arXiv subjects

Jairo Diaz-Rodriguez

Publications and source records attributed to Jairo Diaz-Rodriguez.

18 recordsLinked to original sources

arXiv2026

How Closely Do LLM Reviews Align with Human Peer Review?

Large language models (LLMs) are increasingly used to generate scientific reviews, yet existing evaluations rarely examine whether different providers align with both conference decisions and human reviewing priorities within the same controlled setting. We compare reviews from OpenAI GPT-5.4, Google Gemini 3.1 Pro Preview, and Anthropic Claude Opus 4.6 with human reviews and final decisions for 300 topic-matched ICLR 2026 submissions, equally divided among oral, poster, and rejected papers. Each model reviewed every paper using identical instructions and rating scales after decision information was removed. Our study contributes a cross-provider analysis of three complementary dimensions: alignment with broad and fine-grained decision categories, differences in recommendation-scale usage, and thematic agreement in identified weaknesses. All three LLMs distinguished accepted from rejected papers, but none reproduced the oral versus poster distinction present in human ratings. Scoring patterns were provider-specific: Gemini assigned systematically higher ratings, while OpenAI and Claude were closer to humans for rejected and poster papers but more critical of oral papers. Human and LLM reviews also differed in emphasis, with LLMs more frequently identifying missing baseline comparisons and humans more often raising computational-efficiency concerns. These results show that broad decision alignment does not imply agreement with finer human judgments or reviewing priorities.

cs.CL↗

arXiv2026

How Much Structure Do LLMs Need? Evaluating LLMs for Bibliometric Cluster Description

Large language models (LLMs) can support scientific literature synthesis, but remain prone to hallucinated references, uneven coverage, and weakly grounded thematic organization. We evaluate whether bibliometric structure improves LLM-assisted synthesis by comparing six pipelines for generating cluster descriptions under different levels of evidence and structure. Using 100 published bibliometric analyses, we reconstruct Scopus corpora, extract human-written cluster descriptions, and assess outputs by human alignment, semantic coverage, clustering quality, graph quality, and reference grounding. Results show that LLMs produce descriptions semantically close to human-written ones, but are unreliable when asked to infer bibliometric structure from scratch. Performance improves when bibliometric algorithms define the clusters and the LLM interprets them. Overall, LLM-assisted bibliometric synthesis is most promising as a hybrid workflow in which algorithms provide auditable structure and LLMs generate readable descriptions.

cs.CL↗

arXiv2026

Position: Ideas Should be the Center of Machine Learning Research

Machine learning research increasingly bifurcates into two disconnected modes: benchmark-driven engineering that prioritizes metrics over understanding, and idealized theory that often fails to transfer to modern systems. In this position paper, we argue that the field focuses too heavily on these endpoints, neglecting the central scientific object: the idea. We propose an Ideas First framework in which ideas are valued for the behavioral signatures they predict in modern models, and these signatures are tested through tailored experiments designed to detect the relevant patterns rather than to win leaderboards. This shift not only bridges the gap between theory and practice but also promotes equity by removing the "complexity premium," enabling rigorous scientific contributions from researchers with modest computational, financial, and human resources. Ultimately, we advocate for a research culture centered on ideas, treating benchmarks and theorems as instruments for testing mechanistic hypotheses rather than as ends in themselves.

cs.LG↗

arXiv2026

When Can Digital Personas Reliably Approximate Human Survey Findings?

Digital personas powered by Large Language Models (LLMs) are increasingly proposed as substitutes for human survey respondents, yet it remains unclear when they can reliably approximate human survey findings. We answer this question using the LISS panel, constructing personas from respondents' background variables and pre-2023 survey histories, then testing them against the same respondents' held-out post-cutoff answers. Across four persona architectures, three LLMs, and two prediction tasks, we assess performance at the question, respondent, distributional, equity, and clustering levels. Digital personas improve alignment with human response distributions, especially in domains tied to stable attributes and values, but remain limited for individual prediction and fail to recover multivariate respondent structure. Retrieval-augmented architectures provide the clearest gains, but performance depends more on human response structure than on model choice: personas perform best for low-variability questions and common respondent patterns, and worst for subjective, heterogeneous, or rare responses. Our results provide practical guidance on when digital personas could be appropriate for survey research and when human validation remains necessary.

