Evolve Quick Reviews
Public Paper Reviews
Structured preliminary reviews of recent open-access papers, generated and continuously updated by the Evolve analysis engine. Each review condenses what the paper claims, how solid the evidence is, and the questions a rigorous peer reviewer would ask — English analysis with abstracts in your language. These are independent AI analyses, not peer review; the canonical record always lives at the publisher.
Generalist-Specialist Mixture-of-Experts for Rare Pathology Detection in Multimodal Imaging
Johannes Kaiser, Florian Braunmiller, Daniel Rückert, Georgios Kaissis
GS-MoE adds a cross-modal generalist to modality-routed MoE, reviving six rare pathologies all baselines miss; evidence per class rests on 4-8 test images.
Playing log(N)-Questions over Wikipedia Abstracts: Communication Efficiency Between Paired Frontier Models
Peter Potash
Frontier-LLM self-communication win rate follows p^log2 N (p=0.928) with Opus 5 last; credible paired design and honest reporting, thin per-cell statistics.
Toward Markerless Video-based Tremor Analysis: Objective Quantification of Pathological Tremor in Mouse Preclinical Models
Yota Koshimoto, Akihiro Tsukahara, Yasuhiro Moriwaki, Mariko Isogawa
RGB-only mouse tremor scoring reaches r=0.86 vs accelerometer PSD; well-engineered, but validated only on harmaline-challenged wild-type mice.
Gradient Descent with Stochastic Subspaces via Persistence of Memory
Subhroshekhar Ghosh, Clement Z. Q. Ng, Pierre-Louis Poirion, Akiko Takeda
Persistent gradient-aligned guidance vectors inside random subspaces cut SSD iteration complexity from O((n/d)ε⁻²) to O(ε⁻²); strong theory, small-scale validation.
Motivationally objective versus subjective decision-making: Neural correlates, behavior, and self-report
Modak, P., Brown, J. W.
Within-subject fMRI ties subjective vs objective decision contexts to mid-cingulate/dACC; behavior is convincing but brain-behavior correlations are fragile.
Osmo-hydraulic volume regulation ensures robust endothelial-to-haematopoietic transition
Kondrychyn, I., Chen, Y., Kumar, R., da Silva, J. et al.
Zebrafish study: Piezo→VRAC→Aqp1a.1 hydraulic pathway protects HECs during EHT; evidence strong overall, drug off-targets and attribution confounds remain.
Quenched Ensemble Sampling
David Yallup
Replaces nested sampling's hard energy wall with repulsive potentials, letting gradient-based SMC cross first-order transitions; convincing, well-caveated.
A Decision-Support Audit Protocol for Supervision Drift in Proxy-Labeled Credit-Risk Prediction
Mehrdad Shoeibi, Muhammad Shabanpour, Waldemar Karwowski, Niloofar Yousefi
Pre-locked five-layer supervision-drift audit for credit scoring: exemplary pre-commitment and honesty, but single-dataset with a fragile positive control.
Decoy Direction Optimization: A Post-Hoc Defense Against LLM Abliteration
Aashiq Muhamed, Mona T. Diab, Virginia Smith
A clever and cheap post-hoc defense that corrupts the attacker's contrastive estimator rather than hiding refusal, but it leaks under sustained adaptive re-estimation like every baseline it is compared against.
LLM Inference in a Flash!
Sebastian Zhao, Minseo Kim, Coleman Hooper, Luca Manolache et al.
A clean algorithmic co-design for running LLM decoding on Compute-in-Flash: integer-only execution plus dictionary-coded KV gives 15x dynamic-KV compression at ~1% LongBench loss, but all hardware gains come from an unvalidated analytical model that never quantifies device lifetime.
The Router Within: Eliciting Native Skill Routing from a Frozen LLM
Ruishuo Chen, Xun Wang, Yu Chen, Zhuoran Li et al.
Gavel reads skill routing out of a frozen LLM's mid-layer states with two 7.9M-parameter linear maps and beats retriever pipelines by up to 13.4 points, but its headline metric is an LLM judge with only 86% human agreement and its two smallest benchmarks have 75 and 45 items.
Discrete Beckmann Transport Models for One-Step Language Modeling and Reasoning
Sophia Tang, Shiyi Wang
DBTM replaces teacher distillation with a learned one-step map solving an autonomous conservation equation, and the theory plus low-NFE results are strong, but baseline comparisons hold mainly at 1-2 NFEs with no statistics anywhere.
