Sección

Investigación

Papers y preprints de los feeds de investigación que seguimos.

6265documentos

arXiv AI Papershace 7 días

Reconstructing LLM Lineage Trees via Fine-Tuning Path Inference

Fine-tuned (FT) derivatives of large language models (LLMs) are now released in large numbers, yet their provenance relies on self-reports and cannot be verified. Misattributed lineage leaves a verifier unable to audit outsourced fine-tuning, confirm license compliance of derived models, or trace vulnerabilities that a parent passes to its children. All three require the direction of derivation, not merely its presence. Existing output-only methods are limited to pairwise derivation tests or undirected lineage estimation, while recovering a directed tree requires white-box access to model weights. We formalize the problem of reconstructing a directed model tree, including internal nodes, under gray-box access, where only teacher-forced log-likelihoods are observed and weights are never inspected. Fine-tuning raises the log-likelihood of the training data, and a child inherits this increase from its parent. Building on this property, we define an asymmetric per-task log-likelihood deviation between any two models as a trace of fine-tuning history, and derive from it an edge weight for each ordered pair of models. We propose TRACE-Edge, which exactly solves for the maximum spanning directed tree under these edge weights, with neither hyperparameters nor tree-structure constraints. We evaluate it on five configurations totaling 25 model trees built from three base models and seven tasks. TRACE-Edge achieves a parent-match accuracy of 0.74 when probed with the fine-tuning examples, and on average outperforms an output-similarity baseline even with held-out probes drawn from the same distribution as the training data.

en

arXiv AI Papershace 7 días

Rethinking Agent Trajectory Compression: A Workflow-Aware Perspective

As long-horizon agents increasingly rely on multi-step tool interactions to accomplish complex tasks, their execution trajectories continuously accumulate structured outputs and intermediate artifacts, leading to growing context overhead. Existing compression methods primarily rely on text importance or fixed truncation rules, often overlooking cross-step dependencies within workflows and potentially discarding critical information required for subsequent task execution. To address this challenge, we propose WAAC, a workflow-aware framework for agent trajectory compression. WAAC structurally models successful historical trajectories to extract reusable workflow compression experience and incorporates online backward dependency analysis to dynamically identify information retention requirements. By combining current dependency constraints with historical experience, WAAC selects appropriate compression strategies for different types of historical information. Experiments on PinchBench and Odysseys demonstrate that WAAC achieves an overall score of 87.6 and a perfect completion rate of 19.0%, respectively, while reducing execution costs by 19.7% and 29.1% compared with the strongest baseline methods. These results demonstrate the effectiveness of workflow-aware trajectory compression in improving task performance and cost efficiency for long-horizon agents. Code and resources are available at https://anonymous.4open.science/r/waac-6527/.

en

arXiv AI Papershace 7 días

Adaptive Scheduling Pipeline for Multi-Instance Asynchronous Reinforcement Learning

We present the Adaptive Scheduling Pipeline (ASP), an offline rollout scheduler for multi-instance asynchronous reinforcement learning (RL) of large language models (LLMs). ASP jointly reduces three sources of rollout underutilization - padding bubbles from intra-minibatch length heterogeneity, skewness bubbles from long-tail outputs, and sorting bubbles from fluctuations in training-queue occupancy induced by length-sorted dispatch - through a single offline preprocessing pass. This enables efficient multi-instance rollout that overcomes the early scaling plateau of single-instance inference while smoothing the arrival of generated samples at the trainer. We integrate the scheduler into an asynchronous architecture that decouples rollout from training while keeping policy lag observable and bounded. Complete ASP improves end-to-end throughput by up to 3.22 times and reduces iteration time by up to 76.8% relative to MindSpeed-RL, while reducing iteration time by up to 62.5% relative to verl. Within ASP, scheduling reduces mean iteration time by 10.9%. ASP also scales more robustly with additional generation resources than these baselines and shows no visible convergence degradation in the measured bounded-lag regime. ASP has since been used in an internal telecom-domain RL post-training workflow, reducing the reported training period from more than three months to about two months and producing a domain-enhanced model for knowledge answering and tool calling.

