Reid Hoffman

The Sloan Great Wall Anchors the Largest Known Basin of Gravitational Attraction in the Local Universe

Using Hamiltonian Monte Carlo forward reconstruction constrained by the Cosmicflows-4 galaxy distance and velocity dataset, researchers mapped the gravitational basins of attraction in the local cosmic volume. The Sloan Great Wall emerges not only as a prominent large-scale structure in redshift surveys but as the dominant gravitational sink in the region, anchoring a basin of attraction spanning ~0.13c in diameter — the largest identified in the study volume. The work provides probabilistic density and velocity fields within a ΛCDM framework, with uncertainty estimates propagated from both observational noise and model stochasticity, and explores the spatial relationship of the Ho'oleilana baryon acoustic oscillation feature to this basin.

SVMs Outperform Deep Learning Alternatives for Pre-Harvest Crop Mapping Using Satellite Time Series

This study benchmarks ten machine learning algorithms for in-season crop type mapping using Harmonized Landsat-Sentinel surface reflectance time series and crop rotation history, targeting corn in Iowa and almonds in California at 30m resolution. Using year-wise cross-validation across thousands of model configurations, Support Vector Machines emerged as the top-performing algorithm, achieving mean F1 scores of 0.74 for almonds and 0.59 for corn by early June — before harvest. The work directly addresses a gap in USDA capabilities, where no pre-harvest product currently exists with sufficient accuracy for real-time emergency response to crop threats. Interannual variability was identified as the dominant uncertainty source, with ensemble methods and ancillary data flagged as promising paths to further gains.

Agentic LLMs with Formal Verification Tools Outperform Base Models for Network Misconfiguration Repair

Network misconfigurations are a leading cause of Internet outages, and while LLMs have been explored for automated repair, standalone models consistently fail on large-scale, complex scenarios and often introduce new errors. This paper benchmarks open- and closed-source LLMs augmented with formal network verification and context retrieval tools in an agentic architecture. The agentic setup achieves an average 12% improvement in repair efficacy and 17% improvement in safety over base LLMs, driven by dynamic context management and iterative validation loops.

Active Sensing Subserves Task-Level Control

IceCube Releases 14-Year Muon Track Dataset (IceTracks-DR2) for Open Neutrino Source Searches

The IceCube Collaboration has published IceTracks-DR2, its most comprehensive public muon track dataset to date, spanning 14 years of observations (2008–2022) and superseding the prior 10-year release. The dataset incorporates improved detector calibration for post-2010 data and includes binned instrument response functions and effective areas, enabling third-party researchers to conduct independent point-source and transient neutrino analyses. Benchmark results using publicly available software are provided to support reproducibility, lowering the barrier for community-driven searches.

STM Current Pulses Enable Reversible Nanoscale Ferroelectric Switching in WTe₂

Researchers demonstrate that scanning tunneling microscopy (STM) current pulses can write, reposition, and erase persistent nanometer-scale structural patterns on the surface of WTe₂ — a Weyl semimetal and ferroelectric material. The induced patterns manifest as picometer-scale in-plane and out-of-plane atomic displacements accompanied by local density-of-states changes, with in-plane displacements interpreted as ferroelectric switching and out-of-plane displacements modulating the Peierls-like distortion intrinsic to WTe₂'s Td phase. This establishes a nanoscale, reversible handle on ferroelectric order in a topological quantum material, a capability previously elusive despite prior work on bulk ferroelectric switching and ultrafast transient control.

Expanders for Mean Curvature Flow and Counterexamples to Ilmanen's Genus-Reduction Conjecture

An explicit Galois descent for multiple $t$-values of maximal height

Multi-layer attentive probing improves transfer of audio representations for bioacoustics

Bi-Level Optimization for Contact and Motion Planning in Rope-Assisted Legged Robots

Activated random walk exhibits self-organized criticality

Posterior Augmented Flow Matching