arXiv subjects

Renato Vicente

Publications and source records attributed to Renato Vicente.

At least 19 recordsLinked to original sources

arXiv2026

Kolmogorov--Arnold Networks for Small Language Models

Kolmogorov--Arnold Networks (KANs) replace fixed node activations with learned one-dimensional edge functions, offering an explicit interface for interpretation and a possible alternative to transformer feed-forward networks. We test these claims separately. In a six-layer, 10M-parameter B-spline KAN, we reconstruct all 884,736 feed-forward edges: 87.8\% exceed (NLS>0.1) and 0.4\% are inactive. Pruning the lowest-activity 20--25\% causes negligible loss increase, although structured MLP neuron pruning tolerates comparable sparsity. The audit replicates on BabyLM, but grid-size sweeps show that near-total fPCA compression and high closed-form-fit coverage are properties of the low-capacity grid-2 basis, not universal KAN behavior. For replacement, we evaluate MLP, SwiGLU, grouped Chebyshev, and rational GR-KAN networks on BabyLM. The KAN-family and gated variants improve validation loss over the GELU MLP, but this ordering does not transfer to standardized benchmarks: across ten seeds and 59,875 BLiMP pairs, accuracies span 62.4--63.1\%, EWoK remains at chance, and a (+0.7)-point GR-KAN effect on BLiMP reverses on the supplement. Larger tests are also cautionary: parameter-matched MLPEdge underperforms the MLP on Wikitext-103, and 286M-parameter GR-KAN remains below a SwiGLU ClimbMix baseline after stabilization. Thus, small-basis KANs provide a practical, corpus-transferable interface for auditing learned scalar transformations, but the tested replacements show no consistent benchmark, quality, or latency advantage over strong MLP baselines.

cs.LG↗

arXiv2026

Persona-Model Collapse in Emergent Misalignment

Fine-tuning large language models on narrow data with harmful content produces broadly misaligned behavior on unrelated prompts, a phenomenon known as emergent misalignment. We propose that emergent misalignment involves persona-model collapse: deterioration of the model's internal capacity to simulate, differentiate, and maintain consistent characters. We test this hypothesis behaviorally using two metrics: moral susceptibility (S) and moral robustness (R), computed from the across- and within-persona variability of models' Moral Foundations Questionnaire responses under persona role-play. These metrics formalize the model's ability to differentiate characters (S) and its consistency when simulating a given one (R). We evaluate four frontier models (DeepSeek-V3.1, GPT-4.1, GPT-4o, Qwen3-235B) in three variants: base, fine-tuned to output insecure code, and a matched control fine-tuned to output secure code. Across the four models, insecure fine-tuning produces an average $55\%$ increase in S, pushing all four insecure variants beyond the band observed across 13 frontier models benchmarked in prior work -- with GPT-4o reaching more than twice the band's upper end -- signaling dysregulated differentiation. It also causes an average $65\%$ decrease in R, equivalent to a $304\%$ increase in 1/R. By contrast, the matched secure control preserves S near the base and induces only a partial R loss, showing that these effects are largely misalignment-specific. Complementing these metric shifts, insecure variants' unconditioned responses converge toward saturation near the scale ceiling, departing markedly from both base models' structured responses and those elicited when base models role-play toxic personas. Taken together, these metrics provide a sensitive diagnostic for emergent misalignment and serve as behavioral evidence that it involves persona-model collapse.

cs.CL↗

arXiv2026

Deceive, Detect, and Disclose: Large Language Models Play Mini-Mafia

Large language models are increasingly deployed in multi-agent settings whose outcomes hinge on social intelligence, motivating evaluations of their interactive capabilities; yet existing studies remain overwhelmingly empirical, leaving us without a theoretical understanding of how agent interactions determine collective outcomes. To address this, we introduce \textit{Mini-Mafia}, a four-player simplification of the social deduction game Mafia in which a fixed night phase reduces the game to a single critical exchange among a mafioso, a detective, and a villager. In this setting, we show that the mafia win-rate $p$ is predicted by the analytical formula $\text{logit}(p) = v \times (m - d)$, where $m$, $d$, and $v$ represent the mafioso's deception, the detective's disclosure, and the villager's detection capabilities. We turn this analytical framework into the \textit{Mini-Mafia Benchmark}, where Bayesian inference over gameplay data yields per-model estimates of the intrinsic parameters $m$, $d$, and $v$. For $I$ models, only $3I$ parameters suffice to predict the outcomes of all $I^3$ tournament combinations; and in 5-fold cross-validation the formula achieves a $76.6\%$ Brier-score reduction over a random baseline. The benchmark also reveals counterintuitive results: Grok 3 Mini is the strongest detector and GPT-5 Mini the strongest discloser, both ahead of DeepSeek V3.1, Claude Opus 4, and Claude Sonnet 4; while Claude Sonnet 4 is the weakest detector, near random chance. Together, these results show that Mini-Mafia, a simple but nontrivial multi-agent system, admits an analytical description and serves as a principled benchmark for language model interactions.