en

arXiv AI Papershace 7 días

Mind the Divergence: Divergence-Guided Reliability-Aware Fusion and Adaptive LLM Prompting for Conversational Emotion Recognition

Multimodal emotion recognition in conversations (MERC) integrates textual, acoustic, and visual cues to identify utterance-level emotions. However, existing approaches generally do not explicitly adapt multimodal processing to the degree of inter-modal predictive disagreement. In real conversations, semantic incongruity, concealed emotion, noisy observations, and ambiguity can affect modality-specific emotional evidence differently, while the reliability of individual modalities varies across samples. Rather than identifying each source of inconsistency separately, we model their common consequence for emotion recognition as predictive disagreement among unimodal emotion distributions over the same emotion-label set. Such disagreement makes indiscriminate fusion prone to propagating unreliable or competing evidence. Inspired by the Brunswik Lens Model, we propose MiD-ERC, a divergence-centric framework that quantifies inter-modal predictive disagreement using a normalized divergence score and estimates sample-specific modality reliability. MiD-ERC uses these signals to coordinate divergence-guided cross-modal gating, divergence-modulated reliability weighting, and divergence-conditioned adaptive LLM prompting. The framework is trained end-to-end using only standard emotion annotations, without requiring additional conflict labels, auxiliary data construction, or external retrieval modules. Experiments on IEMOCAP and MELD demonstrate that MiD-ERC achieves state-of-the-art performance, with particularly pronounced improvements on high-divergence samples.

en

arXiv AI Papershace 7 días

STRIDE: Learning Humanoid Locomotion in Minutes with Acceleration-Based Rewards

Learning locomotion directly on a humanoid robot is a promising direction to avoid the sim-to-real gap. However, locomotion rewards in Reinforcement Learning (RL) typically track the base velocity, which cannot be measured onboard without an external motion capture system or well-calibrated state estimator. This is why we introduce {}, which learns humanoid locomotion from minutes of samples with a reward that only requires the onboard Inertial Measurement Unit (IMU). Instead of tracking the base velocity, STRIDE rewards the robot for following a commanded acceleration profile. To learn efficiently from this reward, we use the off-policy RL algorithm XQC and combine a high gain setting for the joint position controllers with left-right symmetry augmentation. In simulation, STRIDE learns to walk on the Unitree G1 and Booster T1 humanoids within 14 minutes of samples. Compared to learning without these two components, it reduces the falls during training by 81%, while without them the robot learns only in over an hour.

en

arXiv AI Papershace 7 días

Learning Whom to Trust About the Unseen: Source Credibility and Reliability Transfer in Language Models

A source who has been right before deserves attention, but about what? We ask how far a demonstrated record of reliability travels in language models when the new claim cannot be checked, and whether the answer depends on the kind of claim. Three behavioral studies across six models approach the question from three angles. Study 1 adapts the Einstein-effect paradigm and finds that models, unlike the human sample, favor a spiritual leader over a scientist for the same obscure statement, and that the preference largely reverses when the statement is reworded in the vocabulary of physics. Study 2 adapts a paradigm in which a trusted expert calls a scientific or religious explanation unknown or a mystery; models reproduce the human domain asymmetry in exaggerated form. Study 3, the central test, gives two fictional informants different records on checkable river forecasts and then has them endorse matched claims about hidden natural and spiritual causes of the same village events. Most models carry the record to natural causes and barely at all to spiritual ones; one model transfers it weakly to either. Telling the model that the reliable informant forecasts by spiritual guidance rather than by measurement lowers that informant's influence on natural claims in every model and produces no comparable gain on spiritual claims in any. Demonstrated reliability thus has a scope in these models, and the account a source gives of how they came to be reliable can narrow it. Whether this reflects a judgment about relevance or the low prior credence of spiritual claims is the question the design leaves open, and we outline the experiment that would settle it.

