The Morning Preprint
Thursday 22 October
Showing the most recent batch — Thursday 22 October. No papers have been ingested today yet.
1144 papers ingested by the estate's stations that day — about 11 minutes to read. The counts and the chart are the estate's own ingest, not arXiv's submission figures.
This day's harvest looks short: 1144 papers against a busiest day of 2416 in the week shown below. Weekends and a broken fetch look the same in the chart, and the chart counts the estate's ingest, not arXiv's.
The 60-second version
Five things from this day's batch, best first — everything below the fold expands on them.
Name prediction improves racial estimation accuracy
Using text embeddings, researchers improved race probability estimation by looking at full names, especially for Hispanic and Asian voters. arXiv 2604.22555 arXiv 2604.19444
LLM confidence calibration stabilizes model trust
The BASC method allows models to recalibrate scores using only two responses, providing a distribution-free guarantee across tasks. arXiv 2604.19444
Harmony framework enhances protein docking pose accuracy
The Harmony framework parameterizes ligand score fields using learned harmonic potentials on the circle, improving pose accuracy on PDBBind. arXiv 2608.14866
VAPOLA tracks polarization changes in active galactic nuclei
The VAPOLA survey mapped polarization and rotation properties of AGNs and Sagittarius A* over five years, showing daily changes in linear polarization. arXiv 2608.14900
Test-Time Optimization guides autonomous driving decisions
TOAD treats planning as a search problem at test time, using a learned reward function to actively search for the best trajectory. arXiv 2608.10413
Across the stations
AI Radio
The day's work centered on improving how large language models handle real-world complexities, with research into name prediction for race estimation being a key focus through the eBISG approach. This method uses text embeddings to estimate race probabilities for names not in census data, showing improvement with full-name embedding.
Other papers explored diverse areas; some focused on image generation, like training-free camouflage synthesis with FreeCam, while others tackled 3D scene reasoning using ReRef-3D and motion forecasting in MolmoMotion.
Finally, studies addressed model reliability and efficiency; Binary Agreement for Self-consistency Calibration (BASC) improves LLM confidence calibration, while mechanisms like PISA reduce the quadratic complexity of diffusion transformers.
Astrophysics Radio
The work on VAPOLA focuses on mapping the polarization and rotation properties of active galactic nuclei and Sagittarius A star across five years of ALMA observations. This study shows that while total intensity remains stable, the linear polarization fraction, electric vector position angle, and Faraday rotation measure often change significantly day to day.
Other papers contribute by exploring different cosmic phenomena. The work on Don't miss the forest for the trees develops an effective field theory for Lyman alpha forest fluctuations, successfully matching hydrodynamic simulation data up to k= 3 h Mpc-1 at z= 2.8 with sub percent accuracy. The work on An empirical derivation of the average dust attenuation law uses JWST spectroscopy to constrain the average dust attenuation law for massive star-forming galaxies between redshifts of two and seven, yielding an R V value of approximately 3.98.
These findings suggest that polarization variability in AGNs is linked to jet physics or accretion flow dynamics, while cosmology is being probed by deriving EFTs for the Lyman alpha forest and constraining dust properties at high redshift. The question remains whether the observed polarization changes in VAPOLA sources are primarily driven by local jet activity or processes within the accretion flow of Sgr A star.
Genomics Radio
The day's work on Genomics Radio focused heavily on improving flexible protein-ligand docking by addressing how diffusion models handle torsional updates. The main contribution is the Harmony framework, which parameterizes ligand and side-chain score fields using learned harmonic potentials on the circle, making periodicity explicit in a way that previous models lacked. This approach shows that aligning score parameterization with diffusion geometry effectively enhances pose accuracy and pocket reconstruction on benchmarks like PDBBind.
Other papers contributed by focusing on different aspects of molecular dynamics and scoring functions. One work investigates how the variance-exploding diffusion process affects the Fourier basis when applied to torsional motion, which is key to Harmony's method. Another study addresses physical validity in complex generation using PoseBusters, showing that Harmony improves this aspect of complex construction. These works collectively explore how modeling rotational motion through harmonic potentials can guide flexible docking simulations more accurately than generic methods.
The day's findings suggest that explicitly modeling periodicity via learned harmonic potentials is a simple and effective way to boost performance in flexible docking tasks like those shown by. The question remaining is how this harmonic bias scales when moving to larger, more complex protein systems beyond the case studies on EBNA1 and KRAS G12D mentioned in that work.
