arXiv subjects

Lijin Zhang

Publications and source records attributed to Lijin Zhang.

4 recordsLinked to original sources

arXiv2026

AI Cartography: Mapping the Latent Landscape of AI Benchmark Ecosystems

While aggregate leaderboard scores drive AI development, they contain substantial measurement noise whose sources and magnitudes remain unquantified, making it unclear when rankings reflect genuine capability differences versus evaluation artifacts. We introduce a framework for measuring the latent landscape in AI benchmark ecosystems. Applying Confirmatory Factor Analysis (CFA) and Generalizability Theory to 4,000+ models from the Open LLM Leaderboard, we decompose sources of ranking variance and establish: (1) structures assumed in current reporting practice underestimate the strength of relationships between benchmarks; (2) evidence of local dependence among leaderboard items, undermining uses of benchmarks as measurement instruments under current scoring systems; (3) contributor metadata explains more rank-relevant variance ($\approx9\%$) than architecture or deployment categories in this context; (4) a manifest-score "scaling law" slope has low reliability ($R_{\beta}=0.53$); by contrast, the latent general-factor size slope is highly stable across ecosystem controls ($R_g=0.97$). We are able to provide unique insights into benchmark dynamics, such as which benchmarks are a function of LLM size and which can be oppositely impacted by post-training practices. We provide actionable diagnostics to determine how benchmark rankings can be trusted and how benchmark design can be improved.

cs.AI↗

arXiv2026

Understanding Student Effort Using Response-Time Propensities During Problem Solving

Adaptive learning systems can produce substantial learning gains, yet many students engage for too brief or too superficial a period to benefit. A central obstacle is measuring effort. Effort during multi-step problem solving is rarely directly observed, and common log-based proxies, such as time on task, cannot distinguish between a student working carefully and a student encountering a harder problem. We examine step-to-step response time as a scalable effort signal by modeling trait-like differences in students' typical response timing during tutoring (while adjusting for skill difficulty). Using step-level logs from eight classroom deployments of algebra tutoring systems (2020 to 2023) across six U.S. schools (794 students), we estimate student- and knowledge-component-level propensities using hierarchical models and relate them to learning efficiency, defined as performance improvement per completed solution step. Response-time propensities show moderate to strong stability within students, supporting their use as an individual differences measure beyond correctness. At the same time, their relationship to learning is not uniform but conditional on the learner and context. Slower propensities predict greater learning efficiency for higher-proficiency students, consistent with constructive processing, whereas for lower-proficiency students, slower propensities are weakly related or even negative, consistent with unproductive struggle or idling. These associations are strongest early in practice sequences and attenuate later in the class period, highlighting an actionable window for detecting emerging disengagement and low persistence. Overall, response-time propensities provide a practical way to incorporate temporal process data into learner models and to target adaptive supports when effort is most diagnostic.

cs.CY↗

arXiv2025

Polytomous Explanatory Item Response Models for Item Discrimination: Assessing Negative-Framing Effects in Social-Emotional Learning Surveys

Modeling item parameters as a function of item characteristics has a long history but has generally focused on models for item location. Explanatory item response models for item discrimination are available but rarely used. In this study, we extend existing approaches for modeling item discrimination from dichotomous to polytomous item responses. We illustrate our proposed approach with an application to four social-emotional learning surveys of preschool children to investigate how item discrimination depends on whether an item is positively or negatively framed. Negative framing predicts significantly lower item discrimination on two of the four surveys, and a plausibly causal estimate from a regression discontinuity analysis shows that negative framing reduces discrimination by about 30\% on one survey. We conclude with a discussion of potential applications of explanatory models for item discrimination.

stat.ME↗

arXiv2022

Convergence of position-dependent MALA with application to conditional simulation in GLMMs

We establish conditions under which Metropolis-Hastings (MH) algorithms with a position-dependent proposal covariance matrix will or will not have the geometric rate of convergence. Some of the diffusions based MH algorithms like the Metropolis adjusted Langevin algorithm (MALA) and the pre-conditioned MALA (PCMALA) have a position-independent proposal variance. Whereas, for other modern variants of MALA like the manifold MALA (MMALA) that adapt to the geometry of the target distributions, the proposal covariance matrix changes in every iteration. Thus, we provide conditions for geometric ergodicity of different variations of the Langevin algorithms. These results have important practical implications as these provide crucial justification for the use of asymptotically valid Monte Carlo standard errors for Markov chain based estimates. The general conditions are verified in the context of conditional simulation from the two most popular generalized linear mixed models (GLMMs), namely the binomial GLMM with the logit link and the Poisson GLMM with the log link. Empirical comparison in the framework of some spatial GLMMs shows that the computationally less expensive PCMALA with an appropriately chosen pre-conditioning matrix may outperform the MMALA.

stat.ME↗