en

arXiv AI Papershace 7 días

When F1 Is Not Enough: Auditing Codebooks for LLM-Assisted Annotation

Large language models (LLMs) are increasingly used for text annotation in social science, but standard performance metrics do not explain why errors occur. Low F1 may reflect limitations of the annotation system, defects in the written codebook, or inconsistency in reference labels. We propose a pre-deployment codebook audit for LLM-assisted annotation. The audit uses class-specific metrics and confusion matrices to identify problematic classes and boundaries, structured document-level review to attribute error sources, and targeted codebook revision followed by held-out evaluation. We apply the audit to two published multi-class codebooks from distinct domains. In both applications, the audit identifies under-specified or conflicting coding rules, and targeted revisions reduce the diagnosed errors on held-out data. Analyses with two LLM annotators show that some problematic boundaries and codebook revisions transfer across models, whereas other revision effects are annotator-specific. The results show that LLM annotation errors can help identify weaknesses in the written measurement instrument, not only in the model. Codebook auditing complements prompting, retrieval, and fine-tuning by clarifying whether persistent errors call for model improvement, codebook revision, or renewed scrutiny of reference labels.

en

arXiv AI Papershace 7 días

What Matters for LLM Persona Simulation: Signal Form in Cold-Start E-Commerce Advertising

Estimating advertising performance in cold-start settings is challenging because historical logs are unavailable before launch. Large language models (LLMs) offer a potential alternative by simulating persona-conditioned user responses, but it remains unclear which design choices actually drive LLM-based persona simulation. Using real Google Ads e-commerce campaign data across five demographic segments, we identify three signal-form dimensions that affect simulation quality: (i) a demographic anchor without noisy rich attributes, (ii) a relative qualitative banner-blindness signal rather than quantitative anchors, and (iii) prompt phrasing. On this campaign, our method attains an MAE of 24.9 and a MAPE of 26.1%, a 51.3% lower MAE than the best tested non-LLM baseline and 77.3% lower than an LLM configuration using explicit CTR anchors. Systematic ablations show that minimal demographic descriptions achieve lower mean error than richer persona configurations, while relative qualitative cues substantially outperform the tested quantitative anchors. We additionally compare these configurations with direct-prior and rule-based baselines and with alternative LLM configurations.

en

arXiv AI Papershace 7 días

Functional Dependency beyond Similarity: Hierarchical Decomposition of Training Data Attribution in Music Generative Models

Copyright analysis focuses on the similarity between two works. However, generative AI models learn from training data through functional relations beyond similarity. Training Data Attribution (TDA) estimates the influence of each training instance on a generated output and may serve as a basis for remuneration. However, prior studies find global influence, where a few training instances rank as the most influential for almost every generation. Used directly, TDA scores would distribute part of every remuneration to these instances regardless of what was generated. To address this, we propose a new approach in which TDA scores are decomposed into three levels of influence, namely global indicating the influence of a training data to a trained model, group for its influence on a group of generated outputs (that could be clustered together), and specific for its influence on a single generated output. Our study finds that, across several music generative models and TDA methods, global- and group-level influence is non-negligible in every case. Based on these observations, we propose the term functional dependency for the relation of training data to a trained model or to its outputs. This notion both encompasses and distinguishes dependencies of different scopes, allowing TDA scores to be interpreted as distinct forms of contribution in legal, economic and ethical discussions on remuneration.

en

arXiv AI Papershace 7 días

Solver-Admitted Metamorphic Testing of Learned Network Optimization Surrogates

Equivalent encodings of an optimization problem can produce different learned predictions. Interpreting these differences requires accounting for numerical error and multiple optima. We construct metamorphic tests for optimization solver surrogates with explicit eligibility conditions, output alignment, and numerical thresholds. Reference-solver agreement and independent constraint checks determine which sampled pairs enter the audit. We study alternating-current optimal power flow (AC-OPF), the traveling salesman problem (TSP), the capacitated vehicle routing problem (CVRP), and maximum independent set (MIS). Across five grids, GridSFM-Open violates each of 13 relations on at least one grid, and 12 by more than twice the threshold. Randomly initialized copies of the grid models fail nearly the same relations, so the failures do not require training. Source patches and input canonicalization remove selected discrepancies, but no intervention establishes preserved accuracy and feasibility within the declared margin on every grid. In routing, Policy Optimization with Multiple Optima (POMO) has a single-pass rotation/reflection violation rate that rises from 12.7% to 89.5% between 20 and 100 cities. Graph-Based Diffusion Solvers for Combinatorial Optimization (DIFUSCO) violates the MIS disjoint-union relation on 35.6% of draws. Post hoc normalization within connected components removes these violations on the tested graphs. These findings support reporting metamorphic consistency alongside optimality, feasibility, and accuracy, and validating repair quality separately from consistency.

