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.
Who Teaches Which Token? Verifier-Gated Multi-Expert On-Policy Distillation for Scientific Reasoning
Xun Xu, Zaixi Zhang
VG-OPD gates multi-expert distillation by per-criterion verifier gains; strong ablations at 4B/8B, but single-run results lack significance testing.
Can AI systems have free will?
Christian List
List argues AI free will is possible via a Dennett-style checklist (agency, alternatives, control); rigorous philosophy but criteria remain qualitative.
Same path, different: a mechanistic comparison of looped and stacked transformer encoders on 12-lead ECG
Pawel Olszowiec, Michal Byra, Grzegorz Gruszczynski, Grzegorz Stefanski et al.
Claims looped bViT matches stacked ViT on ECG with 8.9× fewer params and calmer dynamics; controlled study, though latent-only OOD probes limit the safety claims.
Empirical Evaluation of Task-Based Permission Scoping Architecture for AI Agents
Halil Burak Noyan
Paper: task-scoped gating cuts 84.4% of weighted attack surface; honest eval, all-synthetic benign data, no adversarial tests, own deployability bar unmet.
On The Robustness-Resolution Tradeoff In Temporal Quantization Of Event Streams
Sayeed Shafayet Chowdhury, Ruhi Sharmin, Syed Ishtiaque Ahmed
Claims hard binning is discontinuous and proves a 2/Δ sensitivity floor met by linear interpolation; theory is simple but sound and experiments support it.
A Conservative OCR-Enabled Workflow for R214 Sodium Screening of South African Packaged Foods
Mayimunah Nagayi, Alice Scaria Khan, Tamryn Frank, Rina Swart et al.
Conservative OCR+VLM workflow screens 442 SA foods vs R214 sodium limits; honest uncertainty handling, but only 13 decisive outcomes and no code/data release.
Single-condition neural solvers encode transferable response spaces for parametric differential equations
Wenbo Cao, Weiwei Zhang
Claims a single-condition solver's Jacobian forms a reusable response space for cross-condition PDE transfer; strong controls, but baselines narrow.
Online local learning for generative thermodynamic computing
Huilin Wang, Weibing Deng
Adds a per-step local rule to Whitelam's thermodynamic generator; paired sims (n=5) show 5.6% heat cut, but sign-test n.s. and tiny MNIST task limit scope.
On the role of the tokenizer in ECG transformer models
Jiawei Li, Fabio Bonassi, Johan Sundström, Thomas B. Schön et al.
对照研究发现，心跳对齐的分词器（beat-aligned tokenizers）在 CPSC2018 数据集上将 macro-AUC 从 0.823 提升至 0.891，并将 VRAM 削减了 94.8%；这一结果可信，但仅基于单一数据集，且缺乏显著性检验。
GSLAD: Prototype-Regularized Graph Structure Learning for Multivariate Time Series Anomaly Detection
Zepeng Zhang, Fuad Khuri, Keivan Faghih Niresi, Olga Fink
Prototype-regularized graph structure learning for MTS anomaly detection; strong gains on 4 benchmarks, but no code and unreported variance temper confidence.
Turkish MMLU Pro: Traceable Option Augmentation and Its Validity Limits in Turkish Multiple-Choice Evaluation
M. Ali Bayram
Augments 12,000 Turkish MCQs with borrowed options, shows accuracy drops, and candidly maps validity limits; a traceable resource, not a validated benchmark.
HISPO: Hierarchical Importance-Sampling Policy Optimization with Entropy-Derived Segments
Quoc-Vinh Lai-Dang, Hyo-Sang Shin
HISPO's segment-level entropy-weighted RLVR objective looks promising, but gains are single-seed, checkpoint-selected, and fragile in fresh bootstrap reruns.
Rotation-Based Subspace Tracking for Robust Kernel PCA on Streaming Data
Kris Lokere, John Fossaceca
Adds rotation-based updates to online robust kernel PCA; faster synthetic convergence shown, but no experiment tests actual subspace drift tracking.
Temperature Fragility and the Conditional Benefits of Truncation Sampling
Francesco La Rosa
关于截断采样（Truncation sampling）仅在高温度导致准确性崩溃时才有帮助的断言；这是一项严谨的 232k 生成研究，尽管每个任务仅有 50 个项目。
Data-driven Prediction of Satellite-observed Avalanche Activity from Snowpack Simulations
Jakob Grahn, Filippo Maria Bianchi, Bert Kruyt, Karsten Müller
Uses SNOWPACK simulations to predict SAR avalanche activity (r=0.803 regional, 0.549 per cell); promising, but single run, no baselines, no untouched winter.
The Troy Moment: How LLM Agents Adjudicate the Decision Point Under Impossible Tasks, Claimed Authority, and Peer Information
Ivy Zhang
Three frontier LLM agents show distinct evidence-to-action policies under impossible tasks; well-designed probe but tiny samples and no statistical tests.
Timbre Analysis of the Hulusi, a Southwestern Chinese Free-Reed Instrument, using Machine Learning
Yang Xia, Rolf Bader
Claims hulusi timbre clusters by pitch via SOM on 3 psychoacoustic features; exploratory and useful, but lacks statistical validation and baselines.
Beyond Safe Answers: Segment-Aware Listwise Alignment for Reasoning Safety in Large Reasoning Models
JungMin Yun, Junehyoung Kwon, Hayeong Ryu, Byeonggeuk Lim et al.
Proposes segment-aware listwise DPO for LRM reasoning safety; strong gains look credible but rest on same-family judge scoring and no significance tests.
Option-Aware Retrieval and Task-Specific VLM Adaptation for Medical VQA
Tristan Kirscher, Niklas C. Koser, Soren Pirk
A challenge report: answer-semantic retrieval fixes MCQ label-transfer failure (20→57.5%); LoRA drives gains, retrieval/gating add ≤1 case; OE evidence thin.
Where to Compute and How to Interact: Operator-Readable Adaptation with Gauge-Aware Transport
Zixuan Shen, Quanxu Wan, Bingchuan Wang, Zhi Wang et al.
GA-AMNO adds pre-aggregation feature transport to adaptive-mesh operators and wins all five PDE benchmarks, but single-seed baselines and no code temper it.
Corrupt Plans, Clean Traces: Evading Chain-of-Thought Monitoring with Plan Injection
Keertana Chidambaram, Andrew Ilyas, Vasilis Syrgkanis
Compelling new plan-injection attack that evades CoT monitoring; strong evidence, but headline steering rates are inconsistently reported.
Towards Surrogate Based Dequantization of Quantum Reinforcement Learning
Pablo Rodriguez-Grasa, Sofiene Jerbi, Mikel Sanz, Ryan Sweke
First dequantization guarantees for quantum Q-learning via kernel FQI under uniform sampling; rigorous but strongly conditioned, no experiments.
Authorship attribution and aesthetic evaluation of AI poetry: a case study with Haiku
Livia Oddi, Simone Scardapane, Toru Sugimoto, Donatella Genovese
Finds near-chance human detection of top-model haiku; ratings track perceived, not true, authorship. Sound design, tiny stimulus set limits generalization.
End-to-End Verifiable and Robust Federated Learning
Doryan Lesaignoux, Enrique Mármol Campos, Gabriele Spini, José L. Hernández-Ramos et al.
Proposes NIZK-verified robust FL aggregation with hidden selection; honest, competitive results, but evaluation is small-scale and proofs reach hundreds of MB.