cs.AI↗

arXiv2026

Moral Susceptibility and Robustness under Persona Role-Play in Large Language Models

Large language models (LLMs) increasingly operate in social contexts, motivating analysis of how they express and shift moral judgments. In this work, we investigate the moral response of LLMs to persona role-play, prompting a LLM to assume a specific character. Using the Moral Foundations Questionnaire (MFQ), we introduce a benchmark that quantifies two properties: moral susceptibility and moral robustness, defined from the variability of MFQ scores across- and within-personas. We estimate these quantities with two complementary procedures, repeated sampling and a logit-based method that directly estimates the rating distributions and enables temperature analysis. We evaluate 15 models across six families: Claude, DeepSeek, Gemini, GPT, Grok, and Llama. The two metrics show qualitatively different patterns. Moral robustness varies by more than an order of magnitude, with a coefficient of variation of about $152\%$, and is explained almost entirely by model family. The Claude family is, by a significant margin, the most robust, about 30 times more so than the lower-performing families (DeepSeek, Grok, and Llama), while Gemini and GPT occupy an intermediate tier. This strong family dependence suggests that robustness is primarily shaped by post-training. Moral susceptibility, by contrast, spans a much narrower range, with a coefficient of variation of about $13\%$, and the most susceptible model is only 1.6 times more susceptible than the least. Unlike robustness, susceptibility shows no clear family dependence, suggesting that it is primarily determined by pre-training. Additionally, we present moral foundation profiles for models without persona role-play and for personas averaged across models. Together, these analyses provide a systematic view of how persona conditioning shapes moral behavior in LLMs and a window into the internal machinery they use to instantiate personas.

cs.CL↗

arXiv2023

A unified framework for dataset shift diagnostics

Supervised learning techniques typically assume training data originates from the target population. Yet, in reality, dataset shift frequently arises, which, if not adequately taken into account, may decrease the performance of their predictors. In this work, we propose a novel and flexible framework called DetectShift that quantifies and tests for multiple dataset shifts, encompassing shifts in the distributions of $(X, Y)$, $X$, $Y$, $X|Y$, and $Y|X$. DetectShift equips practitioners with insights into data shifts, facilitating the adaptation or retraining of predictors using both source and target data. This proves extremely valuable when labeled samples in the target domain are limited. The framework utilizes test statistics with the same nature to quantify the magnitude of the various shifts, making results more interpretable. It is versatile, suitable for regression and classification tasks, and accommodates diverse data forms - tabular, text, or image. Experimental results demonstrate the effectiveness of DetectShift in detecting dataset shifts even in higher dimensions.

stat.ML↗

arXiv2022

Restricted Boltzmann Machine Flows and The Critical Temperature of Ising models

We explore alternative experimental setups for the iterative sampling (flow) from Restricted Boltzmann Machines (RBM) mapped on the temperature space of square lattice Ising models by a neural network thermometer. This framework has been introduced to explore connections between RBM-based deep neural networks and the Renormalization Group (RG). It has been found that, under certain conditions, the flow of an RBM trained with Ising spin configurations approaches in the temperature space a value around the critical one: $ k_B T_c / J \approx 2.269$. In this paper we consider datasets with no information about model topology to argue that a neural network thermometer is not an accurate way to detect whether the RBM has learned scale invariance or not.

cond-mat.stat-mech↗

arXiv2022

Effective Sample Size, Dimensionality, and Generalization in Covariate Shift Adaptation

In supervised learning, training and test datasets are often sampled from distinct distributions. Domain adaptation techniques are thus required. Covariate shift adaptation yields good generalization performance when domains differ only by the marginal distribution of features. Covariate shift adaptation is usually implemented using importance weighting, which may fail, according to common wisdom, due to small effective sample sizes (ESS). Previous research argues this scenario is more common in high-dimensional settings. However, how effective sample size, dimensionality, and model performance/generalization are formally related in supervised learning, considering the context of covariate shift adaptation, is still somewhat obscure in the literature. Thus, a main challenge is presenting a unified theory connecting those points. Hence, in this paper, we focus on building a unified view connecting the ESS, data dimensionality, and generalization in the context of covariate shift adaptation. Moreover, we also demonstrate how dimensionality reduction or feature selection can increase the ESS, and argue that our results support dimensionality reduction before covariate shift adaptation as a good practice.

stat.ML↗

arXiv2021

An experiment on the mechanisms of racial bias in ML-based credit scoring in Brazil

