Papers where the original authors have engaged with our plain-language explanation.
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.
977 papers reviewed by authors · 611–620 / 977
This paper introduces a rigorous and computationally efficient measure of model complexity based on input gradient similarities that unifies various existing metrics and provides new insights into the double descent phenomenon across diverse model architectures.
This paper presents a tiered KV cache architecture that enables runtime-certified bounded-error quantized attention by computing online error bounds to trigger adaptive precision selection and deterministic FP16 fallback, thereby guaranteeing recovery to exact dense attention outputs while maintaining high compression for long-context LLM inference.
This study identifies the subgiant star TOI-5882 as a strong candidate for planetary engulfment, evidenced by its significant lithium enrichment which modeling suggests could result from the ingestion of a super-Earth to Neptune-mass planet.
The paper introduces BALLAST, a Bayesian active learning framework that optimizes the placement of Lagrangian sea drifters for inferring time-dependent ocean vector fields by incorporating look-ahead trajectory predictions and a novel efficient Gaussian Process inference method called VaSE.
The paper introduces Sutra, a purely functional programming language that compiles symbolic Vector Symbolic Architecture operations into fused PyTorch tensor graphs, enabling programs to achieve perfect decoding accuracy across diverse frozen embedding substrates and to be trained via backpropagation while remaining fully legible and recompilable as source code.
This paper presents a unified, multi-layer testing framework for cloud-native ELT pipelines that integrates orchestration-level validation, declarative dbt tests, and LLM-generated semantic tests, demonstrating through controlled experiments that this approach achieves a 128.57% improvement in anomaly detection over manual baselines while maintaining operational practicality.
The paper introduces torchtune, a PyTorch-native library designed to streamline the post-training lifecycle of large language models by prioritizing modularity, transparency, and extensibility to enable efficient fine-tuning and rapid research iteration while maintaining competitive performance and memory efficiency.
This paper proposes a novel Fuzzy Convolution Neural Network (FCNN) framework that converts tabular data into fuzzy membership-based images to effectively leverage deep learning for classification, demonstrating competitive or superior performance compared to traditional machine learning algorithms on complex noisy datasets.
This paper proposes that black holes function as condensates of spacetime's thermodynamic degrees of freedom, offering a coherent interpretation of their mass, entropy, and interior structure that is supported by recent black hole merger observations.
This paper presents complete analytic solutions for coherent couplers with arbitrary parameters using Weierstrass elliptic functions, identifies Jensen's coupler as a special case, and establishes a projection from the three-mode degenerate four-wave mixing system to the two-mode coupler, revealing a deeper connection to integrable parametric processes and Kronecker theta functions.