Menu
Sign In Search Podcasts Charts People & Topics Add Podcast API Pricing
Podcast Image

AI可可AI生活

AI前沿:从神经网络的秘密到AI场景生成

11 May 2025

Description

这期《TAI快报》我们聊了五篇前沿AI论文: Towards Quantifying the Hessian Structure of Neural Networks:揭示了神经网络海森矩阵“块对角”结构的真正驱动力是类别数量,而非交叉熵损失,为优化算法设计提供了新视角。 Discrete Spatial Diffusion: Intensity-Preserving Diffusion Modeling:提出离散空间扩散框架,通过颗粒随机游走实现质量守恒,拓展了扩散模型在科学领域的应用。 Steerable Scene Generation with Post Training and Inference-Time Search:开发了可控3D场景生成方法,用强化学习和搜索引导生成,满足机器人训练的特定需求。 Practical Efficiency of Muon for Pretraining:证明Muon优化器在语言模型预训练中比AdamW更省资源,并提出“伸缩式”调参算法,提升训练效率。 What do Language Model Probabilities Represent?:澄清了语言模型概率的三种含义,提醒我们在使用和评估时要明确目标,避免误解。完整推介:mp.weixin.qq.com

Audio
Featured in this Episode

No persons identified in this episode.

Transcription

This episode hasn't been transcribed yet

Help us prioritize this episode for transcription by upvoting it.

0 upvotes
🗳️ Sign in to Upvote

Popular episodes get transcribed faster

Comments

There are no comments yet.

Please log in to write the first comment.