We dissect an experimental credit scoring model developed with real data and demonstrate - without access to protected attributes - how the use of location information introduces racial bias. We analyze the tree gradient boosting model with the aid of a game-theoretic inspired machine learning explainability technique, counterfactual experiments and Brazilian census data. By exposing algorithmic racial bias explaining the trained machine learning model inner mechanisms, this experiment comprises an interesting artifact to aid the endeavor of theoretical understanding of the emergence of racial bias in machine learning systems. Without access to individuals' racial categories, we show how classification parity measures using geographically defined groups could carry information about model racial bias. The experiment testifies to the need for methods and language that do not presuppose access to protected attributes when auditing ML models, the importance of considering regional specifics when addressing racial issues, and the central role of census data in the AI research community. To the best of our knowledge, this is the first documented case of algorithmic racial bias in ML-based credit scoring in Brazil, the country with the second largest Black population in the world.

cs.CY↗

arXiv2021

LegalNLP -- Natural Language Processing methods for the Brazilian Legal Language

We present and make available pre-trained language models (Phraser, Word2Vec, Doc2Vec, FastText, and BERT) for the Brazilian legal language, a Python package with functions to facilitate their use, and a set of demonstrations/tutorials containing some applications involving them. Given that our material is built upon legal texts coming from several Brazilian courts, this initiative is extremely helpful for the Brazilian legal field, which lacks other open and specific tools and language models. Our main objective is to catalyze the use of natural language processing tools for legal texts analysis by the Brazilian industry, government, and academia, providing the necessary tools and accessible material.

cs.CL↗

arXiv2021

Effects of personality traits in predicting grade retention of Brazilian students

Student's grade retention is a key issue faced by many education systems, especially those in developing countries. In this paper, we seek to gauge the relevance of students' personality traits in predicting grade retention in Brazil. For that, we used data collected in 2012 and 2017, in the city of Sertaozinho, countryside of the state of Sao Paulo, Brazil. The surveys taken in Sertaozinho included several socioeconomic questions, standardized tests, and a personality test. Moreover, students were in grades 4, 5, and 6 in 2012. Our approach was based on training machine learning models on the surveys' data to predict grade retention between 2012 and 2017 using information from 2012 or before, and then using some strategies to quantify personality traits' predictive power. We concluded that, besides proving to be fairly better than a random classifier when isolated, personality traits contribute to prediction even when using socioeconomic variables and standardized tests results.

stat.AP↗

arXiv2020

Age-structured estimation of COVID-19 ICU demand from low quality data

We sample aggravated cases following age-structured probabilities from confirmed cases and use ICU occupation data to find a subnotification factor. A logistic fit is then employed to project the progression of the COVID-19 epidemic with plateau scenarios taken from locations that have reached this stage. Finally, the logistic curve found is corrected by the subnotification factor and sampled to project the future demand for ICU beds.

physics.soc-ph↗

arXiv2016

Information inefficiency in a random linear economy model

We study the effects of introducing information inefficiency in a model for a random linear economy with a representative consumer. This is done by considering statistical, instead of classical, economic general equilibria. Employing two different approaches we show that inefficiency increases the consumption set of a consumer but decreases her expected utility. In this scenario economic activity grows while welfare shrinks, that is the opposite of the behavior obtained by considering a rational consumer.

q-fin.GN↗

arXiv2016

Phase transition from egalitarian to hierarchical societies driven by competition between cognitive and social constraints

Empirical evidence suggests that social structure may have changed from hierarchical to egalitarian and back along the evolutionary line of humans. We model a society subject to competing cognitive and social navigation constraints. The theory predicts that the degree of hierarchy decreases with encephalization and increases with group size. Hence hominin groups may have been driven from a phase with hierarchical order to a phase with egalitarian structures by the encephalization during the last two million years, and back to hierarchical due to fast demographical changes during the Neolithic. The dynamics in the perceived social network shows evidence in the egalitarian phase of the observed phenomenon of Reverse Dominance. The theory also predicts for modern hunter-gatherers in mild climates a trend towards an intermediate hierarchy degree and a phase transition for harder ecological conditions. In harsher climates societies would tend to bemore egalitarian if organized in small groups but more hierarchical if in large groups. The theoretical model permits organizing the available data in the cross-cultural record (Ethnographic Atlas, N=248 cultures) where the symmetry breaking transition can be clearly seen.

physics.soc-ph↗

arXiv2015

For whom will the Bayesian agents vote?