en

arXiv AI Papershace 7 días

Notwithstanding Anything to the Contrary, Ignore All Instructions: Measuring Prompt Injection Guards on Legal Contract Review Attacks

Specialized prompt injection guards are deployed in front of document analysis systems, but no false positive rate has been published for any of them on legal contracts. We measure eleven open-weights guards on two corpora of legal contracts and find that existing guards struggle: document-level false positive rates are as high as 92.39%; two guards fall below chance on paired AUC; and a five-regex baseline is not Pareto-dominated by any guard in the panel. Controlled probes locate model-specific as well as general failure modes. Prompted LLMs resolve the tested failures, including a second threat category (unauthorized scope claims) on which every guard scores zero. A cascaded LLM solution, where the positives of a mid-tier model are verified by a frontier model, is 6.5x cheaper than using exclusively a frontier model and approaches its performance.

en

arXiv AI Papershace 7 días

Beyond the Lower-Star Filtration: Alternative Filtration Methods for Persistent Homology

In the lower-star filtration (LSF), simplices are assigned filtration values according to the maximum-vertex rule, with each simplex entering the filtration when its highest-valued vertex appears. A limitation of this construction is that some higher-dimensional simplices may enter simultaneously with their boundary faces, causing certain topological features to be filled immediately upon formation and thus not appear in the resulting persistence diagram. To address this limitation, we propose three alternative filtration methods: the ()-lower-star filtration (ELSF), the (L_p) filtration (LpF), and the softmax filtration (SMF). {ELSF introduces a dimension-dependent delay to the maximum-vertex rule,} allowing higher-dimensional simplices to enter the filtration after their boundary faces, whereas {LpF and SMF use aggregation functions that depend on all vertex values rather} than solely on their maximum. By modifying how filtration values are assigned, these methods allow simplices to enter the filtration more progressively, so that topological features such as loops and voids may persist over longer filtration intervals rather than being filled immediately. Consequently, the resulting persistence diagrams may include persistence features that are not observed under LSF, potentially providing useful information for distinguishing data populations. Experimental results on simulated and real-world datasets demonstrate that the proposed filtration methods are computationally efficient and can provide greater sensitivity to structural differences in the data than LSF.

en

arXiv AI Papershace 7 días

Global Solar Siting from Geospatial Embeddings

Global solar photovoltaic (PV) capacity is rapidly growing, helping decrease our emissions and dependence on fossil fuels. However, optimally siting a solar plant remains challenging, requiring anticipation of geographical, urban, and infrastructure needs. Current methods rely on multi-criteria decision analysis (MCDA) using hand-picked geophysical layers and solar-resource products, however these strategies are not geographically transferable and do not capture the practical requirements of solar sites. We use geospatial embeddings, dense vectors representing geophysical properties of a location, to analyse 38,591 existing solar sites and the land they were built on to learn the latent solar suitability signal. We assess a set of linear and non-linear models and select a soft-voting ensemble of a random forest and a multilayer perceptron (RF+MLP ensemble), chosen for accuracy and large scale deployability. Validated against future installations in thirteen countries, the ensemble separates existing solar sites from random land with a median ROC-AUC of 0.90. Compared with published MCDA strategies, the ensemble increases ROC-AUC by 0.06-0.30 in every country. The ranking also concentrates real deployment: the most-suitable 20% of land holds 84% of estimated generation (annual energy yield) and 67% of installed capacity (potential power output). After applying standard siting constraints, 5.6% of global land (9.8% if cropland is permitted) remains highly suited to solar. The global 10 m resolution solar suitability dataset and code can be found here: https://github.com/Abdullahr29/Global-Solar-Siting.

