For every paper on this page, at least one of the original authors has seen our plain-language explanation and engaged with it — either confirming it reads accurately or requesting corrections that we then applied. An endorsement does not mean the authors formally approve every sentence, but it does mean the explanation has passed the eyes of the people who wrote the paper.

983 papers reviewed by authors · 1–10 / 983

💬 NLP

Beyond Direct Answering: Aligning Educational LLMs as Socratic Guides via Heuristic Reinforcement Learning

This paper introduces HeuristicEdu, a two-phase reinforcement learning framework that aligns the Qwen2.5-7B model to function as a Socratic guide rather than a direct answerer, achieving significant improvements in scaffolding effectiveness and reduced concept leakage through heuristic rewards and Group Relative Policy Optimization.

Xiaokun Wang, Siyu Song, Wentao Liu, Xiaodong Zou2026-07-28✓ Author reviewed
💻 computer science

Removing Noise or Introducing Bias? The Hidden Cost of MSR Filtering

This study analyzes 1.57 million GitHub repositories to demonstrate that common filtering criteria in Mining Software Repositories (MSR) research introduce significant maintenance, ecosystem, and relational biases that distort project abandonment rates and variable relationships, advocating for a shift toward stratified sampling and refined noise detection.

Mohit Kaushik, Jyoti Bawa2026-07-28✓ Author reviewed
💬 NLP

Novel Claim or Déjà Vu? Rethinking "Contamination-Free'' Dynamic Evaluation for Multimodal Automated Fact-Checking

This paper challenges the assumption that dynamic benchmarks are inherently contamination-free by demonstrating that a significant portion of post-cut-off claims remain verifiable via pre-existing knowledge, which can artificially inflate multimodal fact-checking performance and distort system rankings, thereby necessitating stricter evaluation protocols.

Haorui He, Xinwen Chen, Dacheng Wen, Reynold Cheng, Francis C. M. Lau, Yupeng Li2026-07-28✓ Author reviewed
⚛️ phenomenology

Baryogenesis from the Thermodynamic Arrow of Time: a Transfer-Function Bound and an Entropy-Clock Mechanism

This paper proposes a baryogenesis mechanism driven by an "entropy-clock" chemical potential during reheating, establishing a transfer-function bound that demonstrates successful freeze-out requires the entropy production timescale to overlap with the charge-violation scale, thereby constraining the reheating temperature to be comparable to the freeze-out temperature.

Yakov Mandel2026-07-28✓ Author reviewed
🔭 astrophysics

Enhancing weak lensing redshift distribution characterization by optimizing the Dark Energy Survey Self-Organizing Map Photo-z method

This paper demonstrates that optimizing the Dark Energy Survey's Self-Organizing Map photometric redshift method by tailoring the algorithm for redshift estimation and incorporating g-band flux information significantly improves redshift distribution characterization, reducing bin overlap by up to 66% and establishing a robust framework for future DES Year 6 and stage IV surveys.

A. Campos, B. Yin, S. Dodelson, A. Amon, A. Alarcon, C. Sánchez, G. M. Bernstein, G. Giannini, J. Myles, S. Samuroff, O. (…)2026-07-28✓ Author reviewed
🔭 astrophysics

Reconstructing PTA measurements via early seeding of supermassive black holes

This paper investigates how early seeding mechanisms for supermassive black holes, specifically comparing direct collapse black holes and Dark Star collapse, influence the nanohertz gravitational wave background detected by pulsar timing arrays, finding that Dark Star seeds with a specific number density could dominate the signal while offering a pathway to constrain seed populations.

Sohan Ghodla, Cosmin Ilie2026-07-28✓ Author reviewed
🤖 machine learning

Adaptive Data Admission and Retention for Streaming Federated Learning

This paper proposes an Active-Constraint Drift-Plus-Penalty (ACDPP) framework for streaming federated learning with limited client memory, which jointly optimizes server-side data admission and client-side retention to minimize cumulative excess population risk while satisfying sampling-cost and buffer constraints, achieving sublinear regret guarantees validated by experiments.

Zhuoyi Zhao, Ben Liang2026-07-28✓ Author reviewed
💬 NLP

Gubernaut: A Deterministic Homeostatic Controller for Affect-Regulated LLM Agents, Validated Across Independent Model Families

This paper introduces Gubernaut, a deterministic, model-agnostic runtime controller that uses a token-free meta-level monitoring affect telemetry to successfully regulate reactive failure modes in large language model agents across multiple frontier model families, as validated by a rigorous, pre-registered evaluation protocol.

Dushyant Sharma2026-07-28✓ Author reviewed