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.
Beyond Scores: Understanding LLM-as-a-Judge Mechanisms in Summarization Evaluation
Himil Vasava, Ming Jiang
Claims LLM-judges run a two-stage attention→MLP pipeline crystallizing at L25-26; a solid study, limited by n=2 evaluators and detection-filtered data.
When Features Become Instances: Inverted Contrastive Learning for Unsupervised Feature Selection
Utsab Ghosh, Roshni Chakraborty
Claims best clustering accuracy on 10/12 datasets via feature-wise contrastive norm ranking; plausible but gains small, no significance tests, sloppily edited.
CAT-Flow: Curvature-Adaptive sTeps for Flow Matching
Qinchan Li, Pedro Cisneros-Velarde, Keru Fu, Samuel Antunes Miranda et al.
CAT-Flow claims training-free curvature-adaptive step sizes for Flow Matching; theory is thin, but 4-model experiments support the NFE savings.
TEIDAN: A Multilingual Multiparty Dialogue Corpus
Taiga Mori, Koji Inoue, Mikey Elmers, Divesh Lala et al.
TEIDAN: 69-session JA/EN triadic dialogue corpus, well documented and honestly caveated; data release unconfirmed and cross-language claims stay descriptive.
Sparse Readout Prism: Explaining Logit-Lens Scores in Features Instead of Tokens
Matteo He, William F. Shen, Xinchi Qiu, Nicholas D. Lane
SRP: an SAE over LM-head rows decomposes lens scores exactly; strong baselines and intervention checks, but cross-recipe feature stability is limited.
StudyBench: Can Self-Evolution Squeeze Textbooks for Olympiad Capability?
Yinghao Chen, Zixi Chen, Bingxiang He, Ziqing Qiao et al.
Controlled physics benchmark separates absorption from transfer in self-evolution, yet small test sets and Qwen-anchored filtering limit generality.
Ctrl-F-Resist. Practices, Challenges, and Technical Needs of Civil Society Organizations Monitoring the Far-Right Online
Elisabeth Steffen, Helena Mihaljević
Interview study of 15 practitioners at 12 German CSOs: monitoring stays manual, search is top need; solid qualitative work, new concepts partly programmatic.
MADS: A Multiview Acoustic Descriptor Set Beyond Standard Spectral Summaries
Utsab Ghosh, Roshni Chakraborty
Proposes 19-D physics-inspired audio descriptors that beat 26/38-D baselines on ESC-10/50 and MSoS; margins ≤1.6 pts, no significance tests, code withheld.
Instella-MoE Technical Report
Jiang Liu, Sudhanshu Ranjan, Prakamya Mishra, Yonatan Dukler et al.
Instella-MoE is a genuinely fully open 16B MoE (2.8B active) trained on AMD GPUs with credible, verifiable fully-open SOTA claims, but headline margins are tiny, single-run, and error-bar-free.
One Policy, Any Budget: Internalizing Budget-Aware Search via Reinforcement Learning
Xiaowei Sun, Jin Li, Yili Hong, Yikun Fu et al.
AnySearch 训练单一策略以适应任意搜索预算，通过脚手架课程强化学习实现；基准测试和消融实验表现强劲，但性能提升有限且未经过充分测试。
Agentic programs: an emerging form of scientific software in computational materials science
Yunsung Lim, Haekwan Jeon, Jaesun Kim, Jisu Kim et al.
Position paper defining 'agentic programs'; DeMARS case study is compelling but 100-CIF validation is qualitative, with no error rates or baselines reported.
HarnessEvolve: Learning from Reference Trajectories for Reliable Agent Self-Evolution
Wen Jiang, Mingmin Chu, Yimeng Tian, Qianxin Zhang et al.
HarnessEvolve evolves the agent harness via reference trajectories and dual gates, beating GEPA/ACE/SkillOpt on 5 benchmarks, without significance tests.