cs.CL↗

arXiv2026

Inheritance Between Feedforward and Convolutional Networks via Model Projection

Neural-network techniques are often transferred across architecture families by analogy, but such transfer is valid only when the assumptions required by a technique are preserved. We introduce this idea as inheritance between model classes. Using a unified node-level framework with tensor-valued activations, we prove that generalized feedforward networks (GFFNs) form a strict subset of generalized convolutional networks (GCNNs), so GCNN properties transfer directly to GFFNs. The reverse direction is not automatic: standard CNN nodes use spatial kernels, while FFN nodes use one scalar weight per input contribution. We introduce model projection to recover a restricted reverse inheritance path. Projection freezes each convolutional input-channel sub-function and learns one scalar coefficient for each input-output channel contribution, giving projected CNN nodes the GFFN-style trainable structure of scalar-weighted input recombination. This inherited structure leads naturally to parameter-efficient transfer learning. Across multiple ImageNet-pretrained CNN backbones and downstream image-classification datasets, model projection is competitive with standard and PEFT baselines and provides an effective initialization for subsequent full fine-tuning.

stat.ML↗

arXiv2026

Unsupervised Text Segmentation via Kernel Change-Point Detection on Sentence Embeddings

Unsupervised text segmentation is crucial because boundary labels are expensive, subjective, and often fail to transfer across domains and granularity choices. We propose Embed-KCPD, a training-free method that represents sentences as embedding vectors and estimates boundaries by minimizing a penalized KCPD objective. Beyond the algorithmic instantiation, we develop, to our knowledge, the first dependence-aware theory for KCPD under $m$-dependent sequences, a finite-memory abstraction of short-range dependence common in language. We prove an oracle inequality for the population penalized risk and a localization guarantee showing that each true change point is recovered within a window that is small relative to segment length. To connect theory to practice, we introduce an LLM-based simulation framework that generates synthetic documents with controlled finite-memory dependence and known boundaries, validating the predicted scaling behavior. Across standard segmentation benchmarks, Embed-KCPD often outperforms strong unsupervised baselines. A case study on Taylor Swift's tweets illustrates that Embed-KCPD combines strong theoretical guarantees, simulated reliability, and practical effectiveness for text segmentation.

cs.CL↗

arXiv2025

The Curved Spacetime of Transformer Architectures

We present a geometric framework for understanding Transformer-based language models, drawing an explicit analogy to General Relativity. Queries and keys induce an effective metric on representation space, and attention acts as a discrete connection that implements parallel transport of value vectors across tokens. Stacked layers provide discrete time-slices through which token representations evolve on this curved manifold, while backpropagation plays the role of a least-action principle that shapes loss-minimizing trajectories in parameter space. If this analogy is correct, token embeddings should not traverse straight paths in feature space; instead, their layer-wise steps should bend and reorient as interactions mediated by embedding space curvature. To test this prediction, we design experiments that expose both the presence and the consequences of curvature: (i) we visualize a curvature landscape for a full paragraph, revealing how local turning angles vary across tokens and layers; (ii) we show through simulations that excess counts of sharp/flat angles and longer length-to-chord ratios are not explainable by dimensionality or chance; and (iii) inspired by Einstein's eclipse experiment, we probe deflection under controlled context edits, demonstrating measurable, meaning-consistent bends in embedding trajectories that confirm attention-induced curvature.

cs.LG↗

arXiv2025

Structured Output Regularization: a framework for few-shot transfer learning

Traditional transfer learning typically reuses large pre-trained networks by freezing some of their weights and adding task-specific layers. While this approach is computationally efficient, it limits the model's ability to adapt to domain-specific features and can still lead to overfitting with very limited data. To address these limitations, we propose Structured Output Regularization (SOR), a simple yet effective framework that freezes the internal network structures (e.g., convolutional filters) while using a combination of group lasso and $L_1$ penalties. This framework tailors the model to specific data with minimal additional parameters and is easily applicable to various network components, such as convolutional filters or various blocks in neural networks enabling broad applicability for transfer learning tasks. We evaluate SOR on three few shot medical imaging classification tasks and we achieve competitive results using DenseNet121, and EfficientNetB4 bases compared to established benchmarks.

cs.CV↗

arXiv2025

Consistent Kernel Change-Point Detection under m-Dependence for Text Segmentation

Kernel change-point detection (KCPD) has become a widely used tool for identifying structural changes in complex data. While existing theory establishes consistency under independence assumptions, real-world sequential data such as text exhibits strong dependencies. We establish new guarantees for KCPD under $m$-dependent data: specifically, we prove consistency in the number of detected change points and weak consistency in their locations under mild additional assumptions. We perform an LLM-based simulation that generates synthetic $m$-dependent text to validate the asymptotics. To complement these results, we present the first comprehensive empirical study of KCPD for text segmentation with modern embeddings. Across diverse text datasets, KCPD with text embeddings outperforms baselines in standard text segmentation metrics. We demonstrate through a case study on Taylor Swift's tweets that KCPD not only provides strong theoretical and simulated reliability but also practical effectiveness for text segmentation tasks.

