Science – Page 7

Sharper Reasoning: Training Language Models to Think Step-by-Step

20.12.2025 by qfx

The Generative Adversarial Reasoner (GAR) consistently elevates performance on seven mathematical reasoning benchmarks-achieving gains of up to +35.3% on the LiveMathBench-Hard dataset-and demonstrates robust generalization across both Deepseek-R1-Distill-Qwen-7B and Deepseek-R1-Distill-Llama-8B backbones, as evidenced by improvements of +22.9% on AIME24 and +19.5% on AIME25 with the Llama model.

A new co-training framework uses an adversarial approach to refine the reasoning process of large language models, leading to more accurate and efficient problem-solving.

Beyond Models: A New Approach to Pricing Exotic Derivatives

20.12.2025 by qfx

The methodology integrates a Smart Monte Carlo reweighting step into the standard front-office pricing loop of investment banking, enforcing consistency with vanilla options and establishing robust min-max bounds without necessitating alterations to the existing risk library.

A novel framework reconciles the benefits of model-independent pricing with the practical demands of implementation for complex financial instruments.

Beyond Search Results: How Reranking is Reshaping Information Access

20.12.2025 by qfx

The research demonstrates a Retrieval-Augmented Generation (RAG) approach enhanced by a post-retrieval reranking step, designed to refine document selection and improve the precision of generated responses.

This in-depth review charts the progression of reranking techniques, from early algorithmic approaches to the transformative impact of deep learning and large language models.

Connecting the Dots: Graph Networks for Smarter Research

19.12.2025 by qfx

The GRIL framework establishes a structured approach to reasoning, leveraging a cyclical process of generating, retrieving, and iteratively refining information to achieve robust performance in complex tasks.

This review examines how leveraging graph-based approaches can improve the discovery of relevant research papers and enhance academic assistance tools.

Echoes of the Past: How News AI Can Amplify Racial Bias

19.12.2025 by qfx

The evolving usage of the term “blacks” alongside its label “black” demonstrates a discernible shift in terminology over time.

A new study reveals that artificial intelligence systems used in news gathering and dissemination can inadvertently perpetuate outdated racial biases embedded within the historical data they are trained on.

Predicting the Fed: An AI System for Smarter Rate Forecasts

19.12.2025 by qfx

FedSight leverages a multi-agent artificial intelligence system to forecast the decisions of the Federal Open Market Committee, effectively modeling the complex interplay of factors influencing monetary policy.

Researchers have developed an AI-powered multi-agent system capable of forecasting Federal Funds target rates by analyzing a wide range of economic data.

The Quiet Signal of Market Moves

19.12.2025 by qfx

Order-flow entropy demonstrably predicts the magnitude of five-minute returns-declining from 8.14 to 3.75 basis points across quintiles-thereby forecasting volatility, though it provides no indication of return direction, as evidenced by an analysis of 828,907 second-resolution observations over 36 trading days, significantly differing from the unconditional mean of 5.29 basis points.

New research reveals how the complexity of trading activity can foreshadow the size of price swings, even without indicating which way they’ll go.

Pricing Options with a Speed Boost: Machine Learning Meets Fourier Transforms

19.12.2025 by qfx

$The relationship between the point at which a theoretical model is truncated and the resulting option prices demonstrates a fundamental limit to predictive power - as the model simplifies, its capacity to accurately reflect market value diminishes, mirroring the inevitable loss of information beyond the event horizon of any constructed theory, much like the formula $P = f(K, T, \sigma)$ becomes increasingly imprecise with omitted variables.$

A new framework leverages the power of machine learning and fast Fourier transforms to deliver significantly faster and more accurate option pricing compared to traditional methods.

Decoding the Market with Deep Learning

19.12.2025 by qfx

Return predictability significantly improves with the introduction of consensus learning, peaking at a $λ$ value between 0.3 and 0.4, and consistently outperforms benchmarks even at higher $λ$ values, while the accuracy of consensus variable approximation increases steadily with increasing $λ$.

A new framework combines the power of artificial intelligence with analyst insights to predict stock returns and offer economic transparency.

Beating the Market with Quantum-Boosted Prediction

19.12.2025 by qfx

A new hybrid approach combining classical machine learning with quantum-enhanced features significantly improves the accuracy of S&P 500 directional prediction.