1
Unifying Physical Backpropagation
统一物理反向传播框架，连接神经科学与物理计算，为硬件训练算法提供理论新视角。
arXiv:2608.11585v1 Announce Type: cross Abstract: Physical computing systems exploit device dynamics for computation, but their gradient-based optimiz…
2
FORGE: Fused On-Register Gradient Elimination for Memory-Efficient LLM Training
融合反向传播与优化器阶段的梯度处理，大幅降低LLM训练的内存峰值，突破显存瓶颈的新方案。
arXiv:2606.22932v1 Announce Type: new Abstract: Reverse-mode differentiation computes every weight gradient, writes it to memory, and only then lets t…
3
Self-Evolving Multi-Agent Systems via Textual Backpropagation
提出基于文本反向传播的多智能体自我进化框架，让智能体在协作中自动优化策略，无需人工干预。
arXiv:2506.09046v3 Announce Type: replace Abstract: Leveraging multiple Large Language Models (LLMs) has proven effective for addressing complex, high…
4
Show HN: Microcrad – Micrograd Reimplemented in C
用C语言重新实现Andrej Karpathy的微型自动微分引擎，适合学习反向传播和神经网络底层原理。
Article URL: https://github.com/oraziorillo/microcrad Comments URL: https://news.ycombinator.com/item?id=48570364 Points: 1 # Comments: 1
5
A Fully First-Order Layer for Differentiable Optimization
全一阶可微优化层，无缝集成深度学习，让优化问题也能反向传播
arXiv:2512.02494v2 Announce Type: replace Abstract: Differentiable optimization layers enable learning systems to make decisions by solving embedded o…
6
Memory-Efficient Structured Backpropagation for On-Device LLM Fine-Tuning
专为大模型设备端微调设计的内存高效结构化反向传播，突破移动设备6-12GB内存限制，实现隐私保护与性能平衡。
arXiv:2602.13069v2 Announce Type: replace Abstract: On-device fine-tuning enables privacy-preserving personalization of large language models, but mob…
7
Misalignment Between Backpropagation and the Hierarchy of Brain Responses to Images
揭示反向传播与大脑图像处理层级间的错位，为理解AI与生物智能差异提供新视角。
arXiv:2605.28693v1 Announce Type: cross Abstract: Backpropagation is the core learning mechanism underlying deep learning. However, whether and how th…
8
Securing Multi-Agent Systems Against Corruptions via Node Contribution Backpropagation
多智能体系统新防御：用节点贡献反向传播精准隔离腐败节点，提升系统鲁棒性
arXiv:2510.19420v2 Announce Type: replace-cross Abstract: Multi-Agent Systems (MAS) have become a prevalent paradigm for Large Language Model (LLM) ap…