cs.LG↗

arXiv2025

Summaries as Centroids for Interpretable and Scalable Text Clustering

We introduce k-NLPmeans and k-LLMmeans, text-clustering variants of k-means that periodically replace numeric centroids with textual summaries. The key idea, summary-as-centroid, retains k-means assignments in embedding space while producing human-readable, auditable cluster prototypes. The method is LLM-optional: k-NLPmeans uses lightweight, deterministic summarizers, enabling offline, low-cost, and stable operation; k-LLMmeans is a drop-in upgrade that uses an LLM for summaries under a fixed per-iteration budget whose cost does not grow with dataset size. We also present a mini-batch extension for real-time clustering of streaming text. Across diverse datasets, embedding models, and summarization strategies, our approach consistently outperforms classical baselines and approaches the accuracy of recent LLM-based clustering-without extensive LLM calls. Finally, we provide a case study on sequential text streams and release a StackExchange-derived benchmark for evaluating streaming text clustering.

cs.CL↗

arXiv2025

Dynamics of Spontaneous Topic Changes in Next Token Prediction with Self-Attention

Human cognition is punctuated by abrupt, spontaneous shifts between topics-driven by emotional, contextual, or associative cues-a phenomenon known as spontaneous thought in neuroscience. In contrast, self-attention based models depend on structured patterns over their inputs to predict each next token, lacking spontaneity. Motivated by this distinction, we characterize spontaneous topic changes in self-attention architectures, revealing both their similarities and their divergences from spontaneous human thought. First, we establish theoretical results under a simplified, single-layer self-attention model with suitable conditions by defining the topic as a set of Token Priority Graphs (TPGs). Specifically, we demonstrate that (1) the model maintains the priority order of tokens related to the input topic, (2) a spontaneous topic change can occur only if lower-priority tokens outnumber all higher-priority tokens of the input topic, and (3) unlike human cognition, the longer context length or the more ambiguous input topic reduces the likelihood of spontaneous change. Second, we empirically validate that these dynamics persist in modern, state-of-the-art LLMs, underscoring a fundamental disparity between human cognition and AI behaviour in the context of spontaneous topic changes. To the best of our knowledge, no prior work has explored these questions with a focus as closely aligned to human thought.

cs.CL↗

arXiv2024

Differentially Private Boxplots

Despite the potential of differentially private data visualization to harmonize data analysis and privacy, research in this area remains underdeveloped. Boxplots are a widely popular visualization used for summarizing a dataset and for comparison of multiple datasets. Consequentially, we introduce a differentially private boxplot. We evaluate its effectiveness for displaying location, scale, skewness and tails of a given empirical distribution. In our theoretical exposition, we show that the location and scale of the boxplot are estimated with optimal sample complexity, and the skewness and tails are estimated consistently, which is not always the case for a boxplot naively constructed from a single existing differentially private quantile algorithm. As a byproduct of this exposition, we introduce several new results concerning private quantile estimation. In simulations, we show that this boxplot performs similarly to a non-private boxplot, and it outperforms the naive boxplot. Additionally, we conduct a real data analysis of Airbnb listings, which shows that comparable analysis can be achieved through differentially private boxplot visualization.

stat.ME↗

arXiv2024

Tomographic reconstruction of a disease transmission landscape via GPS recorded random paths

Identifying areas in a landscape where individuals have a higher likelihood of disease infection is key to managing diseases. Unlike conventional methods relying on ecological assumptions, we perform a novel epidemiological tomography for the estimation of landscape propensity to disease infection, using GPS animal tracks in a manner analogous to tomographic techniques in positron emission tomography (PET). Treating tracking data as random Radon transforms, we analyze Cervid movements in a game preserve, paired with antibody levels for epizootic hemorrhagic disease virus (EHDV) -- a vector-borne disease transmitted by biting midges. After discretizing the field and building the regression matrix of the time spent by each deer (row) at each point of the lattice (column), we model the binary response (infected or not) as a binomial linear inverse problem where spatial coherence is enforced with a total variation regularization. The smoothness of the reconstructed propensity map is selected by the quantile universal threshold. To address limitations of small sample sizes and evaluate significance of our estimates, we quantify uncertainty using a bootstrap-based data augmentation procedure. Our method outperforms alternative ones when using simulated and real data. This tomographic framework is novel, with no established statistical methods tailored for such data.

