Sección

Investigación

Papers y preprints de los feeds de investigación que seguimos.

6415documentos

SpecMRL: Redesign Matryoshka Representation Learning within Parametric Spectrum

Matryoshka Representation Learning (MRL) is designed to support flexible representation sizes by jointly optimizing nested prefixes. While effective, this design lacks an explicit constraint to ensure that the leading portion provides a suitable compact encoding. To address this, we propose SpecMRL, which reorganizes the feature space to enable more effective prefix readout. Specifically, we extend MRL with a parameter-derived spectral basis organized with feature importance. SpecMRL otherwise follows standard MRL, with prefix readout performed in the learned spectral space. Experiments on MTEB show that SpecMRL outperforms MRL at all evaluated reduced dimensions, with larger gains under more aggressive compression. These results suggest that the learned spectral basis better prioritizes informative directions, enabling more effective representation compression. Notably, SpecMRL extends the performance frontier of MRL, surpassing full-dimensional MRL with only half the embedding dimensions.

arXiv AI Papershace 25 días

AfriBED: a Speech Recognition Bias Evaluation Dataset for Low Resource African Languages

Despite advancements in automatic speech recognition systems (ASR), current systems exhibit speaker-attribute bias, i.e., their speech recognition performance is biased against certain speakers based on their demographic attributes such as gender, accents, and speaker type. However, accurately assessing speaker attribute bias in a given ASR system requires the availability of a bias evaluation dataset with comparable speech across subgroups of a speaker attribute. In this paper, we introduce African speech Bias Evaluation Dataset (AfriBED), a novel speech recognition bias evaluation dataset for two low resource African languages, Bemba and Nyanja, to detect and estimate gender and speaker-type bias for ASR systems. The dataset comprises 9 hours of speech with 3928 utterances recorded by 80 speakers for Bemba, while the Nyanja set includes 6260 utterances with approximately 18 hours of speech recorded by 127 speakers. The results of our baseline experiments demonstrate that the datasets are useful for examining gender and speaker-type attribute bias in the target languages.

arXiv AI Papershace 25 días

ASTOR: Benchmarking and Improving Autoformulation for Soft-Constraint Optimization with a Formal Grammar

Real-world decision problems combine two qualitatively different kinds of constraints: requirements that must be satisfied, called hard constraints, and preferences that can be violated, called soft constraints. Because these problems are often posed to decision-makers in natural language, large language models offer a natural tool for assisting with their solution. This has given rise to autoformulation, which focuses on using LLMs to translate natural-language descriptions of decision problems into executable mathematical optimization models. However, this research often neglects soft constraints like those proposed by consumers, managers, and planners, despite the fact that many practical problems require modeling them explicitly. We introduce ASTOR, an inference-time framework that maps a natural-language decision problem to a grammar-guided abstract syntax tree (AST) before deterministically compiling the resulting representation into executable optimization code. We also introduce SoftOR, a unified benchmark of natural language problems which focuses on soft constraint formulations. On SoftOR, ASTOR achieves the best aggregate accuracy among comparable single-request and multi-agent LLM methods while remaining competitive on standard autoformulation benchmarks that include only hard constraints. Code and data are available at https://anonymous.4open.science/r/astor-softor-iclr-E34C/README.md.

