paper-with-me

홈 › Papers

Composing Global Optimizers to Reasoning Tasks via Algebraic Objects in Neural Nets

2024-10-02 · Yuandong Tian

We prove rich algebraic structures of the solution space for 2-layer neural networks with quadratic activation and $L_2$ loss, trained on reasoning tasks in Abelian group (e.g., modular addition). Such a rich structure enables analytical construction of global optimal solutions from partial solutions that only satisfy part of the loss, despite its high nonlinearity. We coin the framework as CoGO (Composing Global Optimizers). Specifically, we show that the weight space over different numbers of hidden nodes of the 2-layer network is equipped with a semi-ring algebraic structure, and the loss function to be optimized consists of monomial potentials, which are ring homomorphism, allowing partial solutions to be composed into global ones by ring addition and multiplication. Our experiments show that around $95\%$ of the solutions obtained by gradient descent match exactly our theoretical constructions. Although the global optimizers constructed only required a small number of hidden nodes, our analysis on gradient dynamics shows that over-parameterization asymptotically decouples training dynamics and is beneficial. We further show that training dynamics favors simpler solutions under weight decay, and thus high-order global optimizers such as perfect memorization are unfavorable. Code can be found at https://github.com/facebookresearch/luckmatters/tree/yuandong3/ssl/real-dataset.

📄 PDF Abstract BibTeX arXiv:2410.01779

Code (1)

facebookresearch/luckmatters 공식 구현 pytorch

Tasks

Memorization

Similar Papers 제목 키워드 기반

An Algebraically Converging Stochastic Gradient Descent Algorithm for Global Optimization

2022-04-12 · Björn Engquist, Kui Ren, Yunan Yang

We propose a new gradient descent algorithm with added stochastic terms for finding the global optimizers of nonconvex optimization problems. A key component in the algorithm is the adaptive tuning of the randomness base…

global-optimization

Convergence of optimizers implies eigenvalues filtering at equilibrium

2025-10-10 · Jerome Bolte, Quoc-Tung Le, Edouard Pauwels arxiv

Ample empirical evidence in deep neural network training suggests that a variety of optimizers tend to find nearly global optima. In this article, we adopt the reversed perspective that convergence to an arbitrary point …

Syzygy of Thoughts: Improving LLM CoT with the Minimal Free Resolution

2025-04-13 · Chenghao Li, Chaoning Zhang, Yi Lu, Jiaquan Zhang 외

Chain-of-Thought (CoT) prompting enhances the reasoning of large language models (LLMs) by decomposing problems into sequential steps, mimicking human logic and reducing errors. However, complex tasks with vast solution …

GSM8KMath

DC Decomposition of Nonconvex Polynomials with Algebraic Techniques

2015-10-06 · Amir Ali Ahmadi, Georgina Hall

We consider the problem of decomposing a multivariate polynomial as the difference of two convex polynomials. We introduce algebraic techniques which reduce this task to linear, second order cone, and semidefinite progra…

valid

Counterfactual Credit Policy Optimization for Multi-Agent Collaboration

2026-03-23 · Zhongyi Li, Wan Tian, Jinju Chen, Huiming Zhang 외 arxiv

Collaborative multi-agent large language models (LLMs) can solve complex reasoning tasks by decomposing roles, but reinforcement learning for such systems is limited by credit assignment: shared terminal rewards obscure …

Reinforcement LearningMathematical Reasoning