Quantum Radio
The day at Quantum Radio focused on protecting quantum machine learning pipelines from channel substitution attacks, a critical issue as QML moves to the cloud. The work called QML-PipeGuard shows how malicious providers can substitute quantum channels while still passing basic accuracy tests, and it detects this at runtime using measurement statistics.
Other submissions explored different areas; for instance, Tensor computing application programming interface introduces a lightweight API to make tensor network computations more portable across different hardware and software platforms. Additionally, research on Electrical detection of spin-flip transition in metal/ heterostructure observed step-like changes in magnetoresistance spectra during spin-flip transitions in thin metal films.
The day leaves open questions regarding the fundamental limits of quantum state representation and superconducting physics. Slave-boson Formalism for Superconducting Pairing at Strong Coupling maps pairing instabilities to experimental cuprate observations, while Nonlinear Magnon Magnetic Moment Transport in Triangular-Lattice f-Wave Antialtermagnets demonstrates how magnons carry magnetic moments in frustrated systems. Meanwhile, Pseudogap and Non-Fermi-liquid criticality in double Kondo model for bilayer nickelates investigates non-Fermi liquid behavior in bilayer nickelates, suggesting that the pseudogap phase is accessible via electron doping.
Robotics Radio
The day's focus on Robotics Radio is heavily weighted toward understanding how biological systems move, particularly with work like, which shows how tuning posture and muscle stiffness lets animals exploit mechanical resonance for effective gaits. This contrasts with many robots that lack this specific tuning for nonlinear dynamics.
Other papers contribute to the field by addressing different challenges; some look at advanced planning, such as which proposes DA-WAM to align world models with driving decisions through candidate trajectory scoring and safety checks. Others tackle control authority, like which introduces a forward-invariant control barrier function for overactuated multirotors, ensuring robust movement despite actuator limits.
The day leaves open the question of how to translate biological principles of mechanical resonance into predictable control frameworks for physical robots. While shows how nonlinear modes emerge from resonance in a quadruped, and shows robust authority guarantees for multirotors, there is still a gap in developing generalized methods that allow robots to naturally leverage these inherent nonlinearities for agile locomotion without needing specific prior tuning.
Security Radio
The main focus of yesterday's findings involved securing large language models against intellectual property theft, specifically through a novel key-conditioned watermarking framework called SAFESEAL. This work demonstrated that SAFESEAL maintains high utility while providing strong detection capabilities by using context-aware synonyms to preserve semantic fidelity.
Other research addressed different security vectors across various domains. The paper on Anywhere, Any-Stymie details a method for triggering dormant malware on LiDAR sensors using modulated signals to perform false data injection in autonomous systems. Another piece, Quantifying and Defending against the Privacy Risk in Logit-based Federated Learning, analyzed a hidden privacy risk where semi-honest servers could learn private models from logits during federated learning.
The remaining papers cover diverse topics. Memory-Induced Tool-Drift shows how biased memories can cause LLM agents to silently shift tool parameters, while an AI Security Agent for University ACMIS presents a threat detection system with high F1 scores against simulated institution logs. Furthermore, Alexa versus Alexa investigates a self-issuing voice command attack that allows control over smart speakers, and Concealing CAN Message Sequences to Prevent Schedule-based Bus-off Attacks explores obfuscating timing side channels in automotive networks using task skipping.
The day, in brief
Did you know that the way we predict race using names might be much more accurate than we think? Researchers are tackling a huge hurdle: getting individual-level racial data when it’s often missing from official records. They're developing a technique called embedding-powered Bayesian Improved Surname Geocoding, or eBISG. Instead of just looking at surnames, this system takes names and turns them into dense mathematical vectors. These vectors help estimate race probabilities even for surnames that don't appear in the Census. The biggest gains come when they look at full names, especially for Hispanic and Asian voters whose last names are absent from those records. This iterative improvement shows that richer data representations directly translate into better demographic estimations, helping us build more reliable predictions where existing methods fail.
Meanwhile, in the world of artificial intelligence, there's a fascinating breakthrough in making large language models more trustworthy when they don't have perfect answers. The method they developed is called Binary Agreement for Self-consistency Calibration, or BASC. When an AI doesn't have a confidence score because no correct label is available, BASC treats the model’s self-consistency—how much it agrees with itself across different responses—as a proxy for correctness. It allows the system to recalibrate its scores using just two responses per question, providing a distribution-free guarantee across nine models and five tasks. This essentially gives us a way to trust AI outputs even in situations where they can't definitively prove their own accuracy.