arXiv AI Papershace 25 días

Next Best Action: Causal Constrained Offline RL for Business Action Targeting

Next Best Action (NBA) systems must select among customer interventions (e.g., discounts, promotions, and lifecycle actions), maximize long-term economic value, satisfy complex operational constraints, and remain causally grounded. Existing approaches lack a rigorous framework for meeting these requirements jointly. We address this gap by formulating NBA as a Constrained Markov Decision Process (CMDP). We then integrate off-policy causal correction of the estimated customer dynamics directly into the Linear Programming (LP) formulation of the CMDP. The resulting policy selects actions based on customer states to maximize long-term value while explicitly enforcing business guardrails. Furthermore, the LP formulation yields shadow prices that quantify the economic trade-offs of operational constraints (e.g., fairness and user welfare), enhancing policy interpretability. We validate our proposed offline Reinforcement Learning (RL) framework on an industry case study involving rideshare coupon targeting, where customer states are defined by Recency, Frequency, and Monetary (RFM) stages. We evaluate performance using model-based off-policy evaluation, with causal dynamics estimated from a temporal holdout. Under the observed coupon budget and RFM state representation, our optimized policy improves customer lifetime value (LTV) by 19% over the estimated behavior policy and 8% over a myopic baseline. Although the demonstrated action space is binary, the proposed framework is action-agnostic, yielding a general-purpose decision module for targeting systems that manage multiple interventions.

arXiv AI Papershace 25 días

Safety Mid-Training: Internalizing LLM Safety as a Foundational Capability

Safety alignment for large language models (LLMs) is primarily performed during post-training, while recent work has also begun incorporating safety data into pre-training. Yet safety can degrade when later training no longer includes safety data, and gains in safety often come at the cost of general capability. This tension suggests that safety and general reasoning should not be treated as separate training goals. We instead view safety as an LLM capability that can be learned through a reasoning structure shared with general tasks. Both require the model to understand intent or task goals, apply relevant knowledge, and reason about potential consequences. We introduce safety mid-training, a stage between pre-training and post-training that jointly develops safety and general reasoning through this shared structure. Given safety and utility prompts, the base model generates structured documents and reasoning traces that are then used for mid-training. Safety mid-training reduces average attack success rate (ASR) by 72% to 81% relative to Qwen3-1.7B, 4B, and 8B base models, while also improving general capability. The benefits persist through subsequent post-training. After general-only SFT, mid-trained models largely retain their safety, achieving lower ASR than the baselines without sacrificing general capability. They also retain safety under general-only GRPO without safety rewards, while subsequent RL reaches a better safety-utility frontier when initialized from the mid-trained models. Together, these results suggest that learning safety and general reasoning within a shared mid-training stage can reduce the safety-utility trade-off and make safety more robust to subsequent training.

arXiv AI Papershace 25 días

Scaling Parameter-Efficient Fine-Tuning for Brain Encoding

Brain encoding models aim to predict neural responses to stimuli. The most widely used approach fits readouts from frozen pretrained backbones. Using neural data to train large models from scratch or fully fine-tune them after task-pretraining has proven difficult because neural data is costly and limited. We show that parameter-efficient fine-tuning (PEFT) enables us to effectively adapt very large pretrained backbones for brain encoding with a small number of trainable parameters. To make it computationally practical, we exploit the backbone's causal structure to predict multiple timepoints in a single pass instead of using sliding windows, significantly reducing feature-extraction cost. This enables efficient fine-tuning of multimodal backbones with up to 104B parameters, the largest model adapted for brain encoding to our knowledge. Trained with a low-rank adapter (LoRA) and a lightweight readout, our approach consistently improves brain encoding performance across nine diverse multimodal backbones. On the same Qwen3-Omni backbone as the prior state-of-the-art (SoTA), our method BrainLoRA achieves the best published performance on the Algonauts 2025 Challenge. We also show the adapted multimodal backbones remain interpretable: modality attribution recovers the canonical sensory topography that prior work often obtains by construction from separate unimodal encoders, and neural adaptation increases the contribution of audio inputs and cross-modal mixing within the backbone. To scale out the number of adapted models, we develop a multi-LoRA kernel that accelerates joint training of multiple LoRAs on a shared backbone. This enables an ensemble of 200+ adapted models that ranks first on the public Algonauts post-challenge leaderboard at the time of writing. Together, our results highlight PEFT as an efficient and powerful paradigm for brain encoding.