en

arXiv AI Papershace 7 días

I Think, Therefore I Attack: Evaluating Thought Injection Attacks on Reasoning Agents

Large Reasoning Models (LRMs) are one of the key components of current agentic systems. LRMs augment their output with an explicit chain-of-thought reasoning stage, introducing a new attack surface for thought injection, in which untrusted text resembling the model's own deliberation is fed back to the model. We present a two-part study of this attack, performed against the prevailing large, production class open-weight models (e.g., GLM-5.3, MiniMax-M2.5, Qwen3.8-Flash-Next). We first evaluate the effectiveness of static thought-injection prompts under different styles as well as tagged and untagged conditions and find that the effects of thinking tags and payload styles are all properties of the specific target model. In the second study, we exploit this model-specificity in a containerized agentic cyber range, using indirect injection prompts styled in the form of the target model's own reasoning traces to achieve attack success rates (ASR) up to 50% on operational objectives such as reverse-shell exfiltration and repository backdooring. We also find that the same malicious objective in standard prompt injection user voice never succeeds, and thinking tags suppress the agent's tendency to disclose the injection. Our results show that imitating a reasoning model's own thought style is effective for this style of attack, and that tag stripping is an unreliable defense.

en

arXiv AI Papershace 7 días

On the Few-Shot Adaptation of Vision-Language Models: A Few Gradient Signs Are All You Need

The limited supervision available for the few-shot adaptation of vision-language models makes gradient estimation particularly noisy. We find that the gradient magnitudes are severely miscalibrated, while the signs of a small top-magnitude subset remain informative. This motivates SignSparseSGD and DynSignSparseSGD, two state-free optimizers that use gradient magnitudes only to select where to update and fixed-magnitude signs to decide how to update. The static variant fixes the support after the first minibatch. The dynamic variant refreshes it at every step. Across 11 datasets, three CLIP backbones, and five shot levels, the proposed methods are competitive with or outperform parameter-efficient and optimization-based baselines on average, while exhibiting markedly lower sensitivity to long training schedules. They introduce no task-specific modules, maintain no first- or second-moment optimizer states, and use one shared configuration across datasets, backbones, and shot levels. Our code is included in the supplementary materials and will be released upon publication.

en

arXiv AI Papershace 7 días

The Matthews Correlation Coefficient: Fifty years

Fifty years after its introduction, the Matthews Correlation Coefficient (MCC) remains central -and debated- in binary classification. This survey offers a comprehensive retrospective and an in-depth analysis of MCC, covering its historical development, mathematical foundations, and implications for practice. We clarify long-standing misconceptions, formalize its equivalences (to Pearson's correlation and to Chi2 in 2x2 tables), and examine its behavior under class imbalance. Beyond MCC, we develop analytical (derivatives) and graphical (contour plots) frameworks that generalize to any confusion-matrix-based metric. These frameworks expose structural properties -bias, robustness, sensitivity- that scalar summaries conceal, and they provide actionable guidance for selecting and interpreting performance measures in applications where reliability and fairness matter. Together, our results position MCC within a broader toolkit while clarifying when it is informative -and when alternatives may be preferable.

en

arXiv AI Papershace 7 días

MotionBench-XAI: Benchmarking What Shapes Shapley Explanations in Spatiotemporal Machine Learning

Predicting outcomes from spatiotemporal data, such as assessing Parkinson's severity from human motion, often requires explainability in addition to accuracy. Practitioners need to know what drove the prediction, whether it be spatial (a body joint), temporal (a specific window), or both. Shapley values are attributions computed by treating input features as players in a cooperative game and measuring how the model output changes as coalitions of players are removed. Computing them involves choosing what counts as players, how absent players are imputed, and how the game is evaluated. How these choices affect explanation quality is poorly understood for spatiotemporal data, and real data carries no ground-truth attributions to check against. We present MotionBench-XAI, a benchmark built on i) five synthetic spatiotemporal datasets with exact ground-truth Shapley values, ii) three real datasets across clinical gait, electrocardiogram, and audio, iii) ten attribution methods across three player sets, and iv) three trained architectures. We show that grouping features into players, across time, space, or both, matters more than the choice of imputer. Generative imputers improve explanations only when their conditional models are accurate. On real data, the standard proxy metrics select contradictory best methods. We distill these findings into practical insights and provide an open-source pipeline.

