Study detects and explains positional bias in financial LLMs.
problem Positional bias in financial decision-making using LLMs.
method Unified framework and benchmark for detecting and quantifying bias in Qwen2.5 models.
result Positional bias is pervasive, scale-sensitive, and resurfaces under nuanced prompt designs.
New methods improve LLM preference optimization by intelligently weighting multiple reference models.
problem Improving LLM preference optimization with multiple reference models.
method Introducing four new weighting strategies for multiple-reference preference optimization.
result All four new weighting strategies outperform current methods on preference accuracy.
PyFi uses adversarial agents to train VLMs on financial image understanding.
problem Training VLMs to understand complex financial questions.
method PyFi-600K dataset and adversarial MCTS mechanism.
result Fine-tuned VLMs improve by 19.52% and 8.06% on financial question accuracy.
Language models fail to execute simple steps, showing gating and binding errors.
problem Procedural hallucinations in language models, failing to execute simple steps.
method Analyzed long-context binding tasks, identifying gating and binding errors.
result Procedural errors are due to gating and binding failures, with recency bias contributing to the latter.
CoE modularizes LLMs for scalable, cost-effective AI systems.
problem Challenges in scalability, cost, and customization of monolithic LLMs.
method Composition of Experts (CoE) leverages multiple expert LLMs with a router for dynamic selection.
result CoE achieves superior performance with reduced computational overhead and cost.
CoDistill-GRPO improves small models in GRPO by distilling knowledge from a larger model.
problem Small models in GRPO struggle with sparse rewards on difficult tasks.
method Simultaneously trains a large and small model using co-distillation and GRPO objectives.
result Significant improvement in small model performance over standard GRPO on mathematical benchmarks.
LLMs cause inconsistent financial outputs, smaller models are more reliable.
problem Inconsistent outputs from LLMs undermine auditability and trust in financial workflows.
method Finance-calibrated deterministic test harness, task-specific invariant checking, model classification, and cross-provider validation.
result Smaller models (Granite-3-8B, Qwen2.5-7B) achieve 100% output consistency, while larger models like GPT-OSS-120B have high drift.
Bias correction improves language model training performance.
problem Stochastic update bias in preconditioned optimizers.
method Cross-fitted preconditioning and variance-corrected inversion.
result Reduces held-out pretraining loss by 0.15 nats.
Adaptive speculative decoding framework for LLMs using bandit algorithms.
problem Adaptive speculative decoding for LLMs to balance speed and quality.
method Formulated as a Multi-Armed Bandit problem, proposed UCBSpec and EXP3Spec algorithms.
result UCBSpec algorithm achieves optimal regret performance up to universal constants.
Unified framework detects overfitting in crash classification models.
problem Evaluation metrics fail to detect overfitting in crash classification models.
method Random Matrix Theory and Heavy-Tailed Self-Regularization framework applied to various model types.
result Power-law exponent α reliably distinguishes well-regularized from overfit models.
IGSD separates task-specific content channels in transformer components by comparing activation replacement with zero ablation.
problem Mechanistic interpretability of transformer components
method IGSD: paired-intervention framework for comparing activation replacement with zero ablation
result IGSD identifies an early-layer content channel in transformer components that standard importance methods underestimate.
ABMLL adapts LLMs to multiple datasets efficiently.
problem Limited generalization across multiple datasets in LLMs.
method Amortized Bayesian Meta-Learning for LoRA (ABMLL).
result ABMLL supports effective generalization across datasets and scales to large models.
Transformer models show distinct spectral fingerprints under voice changes.
problem Detecting architectural biases in transformer models.
method Spectral analysis of attention-induced token graphs.
result Clear architectural signatures in model fingerprints correlate with language-specific behavior.
ORCA calibrates LLMs for efficient, generalizable reasoning.
problem Miscalibration of large language models leading to inefficiencies.
method Online Reasoning Calibration (ORCA) using conformal prediction and test-time training.
result ORCA provides higher efficiency and generalization across different reasoning tasks.
Training-free method improves large language model sequence quality via reward-guided sampling.
problem Optimizing large language model sequence quality over token likelihood.
method Reward-augmented target distribution combined with Sequential Monte Carlo sampling.
result Significant gains in sequence generation and mathematical reasoning tasks.
The paper analyzes RLVR's training dynamics, proving convergence depends on aligning update direction with Gradient Gap.
problem Understanding why RLVR works and its limitations.
method Analysis of RLVR's training process at trajectory and token levels, introducing Gradient Gap.
result Convergence depends on aligning update direction with Gradient Gap, with a sharp step-size threshold.
