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.
LazyAgent: Demand-Driven Materialization and Physical Optimization of Agentic Programs
Xin Heng
Paper proposes demand-driven backward-closure permission for agent programs, claiming 42–52% cuts vs stop-eager; evaluation is unusually rigorous but the regime itself is unproven.
Event Signature Transfer: Model-Agnostic Forecast Scenario Construction from Historical Events
Karthik Sridhar, Aaditya Jain, Murari Mandal, Saurabh Deshpande
EST is a training-free, quantile-only operator transferring past event effects onto forecasts; promising but entirely in-sample, with hand-set controls and no prospective test.
Directing large language models to follow the letter or spirit of the law
Peng Qian, Andrew Li, Sam Chen, Sonia K. Murthy et al.
Anchored-LoRA adapters steer LLMs to letter- or spirit-based judgment across vignettes, Reddit, caselaw; solid evidence, letter-side effects weaker.
Tutoring Large Language Models to be Domain-adaptive, Precise and Safe
Somnath Banerjee
PhD thesis compiling six peer-reviewed works on domain-adaptive, safe, culturally-aware LLMs; broad and credible, though evaluation rigor varies across parts.
FireWorldBench: Benchmarking Complex Physical World Intelligence through Coupled-Field Fire Dynamics
Qiang Chen, Hao Guo, Huatai Zhu, Tairan Huang et al.
Solid FDS-grounded benchmark (520 fire worlds, 9,074 QA) exposing MLLM physical-reasoning limits; caveat: real-world gold is simulation-derived, not measured.
AirGC-CD: Gaussian-Circulant Precoding for Exactly Debiasable PAPR Reduction in Over-the-Air Federated Learning
Jonggyu Jang, Hyeonsu Lyu, Hyun Jong Yang
Elegant Gaussian-circulant sketch makes clipped over-the-air FL exactly unbiased at O(d log d); sound theory, but evaluation is AWGN-only without error bars.
Measured Joules, Learned Routes: Learning to Route for Energy-Efficient LLM Serving
Muhammad Abdur Rab Siddiqui, Daniela Rojas, Chen Yang, Wenqi Cui et al.
Proposes measured-joule LLM routing (KL-GRPO, 60.9% acc, 1692 J); promising, but a fixed Qwen-7B arm (64.0%, 282 J) beats all learned routers on both metrics.
MolSC: Leveraging Substituent Contributions to Enhance Fine-grained Molecular Understanding in LLMs
Hyuntae Park, Sooyeon Kim, Jiwon Park, SangKeun Lee
MolSC (181K substituent contributions) sharply improves LLM substituent-effect prediction, though downstream gains partly stem from its dual-view training.
Neural Spectral Capacity: Measuring and Designing Architectures from Network Specification Alone
Chenyu Zhu, Ruoyu Zhao, Zhichao Lu
Proposes NSC, a closed-form spec-only capacity proxy with exact DP search; results are broad and fast, but GPT-2 ranking numbers temper the headline claim.
OmniEdu: Open Foundation Models for Learning and Teaching
Hao Liang, Qihan Lin, Meiyi Qiang, Linzhuang Sun et al.
Capability-balanced SFT lifts Qwen 4-27B across K-12 benchmarks; gains plausible, but same-named train/eval sources and missing ablations temper the claims.
The Role of Coordinates in Pareto Regret for Adversarial Multi-Objective Bandits
Changkun Guan, Mengfan Xu
Minimax Pareto regret in adversarial MO-bandits is Θ(min{U0,√(KU0)}), proved with matching reward-doubling Poly-INF; rigorous, K≥4, no experiments.
When Does Adversarial Refinement Help? A Negative Result and Open Problem in Adapting R3GAN to Time Series Imputation
Yufeng He
Honest negative result: stable modern GAN refinement beats trivial fills but not linear interpolation; sharp diagnosis but its key ablation is missing.