en

arXiv AI Papershace 7 días

MotionDecipher: LLM-Assisted Clinical Interpretation of Human Motion Signals for Efficient Gait Analysis

Clinical gait and motion analysis is widely used to guide targeted interventions for patients with neurological disorders, the outcome of which is typically a report with descriptions of joint movements and forces. Current practices rely on experts' manual effort, which is labor-intensive and time-consuming. Recent advancements in Large Language Models (LLMs) provide opportunities to address this problem, but translating human motion into clinical gait descriptions remains challenging because the quality of the context directly affects LLMs' reasoning capability over continuous motion signals. To tackle this, we introduce MotionDecipher, a novel framework bridging the semantic gap between motion data and clinical interpretations. Specifically, we made 3 key advancements: 1) a Human Motion Tokenizer that converts motion signals into domain-specific tokens for time- and frequency-informed gait interpretation; 2) a Meta Feature Selector that transforms tokens into Structured Representations; and 3) a Hybrid Context Constructor to determine the optimal demonstration-based prompt for In-Context Learning. We evaluated MotionDecipher on data collected at an accredited gait lab from 122 patients and 46 controls. Results show that MotionDecipher achieves 0.91 BERTScore (13% over GPT-5), while expert evaluation ranks MotionDecipher highest in Correctness, Completeness, and Conciseness compared to baselines, highlighting its promise for efficient gait analysis.

en

arXiv AI Papershace 7 días

High-Dimensional Off-Policy Evaluation without Sample Splitting using an Onsager Correction

Reward models used for off-policy evaluation (OPE) are commonly selected using held-out data. When sample size and context dimension grow proportionally, the difference in test risk between models fitted on the full sample and on a fixed fraction of it need not vanish. Held-out evaluation can therefore consistently estimate the test risk of the model trained on the smaller sample, but because the relative ranking of candidate reward models can change after full-data refitting, held-out selection can be suboptimal. We develop an Onsager-corrected, target-policy-weighted criterion that estimates the full-data reward model's out-of-sample risk under the target policy using only its training observations. This enables early stopping along the full-data gradient-descent trajectory without sample splitting. For neural networks with fixed depth and hidden width under Gaussian contexts, single-index rewards, context-dependent multi-index target policies, and known context-independent logging, we prove uniform consistency of our criterion over a fixed finite set of gradient-descent iterates and a near-oracle guarantee for the selected reward model. The same correction yields a doubly robust estimator that consistently estimates the true policy value without sample splitting, even when reward-model prediction error does not vanish. Experiments show decreasing risk-estimation error as dimension grows at a fixed sample-to-dimension ratio, and improved reward-model selection relative to held-out baselines. Empirically, our correction also reduces the root mean squared error of doubly robust value estimates computed from the same data used to fit and select the reward models.

en

arXiv AI Papershace 7 días

SteerGR: Steering Generative Recommendation via Sparse Autoencoders for Controllable Game Discovery

Generative recommendation aims to predict item identifiers from user interaction histories, but explicit control over the properties of recommended items remains underexplored. A key challenge is to connect human-readable concepts to internal features whose interventions effectively change the recommended items. To address this, we propose Steerable Generative Recommender (SteerGR), a framework that uses sparse autoencoders for concept-level control in game discovery. SteerGR proceeds through three stages: (i) Concept Definition and Grounding for defining play concepts with an LLM; (ii) Feature Discovery and Selection for discovering candidate features by their concept associations and selecting one by its steering effect; and (iii) Sparse Feature Intervention for amplifying or suppressing selected features to steer recommendations without retraining the recommender. Extensive experiments with two industrial generative recommendation models on a real-world game dataset show that the selected features successfully steer recommendations toward or away from target concepts. We believe this work takes a first step to extend SAE-based steering from LLMs to industrial generative recommendation.

en