating in 3D hybrid simulations of decaying balanced and imbalanced Alfvénic turbulence
The same large-scale Alfvénic amplitude can heat protons very differently once you change cross-helicity. Balanced decaying turbulence dissipates harder and he…
Read free explainer →