Proportional-Fair Resource Allocation and Dual-Threshold Early-Exit Inference for Secure Cooperative Multi-Layer Edge Intelligence
Thai T. Vu, John Le, Tu N. Nguyen, Jun Shen et al.
A rigorous proportional-fair formulation with genuine conic-reformulation and monotonicity theory, undermined by a utility that counts only true positives and by thresholds tuned on the same test set used to score them.
CiteGuard-RAG: A Validation-Centered AI System for Evidence-Grounded Question Answering
Sumit Barua, Guan Hong, Halil Dursunoglu, Charles Rodgers et al.
CiteGuard-RAG builds a sensible validation layer, but its 98.3 percent and 0.0 percent headline numbers are produced by that same layer and are close to circular.
Sharp Rates and a One-Line Correction for Spectral Representation Learning
Dier Tang, Jing Yee Tan, Guangyue Han
A sharp two-sided theory replacing isotropy with alignment as the condition for spectral universality, with an exact one-line correction - but validated on real data only by correctly declining to correct.
Structural heterogeneity of CTE Type I tau filaments with an interface-shifted polymorph
Watanabe, R., Lee, E. B.
Public CTE cryo-EM reanalysis finds an interface-shifted Type I-like tau polymorph (2.96 Å) coexisting with Type I in single fibrils; single-dataset evidence.
Cyclic parthenogenesis helps populations cross fitness valleys
Yang, Z., Hardy, N. B.
Claims cyclic parthenogens cross fitness valleys faster via clonal amplification; mechanism credible but statistics absent, rarity-cost argument speculative.
Test-Time Unlearning via Sparse Autoencoder
Pingzhi Li, Jinhao Duan, Vaishnav Tadiparthi, Nakul Agarwal et al.
ARIA reframes LLM unlearning as SAE-gated refusal over frozen weights with impressive WMDP numbers, but its intervention is a canned refusal and the standard TOFU forget-quality metric scores near zero.
How I learned to stop worrying and love StopGrads: Stationarity, Convergence, and a case study on Flow Map Learning
Max W. Shen, Mark Goldstein, Zichu Wang, Aahlad Puli et al.
A sharp variational theory of stop-gradient objectives that pins down stationary points and proves convergence for flow-map training, with a clean 2x memory win but a thin empirical validation of its own theory.
A Sentinel-2 benchmark dataset for deep-learning active-fire segmentation across 25 California wildfires
Shreyan Mitra, Mohammadreza Narimani, Parastoo Farajpoor
A well-documented, openly licensed Sentinel-2 active-fire segmentation benchmark of 2,148 chips across 25 California fires, whose labels are algorithmic and therefore measure rule reproduction rather than fire-detection truth.
Computer-assisted global regularity across nonlinear families of three-dimensional periodic Navier-Stokes flows
Jose Luis Lima de Jesus Silva
Computer-assisted global regularity on explicit neighbourhoods of 3D periodic Navier-Stokes flows; internally careful but an unverified single-author Millennium-problem claim.
Permutation-Based Stegomalware in Large Language Models: Threats and Countermeasures
Danny Wood, James Stringer
A clean dual-use result: an information-theoretic stegomalware embedding that is provably behavior-preserving, plus a derangement-based defense reaching 100% of LLM parameters.
Occupancy Network-Guided Autonomous Robotic Partial Nephrectomy
Ethan Kilmer, Pit Henrich, Jiawei Ge, Paul M. Scheikl et al.
A carefully engineered dual-arm system completes eight autonomous phantom kidney tumor resections with all-negative margins, but the claim rests on four patient models and a weak single-shot baseline.
Multisource Remote Sensing and Geospatial Analysis of Vineyard Wildfire Impacts and Resilience: The 2019 Kincade Fire
Parastoo Farajpoor, Mahla Ardebili Pour, Mohammad Bagher Ghiasi, Mohammadreza Narimani
A careful, openly reproducible geospatial audit of 4,581 Sonoma vineyard fields in the 2019 Kincade Fire: lower dNBR over vines is real descriptively but dissolves under boundary, donut, placebo and residual-spatial diagnostics, so no firebreak effect is claimed.