Shifting gears entirely, we have some mind-bending physics happening at the edge of what we thought was possible in superconductivity. Scientists have found evidence for a non-unitary triplet pairing state in a material called TaRuSi. This suggests a complex superconducting state that breaks time-reversal symmetry—a pairing mechanism far more intricate than the simple s-wave states we usually study. What makes this significant is that the researchers linked this to how spin-orbit coupling interacts with the material's structure, suggesting that these subtle structural differences dictate whether you get unitary or non-unitary pairing.
And in a completely different area, when it comes to autonomous driving, the challenge isn't just predicting where things are; it’s knowing when to stop guessing. A new approach called Test-Time Trajectory Optimization, or TOAD, solves this by treating the planning process itself as a search problem that happens right at test time. Instead of generating a whole path and then scoring it afterward, TOAD uses a learned reward function to actively search for the best trajectory while it's being calculated. This means the robot is making better decisions mid-calculation, leading to significant improvements in how safely and efficiently self-driving systems navigate complex roads.
The week in the estate's ingest
Seven days of papers the estate ingested. Weekends are striped — the weekend is visible in the data.
Show the answer
1144 papers were ingested on Thursday 22 October.
1144 papers, three ways
Where today's batch landed
| cs.LG | 194 |
|---|---|
| cs.CV | 154 |
| cs.CL | 124 |
| cs.RO | 89 |
| cs.AI | 79 |
| quant-ph | 74 |
| astro-ph.GA | 35 |
| eess.SY | 31 |
The index
The day's strongest 60, by score. Every title links its paper here and on arXiv.
- Radiative spin caloritronics in nonreciprocal many-body systems [arXiv] [episode]
- Edge universality in Floquet sideband spectra [arXiv] [episode]
- High Fidelity Single-NV Qubit Quantum State Tomography by Photoelectric Readout [arXiv] [episode]
- Dipolar excitonic quantum wires at atomically sharp lateral interfaces [arXiv] [episode]
- VDDP: Verifiable Distributed Differential Privacy under the Client-Server-Verifier Setup [arXiv] [episode]
- Constructing and characterising anharmonic potentials in a harmonic oscillator [arXiv] [episode]
- Single-step phase-engineered pulse for active readout cavity reset with a known qubit state [arXiv] [episode]
- Topological and Trivial Valence-Bond Orders in Higher-Spin Kitaev Models [arXiv] [episode]
- Improved Nonlocality Certification via Bouncing between Bell Operators and Inequalities [arXiv] [episode]
- Demonstration of tripartite cat states in two distinct classes of entanglement [arXiv] [episode]
- Robust Signatures of Fragile Topology in Nonmagnetic Materials [arXiv] [episode]
- VertMark: A Unified Training-Free Robust Watermarking Framework for Vertical Domain Pre-trained Language Models [arXiv] [episode]
- AuthOr: Lower Cost Privacy-Free Garbling of Arbitrary Boolean Circuits [arXiv] [episode]
- Depth dependence of entanglement and circuit complexity in random linear optical networks [arXiv] [episode]
- Geometrically Frustrated Quadrupoles on the Pyrochlore Lattice and Generalized Spin Liquids [arXiv] [episode]
- Spatiotemporal spin transport from first principles [arXiv] [episode]
- A parity obstruction to degree-optimal syndrome extraction in bivariate bicycle codes [arXiv] [episode]
- Multi-User Multi-Key Image Steganography with Key Isolation [arXiv] [episode]
- Population Neuroscience 3.0: Bridging large-N and small-N paradigms in brain-wide association studies [arXiv] [episode]
- Bootstrapping Life-Inspired Machine Intelligence: The Biological Route from Chemistry to Cognition and Creativity [arXiv] [episode]
- Microscopic Parametric Correlations and Spectral Rigidity in the Non-Hermitian Threefold Way [arXiv] [episode]
- Highly integrated quantum key distribution transmitter enabled by silicon photonics [arXiv] [episode]
- Discovery of connectivity-trainability trade-off of IQP Circuits for Hamiltonian Optimization [arXiv] [episode]
- Slave-boson Formalism for Superconducting Pairing at Strong Coupling [arXiv] [episode]
- Experimental subdiffraction source discrimination enabled by spatial demultiplexing and single-photon detectors [arXiv] [episode]