Towards a Reliable and Practical Eval Pipeline
Emma Thuong Nguyen, Abhishek Ghose
Checklist + GBDT aggregator, SHAP and conformal intervals for LLM judges: solid engineering, but validated on SummEval only, without significance tests.
Verifiable Disaster Storylines and Causal Knowledge Graphs: A Citation-Grounded Pipeline from Heterogeneous Humanitarian Sources
Ivan Decostanzi, Michele Ronco, Sergio Consoli, Christina Corbane et al.
RAG pipeline for citation-grounded disaster storylines and causal KGs from EM-DAT, ReliefWeb, EMM; promising, but only 3 events and weak annotator agreement.
Situation Awareness for Intelligent Data Distribution in Connected Vehicles
Falk Dettinger, Akshay Narla, Michael Weyrich
BEV traffic-situation classifier (DeepLabv3+, Cam2BEV, VGG16): 76.7% CARLA, 61-65% real-world; useful concept, but V2X data-distribution link is unevaluated.
AnalysisBank: An Expert Analysis Pattern Library for Financial Report Generation
Yajing Yang, Yunshan Ma, Kelvin J. L. Koa, Min-Yen Kan
Expert-report pattern library lifts LLM insight rates 1.7–3.7×; convincing ablations, but LLM-judged metrics and possible benchmark overlap need scrutiny.
Polished but Unresolved: Identifying Late-Stage Pressure States in Long-Horizon Tool-Use Agents
Haoyang Chen, Yi Liu, Jianzhi Shao, Xiaozhou Xu et al.
Claims a linearly decodable 'late-stage pressure' state drives premature agent submissions; controls are solid but gains are modest and labels are LLM-defined.
TWIX: a Two-Stage Approach for End-To-End Named Entity Recognition and Relation Extraction
Marco Martinelli, Laura Menotti
TWIX's two-stage pipeline tops all four GutBrainIE subtasks by big margins; strong engineering, but no significance tests and heavy dev-set config tuning.
FLaG: Frequency-Domain Latent-attention Gated Pooling for Token Aggregation
Kewei Li, Rongying Zhang, Xueli Wang, Xiwen Gong et al.
FLaG adds Fourier-domain latent-attention gating before pooling; gains are real but small and task-dependent, and key ablations weaken the full design.
Visual Attention Faithfulness in Vision-Language Models is Heterogeneous
Xurui Song, Weishi Wang, Zhongqi Yue, Kuluhan Binici et al.
Perturbation study finds VLM visual attention faithfulness heterogeneous (3 modes); novel but small n and manual mode partition temper the conclusions.
Probabilistic Model Checking of Autoregressive Neural Sequence Models
Helge Spieker, Dennis Gross, Arnaud Gotlieb
Turns autoregressive generators into PRISM-checkable DTMCs with certified under-approximation bounds; sound and honest, but dmax=20 and T=1.0 limit scope.
Dense Process Supervision for Search Agents via Fact Utility Estimation
Rongzhi Zhu, Xiangyu Liu, Yi Liu, Shuo Zhang et al.
FactAgent converts outcome rewards into Bayesian fact-utility step rewards (51.2 avg EM, 7B); promising method, but per-dataset claims outrun the evidence.
Reflect-SQL: A Self-Reflection Based Framework for Text-to-SQL
Anupreksha Jain, Manish Shrivastava
Self-reflection loops lift BIRD to 72.03% (Claude 4.5); sensible design, but eval subset unclear, no significance tests, and SOTA claim overstated.
A Checklist to assess the energy and carbon impacts of ML/AI applications in Earth System Modeling
Filippo Dainelli, Amirpasha Mozaffari, Marina Castaño, Aina Gaya i Àvila et al.
Dainelli et al. distill Green AI guidance into an 11-question, stage-tagged ESM checklist plus metrics; a useful, well-sourced bridge, but unvalidated, thin at Deployment, and in need of copyediting.