arXiv AI Papershace 25 días

Mise-en-Video: Fine-Grained 3D Control for Multi-Shot Video Generation

Recent advances in text-to-video generation have enabled high-fidelity single-shot synthesis, yet practical storytelling creation requires multi-shot videos with both strong inter-shot consistency and fine-grained controllability. Existing multi-shot video generation approaches achieve strong consistency but still fall short in controllability. In this paper, we propose Mise-en-Video (MeV), a highly controllable multi-shot video generation framework that allows video creators to conduct staging the multi-shot in the 3D space and then generate the corresponding multi-shot video. To achieve this, our framework combines a hierarchical attention mechanism for multi-level narrative consistency with 3D-aware layout guidance for precise, per-shot spatial control. To support training and evaluation, we develop standardized annotation pipelines and construct a 160K-video dataset with hierarchical multi-shot texts and oriented 3D bounding boxes. Extensive experiments demonstrate that MeV consistently outperforms strong baselines in video quality, inter-shot consistency, and subject-level controllability.

arXiv AI Papershace 25 días

Learning to Refer: Inference-Time Collaboration in Decentralized Federated Learning

In decentralized federated learning, heterogeneous clients may develop complementary expertise, yet each query is typically evaluated only by the client where it arrives. To exploit this distributed expertise without global coordination, we formulate inference-time collaboration as budgeted local expert referral, a graph-constrained routing process in which a query can move only between neighboring clients and only visited clients contribute to its final prediction. EdgeDRM implements this framework by learning a density-based routing potential for each client: differences between neighboring potentials provide the comparisons used to select successive referrals, while local task models generate predictions that are aggregated along the resulting route. Both training and inference are decentralized, using only neighbor-local information without a central router or graph-wide client scoring. We establish estimation guarantees for the learned comparisons and characterize how local improvement and graph structure govern prior-weighted density gains from referral. Experiments on benchmark and real-world medical data show that EdgeDRM outperforms graph-matched decentralized baselines in most settings, with the clearest benefits under sparse connectivity.

arXiv AI Papershace 25 días

Teacher Supervision, Fixed Representations and Calibration: Exact Risk Calculations for Candidate-Scoring Decision Models

Candidate-scoring decision models return a probability distribution over a presented set of answers. Public documentation of one such system, Jev, and of an open counterpart, JevAny, describes three training ingredients that determine what these probabilities estimate: soft targets from larger models, a frozen pretrained backbone and temperature calibration. We analyze each ingredient in Gaussian surrogates where the finite-sample risk can be computed exactly. A student trained on teacher targets inherits the teacher's bias, which more student data does not remove and agreement with the teacher does not reveal. A fixed representation lowers the risk only below an explicit sample-size threshold. In a probit model the exact recalibration map is a temperature scaling, which removes all excess risk only when the teacher's coefficient has the correct direction and otherwise leaves a residual that calibration diagnostics do not detect. Under Luce's choice axiom, candidate scores are coherent across refinements of an event if and only if they aggregate by log-sum-exp. On 231,521 credit card complaints from the Consumer Financial Protection Bureau, where the decision is whether a complaint ends in monetary relief, a student trained on a logistic teacher's probabilities ends at the teacher's decision cost while a student trained on outcomes goes 8% below it, a 30B model read zero-shot remains worse than the teacher after recalibration, and that model departs from the coherence condition by 17.5 decibans. Pretrained representations reduce the labeled data required and calibration corrects the scale of the probabilities, but neither corrects a biased teacher, so the probabilities have to be checked against realized outcomes.

arXiv AI Papershace 25 días

Two hidden layer suffice: the caterpillar net

Every continuous piecewise linear (CPWL) function is a finite signed sum of maxima of affine functions, making maximum representation a central problem in the study of the expressiveness of ReLU networks. A now-disproved conjecture predicted logarithmic depth for maxima. Inspired by the proof by Brunck et al. (2026) that two hidden layers suffice for general CPWL functions, we give the Caterpillar Net, a nonrecursive construction for the maximum of N inputs with two hidden layers of widths O(N^2) and 2^{O(NN)}. The construction has a direct geometric interpretation.

arXiv AI Papershace 25 días

On Data Scaling in Tabular ML: More Data, Better Generalization?