- Enabling Deterministic Passive Quantum State Transfer with Giant Atoms [arXiv] [episode]
- Mixed-State Topology in One-Dimensional Open Floquet Systems [arXiv] [episode]
- Spin Kerr-cat qubits [arXiv] [episode]
- Analysis of Io's tidal response as a function of the properties of the partially molten layer [arXiv] [episode]
- Nonperturbative Semiclassical Spin Dynamics for Ordered Quantum Magnets [arXiv] [episode]
- Mismatch between Raman shear modes and ferroelectric polarization in 3R-MoS 2 [arXiv] [episode]
- Entanglement Dimensionality of Continuous Variable States From Phase-Space Quasi-Probabilities [arXiv] [episode]
- Exact quantum circuits for a lattice Boltzmann realization of the Dirac equation [arXiv] [episode]
- Pushing Biomolecular Utility-Diversity Frontiers with Supergroup Relative Policy Optimization [arXiv] [episode]
- Classification of Kirkwood-Dirac positivity-preserving unitaries [arXiv] [episode]
- Energetic Costs of Subspace Quantum Error Correction [arXiv] [episode]
- Optimal phase estimation in the presence of correlated dephasing [arXiv] [episode]
- Interpretable Embeddings of Speech Enhance and Explain Brain Encoding Performance of Audio Models [arXiv] [episode]
- Spectral decomposition and high-accuracy Greens functions: Overcoming the Nyquist-Shannon limit via complex-time Krylov expansion [arXiv] [episode]
- Bardeen-Cooper-Schrieffer State Representation and Pairing in the Fermionic Tonks-Girardeau Gas [arXiv] [episode]
- Quantifying and Defending against the Privacy Risk in Logit-based Federated Learning [arXiv] [episode]
- Predictive and feedback signals differently shape the formation of language representations [arXiv] [episode]
- Numerical study of the free energy monotonicity for the Enskog equation: The thermal relaxation of the dense gas between two parallel plates [arXiv] [episode]
- Diffusive Instabilities in Dusty Disks: Linear Growth and Nonlinear Breakdown [arXiv] [episode]
- Learning Hamiltonians for O(1) Oracle-Query Quantum State Preparation [arXiv] [episode]
- Electronic crystals and quasicrystals in semiconductor quantum wells: an AI-powered discovery [arXiv] [episode]
- Pseudomagnetic focusing of valley-polarized beams in strained graphene [arXiv] [episode]
- Realization of a Rydberg-dressed extended Bose Hubbard model [arXiv] [episode]
- Reducing strain fluctuations in quantum dot devices by gate-layer stacking [arXiv] [episode]
- Variational quantum-neural hybrid imaginary time evolution [arXiv] [episode]
- Quantum mechanical Gaussian wavepackets of single relativistic particles [arXiv] [episode]
- Microscopic study of topological phase transitions: Percolation point of view [arXiv] [episode]
- Entropy transport through a superfluid quantum point contact: A Keldysh field-theory approach [arXiv] [episode]
- Asteroseismology of the High-amplitude delta Scuti Star EH Librae [arXiv] [episode]
- Superfluid rings as quantum pendulums [arXiv] [episode]
- Optimal skyrmion stability in antisymmetric ultrathin ferromagnetic bilayers [arXiv] [episode]
- Resonating-valence-bond superconductor from small Fermi surface in twisted bilayer graphene [arXiv] [episode]
- Alexa versus Alexa: Controlling Smart Speakers by Self-Issuing Voice Commands [arXiv] [episode]
- Electrical detection of spin-flip transition in metal/ heterostructure [arXiv] [episode]
- Pseudogap and Non-Fermi-liquid criticality in double Kondo model for bilayer nickelates [arXiv] [episode]
Listen
Everything the estate published on 2026-10-22, in the player. Playing from this page is how the edition learns whether anyone did.
- Daily Summary for 2026-10-22
- Electrical detection of spin-flip transition in metal/ heterostructure
- Daily Summary for 2026-10-22
- Nonlinear Magnon Magnetic Moment Transport in Triangular-Lattice f-Wave Antialtermagnets
- Slave-boson Formalism for Superconducting Pairing at Strong Coupling
- Daily Summary for 2026-10-22
- Daily Summary for 2026-10-22
- Evidence for non-unitary triplet-pairing superconductivity in noncentrosymmetric TaRuSi and comparison with isostructural TaReSi
- Daily Summary for 2026-10-22
- Alexa versus Alexa: Controlling Smart Speakers by Self-Issuing Voice Commands
- Daily Summary for 2026-10-22
- AuthOr: Lower Cost Privacy-Free Garbling of Arbitrary Boolean Circuits