stat.AP↗

arXiv2023

Adjoint-Free 4D-Var Methods Via Line Search Optimization For Non-Linear Data Assimilation

This paper proposes two practical implementations of Four-Dimensional Variational (4D-Var) Ensemble Kalman Filter (4D-EnKF) methods for non-linear data assimilation. Our formulations' main idea is to avoid the intrinsic need for adjoint models in the context of 4D-Var optimization and, even more, to handle non-linear observation operators during the assimilation of observations. The proposed methods work as follows: snapshots of an ensemble of model realizations are taken at observation times, these snapshots are employed to build control spaces onto which analysis increments can be estimated. Via the linearization of observation operators at observation times, a line-search based optimization method is proposed to estimate optimal analysis increments. The convergence of this method is theoretically proven as long as the dimension of control-spaces equals model one. In the first formulation, control spaces are given by full-rank square root approximations of background error covariance matrices via the Bickel and Levina precision matrix estimator. In this context, we propose an iterative Woodbury matrix formula to perform the optimization steps efficiently. The last formulation can be considered as an extension of the Maximum Likelihood Ensemble Filter to the 4D-Var context. This employs pseudo-square root approximations of prior error covariance matrices to build control spaces. Experimental tests are performed by using the Lorenz 96 model. The results reveal that, in terms of Root-Mean-Square-Error values, both methods can obtain reasonable estimates of posterior error modes in the 4D-Var optimization problem. Moreover, the accuracies of the proposed filter implementations can be improved as the ensemble sizes are increased.

stat.AP↗

arXiv2018

Thresholding tests

We derive a new class of statistical tests for generalized linear models based on thresholding point estimators. These tests can be employed whether the model includes more parameters than observations or not. For linear models, our tests rely on pivotal statistics derived from model selection techniques. Affine lasso, a new extension of lasso, allows to unveil new tests and to develop in the same framework parametric and nonparametric tests. Our tests for generalized linear models are based on new asymptotically pivotal statistics. A composite thresholding test attempts to achieve uniformly most power under both sparse and dense alternatives with success. In a simulation, we compare the level and power of these tests under sparse and dense alternative hypotheses. The thresholding tests have a better control of the nominal level and higher power than existing tests.

stat.ME↗

arXiv2017

Quantile universal threshold for model selection

Efficient recovery of a low-dimensional structure from high-dimensional data has been pursued in various settings including wavelet denoising, generalized linear models and low-rank matrix estimation. By thresholding some parameters to zero, estimators such as lasso, elastic net and subset selection allow to perform not only parameter estimation but also variable selection, leading to sparsity. Yet one crucial step challenges all these estimators: the choice of the threshold parameter~$λ$. If too large, important features are missing; if too small, incorrect features are included. Within a unified framework, we propose a new selection of $λ$ at the detection edge under the null model. To that aim, we introduce the concept of a zero-thresholding function and a null-thresholding statistic, that we explicitly derive for a large class of estimators. The new approach has the great advantage of transforming the selection of $λ$ from an unknown scale to a probabilistic scale with the simple selection of a probability level. Numerical results show the effectiveness of our approach in terms of model selection and prediction.

stat.ME↗

arXiv2017

Nonparametric estimation of galaxy cluster's emissivity and point source detection in astrophysics with two lasso penalties

Astrophysicists are interested in recovering the 3D gas emissivity of a galaxy cluster from a 2D image taken by a telescope. A blurring phenomenon and presence of point sources make this inverse problem even harder to solve. The current state-of-the-art technique is two step: first identify the location of potential point sources, then mask these locations and deproject the data. We instead model the data as a Poisson generalized linear model (involving blurring, Abel and wavelets operators) regularized by two lasso penalties to induce sparse wavelet representation and sparse point sources. The amount of sparsity is controlled by two quantile universal thresholds. As a result, our method outperforms the existing one.

stat.AP↗

arXiv2014

Quantile universal threshold: model selection at the detection edge for high-dimensional linear regression

To estimate a sparse linear model from data with Gaussian noise, consilience from lasso and compressed sensing literatures is that thresholding estimators like lasso and the Dantzig selector have the ability in some situations to identify with high probability part of the significant covariates asymptotically, and are numerically tractable thanks to convexity. Yet, the selection of a threshold parameter $λ$ remains crucial in practice. To that aim we propose Quantile Universal Thresholding, a selection of $λ$ at the detection edge. We show with extensive simulations and real data that an excellent compromise between high true positive rate and low false discovery rate is achieved, leading also to good predictive risk.

stat.ML↗