Recent work in tabular machine learning has emphasized the importance of diverse and realistic evaluation settings. The niche of large-scale (1M+) data, however, remains under-explored. To study training-data scaling, we assemble eight real-world tabular datasets with up to around ten million training examples, retaining the natural distribution shifts of the tasks and using random-split evaluation as a control. We find that GBDTs and MLP-based models show broadly similar scaling trends, with gains from both additional training data and algorithmic improvements. Tabular foundation models show more varied scaling behavior, with recent versions benefiting more consistently from larger contexts even under distribution shift. However, scaling training data under distribution shift introduces new challenges: in-distribution and out-of-distribution performance mostly improve together at smaller scales but can diverge at larger scales. We examine this through MLP capacity sweeps and comparisons of the validation distributions used for tuning and early stopping in MLPs and GBDTs. We also compare common practices of using validation data, contrasting its use for model selection in MLP pipelines with using it for in-context fitting in tabular foundation model pipelines. Under stronger shifts, MLPs selected using validation representative of the evaluation conditions generally maintain or improve performance as training data grows. In contrast, including such examples in context improves foundation-model performance, but does not prevent deterioration with increasing training size in our experiments.

arXiv AI Papershace 25 días

SOMA: Continuous Monitoring and Intervention for System Two Reasoning with System One Models

Reasoning models often run into problems without noticing them, even when symptoms are visible in their own text. In humans, metacognition separates monitoring from control: fast, intuitive signals keep watch while slow, deliberate reasoning unfolds and help decide whether to continue, change course, or stop. System One (S1) decision models, such as Jev, are a natural candidate for this role: they return a calibrated probability for a natural-language question about a text in one fast call, yet they are used only as gates before reasoning. We introduce SOMA (System One Metacognitive Awareness), in which an S1 model continuously monitors a System Two (S2) reasoning model and intervenes when needed. SOMA has three layers: a scenario layer states each objective in natural language with predefined actions, an awareness layer lets S1 judge whether the described behavior appears, and a control layer accumulates these judgments, first reminds S2, and takes over only if the behavior persists. Sharing one S1 model and one controller across tasks, without training, SOMA addresses two different failures on two reasoning models: on overthinking, it cuts reasoning tokens by up to 36\% while improving average accuracy, and on unanswerable questions, it raises abstention from 70.5\% to 87.2\%.

arXiv AI Papershace 25 días

Training with ML-labeled Data via Stochastic First- and Zeroth-Order Optimization

Training a machine learning (ML) model is an optimization process that typically requires a set of examples with ground-truth labels beforehand. Obtaining these ground truths usually requires laborious human effort and can be costly. The scalability of human labeling becomes a concern when there are a large number of data points that need to be labeled. To address this practical issue, we consider an optimization setting where a subset of samples has ground-truth labels, while the remaining data points were labeled by an ML predictor, which could potentially assign incorrect labels to those points. We propose both first-order and zeroth-order stochastic optimization methods for training on datasets consisting of a mix of true-labeled and ML-labeled samples. We further conduct comprehensive empirical studies from training a linear regression model to fine-tuning a large language model (LLM) to showcase the effectiveness of the proposed algorithms for training with ML-labeled data.

arXiv AI Papershace 25 días

Cooperative Multi-Agent Reinforcement Learning for Market Making on Prediction-Market Order Books

