Enhancing Mathematical Reasoning and Problem-Solving Abilities in Large Language Models through a Sequential Learning Approach
This study presents a novel sequential learning approach, named SAAS (Solving Ability Amplification Strategy), which strategically transitions from Chain-of-Thought (CoT) learning to Program-of-Thought (PoT) learning to enhance both mathematical reasoning and problem-solving abilities in Large Language Models.
Latent Representation of Mathematical Operations for Expression Derivation
This paper investigates the possibility of approximating multiple mathematical operations in latent space for expression derivation. It introduces different multi-operational representation paradigms, modeling mathematical operations as explicit geometric transformations, and analyzes the properties of each paradigm when instantiated with state-of-the-art neural encoders.
Outcome-supervised Value Models for Efficient Multi-Step Mathematical Reasoning
Outcome supervision can be leveraged to train a value model that prioritizes steps leading to accurate final answers, enabling efficient guided decoding for multi-step mathematical reasoning.
Enhancing Mathematical Reasoning in Large Language Models through Efficient Data Selection and Composition
Selecting influential data for fine-tuning on mathematical reasoning tasks is crucial for both performance and computation efficiency. The authors propose a Quality-aware Diverse Selection (QaDS) strategy to select influential data, and explore an optimal influential data composition for mathematical reasoning tasks.
Evaluating LLMs' Mathematical Reasoning in Financial Document Question Answering
The author explores the mathematical reasoning capabilities of Large Language Models (LLMs) on financial tabular datasets, focusing on sensitivity to table complexity and performance variations with arithmetic reasoning steps.