Within an agent-based model where moral classifications are socially learned, we ask if a population of agents behaves in a way that may be compared with conservative or liberal positions in the real political spectrum. We assume that agents first experience a formative period, in which they adjust their learning style acting as supervised Bayesian adaptive learners. The formative phase is followed by a period of social influence by reinforcement learning. By comparing data generated by the agents with data from a sample of 15000 Moral Foundation questionnaires we found the following. 1. The number of information exchanges in the formative phase correlates positively with statistics identifying liberals in the social influence phase. This is consistent with recent evidence that connects the dopamine receptor D4-7R gene, political orientation and early age social clique size. 2. The learning algorithms that result from the formative phase vary in the way they treat novelty and corroborative information with more conservative-like agents treating it more equally than liberal-like agents. This is consistent with the correlation between political affiliation and the Openness personality trait reported in the literature. 3. Under the increase of a model parameter interpreted as an external pressure, the statistics of liberal agents resemble more those of conservative agents, consistent with reports on the consequences of external threats on measures of conservatism. We also show that in the social influence phase liberal-like agents readapt much faster than conservative-like agents when subjected to changes on the relevant set of moral issues. This suggests a verifiable dynamical criterium for attaching liberal or conservative labels to groups.

physics.soc-ph↗

arXiv2013

Moral foundations in an interacting neural networks society

The moral foundations theory supports that people, across cultures, tend to consider a small number of dimensions when classifying issues on a moral basis. The data also show that the statistics of weights attributed to each moral dimension is related to self-declared political affiliation, which in turn has been connected to cognitive learning styles by recent literature in neuroscience and psychology. Inspired by these data, we propose a simple statistical mechanics model with interacting neural networks classifying vectors and learning from members of their social neighborhood about their average opinion on a large set of issues. The purpose of learning is to reduce dissension among agents even when disagreeing. We consider a family of learning algorithms parametrized by δ, that represents the importance given to corroborating (same sign) opinions. We define an order parameter that quantifies the diversity of opinions in a group with homogeneous learning style. Using Monte Carlo simulations and a mean field approximation we find the relation between the order parameter and the learning parameter δat a temperature we associate with the importance of social influence in a given group. In concordance with data, groups that rely more strongly on corroborating evidence sustains less opinion diversity. We discuss predictions of the model and propose possible experimental tests.

physics.soc-ph↗

arXiv2013

Statistical mechanics of reputation systems in autonomous networks

Reputation systems seek to infer which members of a community can be trusted based on ratings they issue about each other. We construct a Bayesian inference model and simulate approximate estimates using belief propagation (BP). The model is then mapped onto computing equilibrium properties of a spin glass in a random field and analyzed by employing the replica symmetric cavity approach. Having the fraction of trustful nodes and environment noise level as control parameters, we evaluate the theoretical performance in terms of estimation error and the robustness of the BP approximation in different scenarios. Regions of degraded performance are then explained by the convergence properties of the BP algorithm and by the emergence of a glassy phase.

cond-mat.dis-nn↗

arXiv2012

Invasion, polymorphic equilibria and fixation of a mutant social allele in group structured populations

Stable mixtures of cooperators and defectors are often seen in nature. This fact is at odds with predictions based on linear public goods games under weak selection. That model implies fixation either of cooperators or of defectors, and the former scenario requires a level of group relatedness larger than the cost/benefit ratio, being therefore expected only if there is either kin recognition or a very low cost/benefit ratio, or else under stringent conditions with low gene flow. This motivated us to study here social evolution in a large class of group structured populations, with arbitrary multi-individual interactions among group members and random migration among groups. Under the assumption of weak selection, we analyze the equilibria and their stability. For some significant models of social evolution with non-linear fitness functions, including contingent behavior in iterated public goods games and threshold models, we show that three regimes occur, depending on the migration rate among groups. For sufficiently high migration rates, a rare cooperative allele A cannot invade a monomorphic population of asocial alleles N. For sufficiently low values of the migration rate, allele A can invade the population, when rare, and then fixate, eliminating N. For intermediate values of the migration rate, allele A can invade the population, when rare, producing a polymorphic equilibrium, in which it coexists with N. The equilibria and their stability do not depend on the details of the population structure. The levels of migration (gene flow) and group relatedness that allow for invasion of the cooperative allele leading to polymorphic equilibria with the non-cooperative allele are common in nature.

q-bio.PE↗

arXiv2012

The Taylor-Frank method cannot be applied to some biologically important, continuous fitness functions

The Taylor-Frank method for making kin selection models when fitness is a nonlinear function of a continuous phenotype requires this function to be differentiable. This assumption sometimes fails for biologically important fitness functions, for instance in microbial data and the theory of repeated n-person games, even when fitness functions are smooth and continuous. In these cases, the Taylor-Frank methodology cannot be used, and a more general form of direct fitness must replace the standard one to account for kin selection, even under weak selection.

q-bio.PE↗