A market maker provides liquidity by quoting prices to buy and sell an asset, earning the spread while managing inventory risk. We study market makers trained with multi-agent reinforcement learning on prediction markets, specifically Polymarket, whose thin order books pose distinct challenges. Three challenges underpin this research: adverse selection may cause a policy trained on real data to stop quoting; the consequences of a quote can arrive many steps after the action; and prediction-market L2 data does not identify individual orders needed for simulation. We address these challenges by reconstructing order-level LOBSTER-format data from Polymarket and training agents in JaxMARL-HFT on 417 real markets. We enrich the baseline market maker's observation space with the book and recent flow, redesign its actions and inventory controls, and propose a cooperative multi-agent decomposition into a quoter and a risk manager trained using recurrent Independent Proximal Policy Optimisation (IPPO) with RNN policies. We investigate reward shaping, particularly reward blending and post-fill markout, to mitigate credit-assignment difficulties between the agents. Our redesigned single-agent market maker quotes more frequently and incurs smaller held-out losses than the retrained JaxMARL-HFT baseline, although the contributions of jointly changed observations, actions and risk constraints are not separately identified. Decomposition alone performs poorly; careful reward design substantially improves the desk relative to the unshaped split and yields our best result. Models are evaluated on realised end-of-episode portfolio value on held-out markets. None establishes positive average portfolio value in periodic held-out evaluation, and each configuration was trained with one seed. The environment, data and protocol lay a foundation for makers that quote many linked markets at once.

arXiv AI Papershace 25 días

LG-Bench: A Graph-Structured Evaluation Benchmark for Life Sciences

Traditional evaluation benchmarks reduce inherently interconnected scientific knowledge in life sciences into flat lists of questions, disregarding the underlying topological structure of the knowledge. We introduce, the first graph-structured benchmark for life sciences, featuring over 10,000 high-quality multiple-choice questions across medicine, biology, and chemistry. Our approach constructs a weighted evaluation graph using bidirectional matching and semantic similarity algorithms, where nodes represent questions and edge weights capture their semantic relationships. Leveraging this graph topology, we design two novel evaluation metrics. The Global Coherence Score (GCS) measures a model's consistency within semantically related neighborhoods, while Knowledge Balance Score (KBS) analyzes how model errors are distributed across the graph to reveal conceptual blind spots. LG-Bench facilitates fine-grained comparison of LLMs by surfacing differences in conceptual coherence and patterns of knowledge organization across models. Our framework shifts the evaluation paradigm from flat accuracy metrics to structure-aware analysis, offering a new lens for diagnosing and improving LLM performance in the life sciences domain.

arXiv AI Papershace 25 días

RESIN: Multi-Chain Reasoning with Entity-Structured Indexing

Multi-chain reasoning is a challenging task not only because the multi-chain evidence is dispersed, but also because different parts of reasoning may require distinct, interacting chains of support. Flat retrieval can flood a model with loosely related text, whereas early selection of a single reasoning path can discard evidence that becomes useful only after other facts are considered. We introduce RESIN (Reasoning with Entity Structured Indexing), which uses structure to locate evidence while keeping the original source as the basis for reasoning. RESIN decomposes a question into focused one-hop needs, retrieves a small set of source-grounded units for each hop, and places only this relevant evidence in a compact reasoning buffer. The buffer preserves alternative supports until final composition, making intermediate decisions explicit without forcing a brittle path. On the full HolisQA benchmark, RESIN achieves 82.92 context recall and 76.70 answer relevancy over all questions, with higher reported CR and AR across five domains and higher F1 in four domains than the published baselines. Ablations show that both typed evidence diversity and hop-wise decomposition are essential, supporting a clear conclusion: Precise evidence organization and delayed commitment make multi-chain reasoning effective and auditable. Our code is available at https://github.com/RamonMeng/RESIN.

arXiv AI Papershace 25 días

Game of Phones: Harnessing Game Theory and LLMs in 'Bot Wars' to Counteract Phone Scams

Phone scams impose substantial and growing costs on individuals and financial systems worldwide, yet existing countermeasures impose no resource cost on perpetrators. Strategic scam-baiting, in which automated agents engage scammers to deplete their operational capacity, offers a proactive alternative but requires dynamic strategy adaptation that static systems cannot provide. We introduce a game-theoretic framework enabling LLMs to learn effective counter-strategies through in-context experience without fine-tuning. The framework models scammer-baiter interactions as a two-player game with empirically grounded strategies and utility functions, using Nash equilibrium as a formal benchmark for evaluating learned behaviour. Across 300 simulated dialogues and three LLM architectures, adaptive strategy selection achieves 28.9-79.7% closer alignment to Nash equilibrium than non-adaptive baselines. The code and an annotated dataset of 300 dialogues with turn-level strategy labels and utility scores are released at https://github.com/nardinebasta/GameofPhones_/ .