SAGE enhances reinforcement learning by injecting hints to prevent model stagnation.
problem Sparse rewards cause large language models to stall under relative policy optimization.
method SAGE injects privileged hints during training to increase within-group outcome diversity.
result SAGE consistently outperforms GRPO on 6 benchmarks with LLMs, achieving significant improvements.
LoRA-Curve connects independent LoRA optima through continuous low-loss valleys, improving Bayesian model averaging.
problem Challenges in estimating epistemic uncertainty in LoRA-based Bayesian inference.
method Introduces LoRA-Curve, a segmented Bézier curve parameterization in the LoRA space, with free and anchored configurations.
result Empirically shows that connecting independent LoRA optima through continuous low-loss valleys improves mutual information of the predictive distribution.
New method for RLHF reduces costs by integrating new data in one pass.
problem Continuous integration and re-optimization of models in RLHF leads to high computational and storage costs.
method Proposes a one-pass reward modeling method using online mirror descent with a tailored local norm.
result Achieves constant-time updates per iteration, enhancing both statistical and computational efficiency.
ALIEN improves uncertainty estimation of language models by refining entropy-based methods.
problem Overconfidence in uncertainty estimation for language models, especially for difficult inputs.
method ALIEN refines entropy-based uncertainty by aligning it with prediction reliability, using a lightweight uncertainty head.
result ALIEN consistently outperforms strong baselines in detecting incorrect predictions and achieving the lowest calibration error.
This paper uses CausalGANs and RL with LLM to predict bond yields.
problem Challenges in financial bond yield forecasting due to data scarcity and market conditions.
method Proposes a novel framework combining CausalGANs, RL, and LLM for synthetic data generation and trading signals.
result Improves forecasting performance over existing methods with low Mean Absolute Error.
Classifier-based AI safety gates fail in self-improvement, even with advanced verification methods.
problem Maintaining reliable oversight of AI systems as they improve over iterations.
method Comprehensive empirical testing on neural controllers and MuJoCo benchmarks, using various classifiers and verification methods.
result Classifier-based safety gates fail in maintaining reliable oversight, even with advanced verification methods.
Chronicals boosts LLM fine-tuning speed by 3.51x over Unsloth.
problem Memory bottleneck in fine-tuning large language models.
method Four optimizations: fused kernels, cross-entropy reduction, LoRA+, and sequence packing.
result 3.51x speedup on A100-40GB with Qwen2.5-0.5B.
This study improves stock price prediction using multimodal data.
problem Improving financial asset price forecasting accuracy.
method Combining candlestick time series and textual news flow data using LSTM and pre-trained models.
result Textual modality reduces MAPE by 55%.
SC unifies ICL calibration methods and improves LLM performance.
problem Systematic biases in LLM predictions leading to unstable performance.
method Supervised Calibration (SC) learns optimal affine transformations in logit space.
result SC delivers state-of-the-art performance across multiple datasets.
StatLoRA uses statistical inference to allocate ranks in LoRA fine-tuning, improving performance.
problem Balancing efficiency, expressiveness, and generalization in LoRA rank allocation.
method Formulates LoRA rank allocation as a statistical hypothesis testing problem, using estimated p-values to determine component retention or pruning.
result StatLoRA achieves comparable or better performance than existing methods under matched rank budgets.
G1 uses RL to enhance LLMs' graph reasoning, improving performance on diverse tasks.
problem Limited graph reasoning abilities of LLMs, especially in synthetic graph-theoretic tasks.
method Curated synthetic graph dataset, RL training on LLMs.
result Significant improvements in graph reasoning, zero-shot generalization to unseen tasks.
A geometric theory explains loss functions for robust representation learning.
problem Treats robustness, domain adaptation, and sensor drift as separate literatures.
method Estimates covariance Sigma_task and uses it to pin Jacobian penalties.
result Proves optimality and necessity of range coverage for penalty matrices.
Transformers approximate Bayesian posteriors but not exactly.
problem Bayesian accounts of in-context learning face challenges due to task-preserving order changes in transformers.
method Showed that excess prequential code length is exactly cumulative predictive KL, decomposing expected regret into order-averaged predictor and order-averaging gain.
result Transformers approximate Bayesian posteriors but not exactly, priced by log loss.
Weibull framework diagnoses transformer weight distributions, revealing distinct patterns across modules.
problem Diagnosing weight distributions in transformer models for better understanding of training dynamics.
method Applied Weibull distribution to diagnose weight magnitude distributions in transformer models, fitting each weight matrix independently.
result Distinct patterns of weight distributions across different transformer modules were identified.