An LLM-Assisted AutoML Framework for Intrusion Detection in IoT Networks
Li Yang
LLM-guided compact AutoML beats 10-trial TPE on two IDS benchmarks; honest but single-seed results lack a no-LLM fixed-policy control; attribution unclear.
Counting and Covering in Nearest-Neighbour Representations of Boolean Functions
Martin Anthony
Exact covering-design formulas and sharp probabilistic limits (11/20) for Boolean nearest-neighbour complexity; rigorous theory paper, partly incremental.
MM-ContextFold: Context Folding for Multimodal Agentic Retrieval
Yang Tian, Fan Liu, Jingyuan Zhang, Zhenyang Li et al.
Claims training-free dual-state context folding boosts MAR accuracy 6.3pp over ReAct at 27.5% shorter context; broad, mostly convincing evidence.
DiagGen: Agentic Generation of Deformable Assets with Sim-based Diagnostics for Robotic Simulation
Guanxiong Chen, Yiduo Qu, Qianjun Xia, Pengyu Jing et al.
DiagGen closes the loop from image to simulatable deformable via sim-based stage-routed repair; credible but small-n, VLM-judged, no significance tests.
Whitening Inverts the Hierarchy: What the Norm of a Whitened Embedding Measures
Mohammed Ahnouch, Lotfi Elaachak
Whitened-embedding norms are Mahalanobis atypicality, not likelihood: matched-clone evidence over six encoders strongly supports this anti-Gaussianity critique.
Perplexity Cost Understates What Activation Quantisation Breaks
Anish Sathyanarayanan
Claims perplexity can mask capability-specific quantisation damage; matched controls are strong, but probe noise and an unmatched repair temper the evidence.
CraftBench-UE: Deterministic Evaluation for Coding Agents in Unreal Engine
Shutong Wu, Kevin Calderone, Andy Tsen
Solid no-LLM-judge Unreal agent benchmark; C++ beats Blueprint by 30-43pp; single-attempt results lack significance tests and MCP2 comparison is confounded.
From Inference Engine to Inference Control Plane: Connecting vLLM, llm-d, and the Evolution of Efficient Distributed LLM Serving
Twinkll Sisodia
对 LLM 服务从引擎到控制平面的演进进行的一次细致的系统综合（无新基准测试）；归因诚实，但其规划器提案尚未经过测试。
Provably Efficient Reinforcement Learning in Continuous-Time Episodic MDPs with Poisson Decision Epochs
Kenny Guo, Valentio Iverson, Sahan Wijetunga, William Chang
Extends UCRL and Q-learning to Poisson-epoch Lipschitz-in-time MDPs, proving Õ(T^{2/3}) regret with matching lower bounds; theory-only, some proof gaps.
Real-Time Plasma State Prediction via FPGA-Accelerated Quantized Recurrent Probabilistic Neural Networks
Daniel Gaytan-Villarreal, Aiken Xie, Tu Pham, Rohit Sonker et al.
谨慎的FPGA部署工作流将DIII-D等离子体RPNN缩减了94.9%，实现了2.64 μs的延迟，但硬件数据仅基于时序仿真；闭环EV会衰减。
QwenVLConnector: A Fast, Unified Medical VLM Chatbot for Fine-Grained Clinical Perception and Text Generation
Le Thien Phuc Nguyen, Thien Nguyen, Thanh-Huy Nguyen, Gia Minh Hoang et al.
Claims a dense connector boosts FLARE-2D medical tasks on Qwen2.5-VL; gains are plausible but attribution confounded; regression and multi-label worsen.
Ask for Any Appliance: A Prompt-Programmable Foundation Model for Non-Intrusive Load Monitoring
Xudong Wang, Jiacheng Cui, Junyu Xue, Tongxin Li et al.
Claims a 3.7M prompt-programmable model matches NILM specialists (F1 0.556, MAE 251.8 W); strong evidence, but energy ratio 2.54 invites scrutiny.