arXiv AI Papershace 25 días

Topological Sheaves of Irreducibility: A Theoretical Framework for Multi-Agent Task Obstructions

Multi-agent systems can differ under identical agents and communication budgets when they support different forms of joint interaction. We introduce Topological Sheaves of Irreducibility (TSI), a sheaf-theoretic framework separating task-level compatibility obstructions from system capability. A task is represented by an interaction complex K and a capacity sheaf {F}; nonzero H^n(K;{F}) defines a structural signature, while registered interventions test capability. For cocycle data, cochain realization is solvable exactly when the represented cohomology class vanishes. Examples show that the signature is sheaf-relative, and the degree-1 reduction yields cycle-space witnesses and observation requirements. We evaluate two finite one-round models. Across 15,625 degree-1 inputs, an all-agent additive primitive solves decision and repair exactly; the optimal pairwise system reaches 88.80% on decision, one input above the 88.79% majority-class baseline, and 26.63% on repair. On a degree-2 tetrahedral-boundary task over 531,441 inputs, the same-2-skeleton system retains all four three-agent primitives yet matches the pairwise-only optimum throughout the registered separation window, whereas adding the class-filling four-agent primitive makes both tasks exact. Status-preserving robustness checks retain exact performance but are not cost-matched to the class-restoring ablation. Conditional on the registered primitives and cost models, TSI provides task-relative, falsifiable tests of higher-order interaction.

arXiv AI Papershace 25 días

Monsoon: Scalable Spontaneous Speech Collection for Low-Resource ASR in 50 Languages

For most languages, automatic speech recognition (ASR) is limited less by model architecture than by a lack of accurately transcribed spontaneous speech. We present Monsoon, a framework for collecting such data at scale. It elicits unscripted conversations between paired contributors, screens recordings with automated classifiers, corrects transcripts over five levels of human review, and selects segments for lexical and acoustic coverage. It has collected speech in 50 languages; we analyse seven 3,500-hour corpora. At equal volume, its quality stages reduce semantic word error rate (WER) by 10-39% relative to a conventional protocol. Compared with an architecture-matched baseline trained on public corpora, Whisper fine-tuned on Monsoon performs comparably on the baseline's own test sets and reduces semantic WER by 20-44 points on conversational benchmarks. The gains transfer to a second architecture and grow with data volume. We release the framework, its quality-control models, and the fine-tuned models.

arXiv AI Papershace 25 días

HarnessPro: Recursive Harness Self-Improvement with Pareto Optimization for Performance-Cost Trade-offs

Harness development plays a crucial role in building capable AI agents. Recent approaches improve harnesses through recursive self-improvement (RSI), but their performance gains often come with higher execution costs. We introduce HarnessPro, a label-free method for evolving harnesses to improve task performance while reducing execution cost. In this setting, candidate harnesses are evaluated using LLM-generated tests without access to ground-truth solutions or official tests. Specialized agents diagnose execution trajectories, plan updates with different performance-cost priorities, and implement them as candidate harnesses. Rather than returning a single harness, HarnessPro uses normalized weighted squared-distance selection to provide cost-focused (lite), balanced (medium), and performance-focused (heavy) harnesses for different deployment budgets. Selected harnesses enter the next iteration, where execution feedback guides further updates. Our experiments cover Terminal-Bench 2, SWE-Bench Pro, and ProgramBench, using GPT-5.5 with Codex and GLM-5.2 with Claude Code. Across these benchmarks using GPT-5.5 with Codex, the heavy mode of HarnessPro achieves an average pass-rate gain of 2.0 percentage points and an average per-task cost reduction of 26.1% relative to RHO. This advantage also holds when using GLM-5.2 with Claude Code. Code is available at https://github.com/pzs19/HarnessPro