AstroPT对星系的认知，以及这能为我们提供关于大型语言模型（LLMs）的启示
What AstroPT knows about galaxies, and what that can teach us about LLMs
浏览论文内容
中文总结 AI 辅助
该研究利用基于星系图像训练的类LLM模型AstroPT作为测试平台，探究星系属性的出现顺序，为校准应用于LLMs的机械可解释性方法提供了受控沙盒。
中文摘要 AI 辅助
可解释性研究日益关注训练过程中概念的出现时机，以及线性探测能否恢复真实结构，但在语言模型中，这些主张难以验证，因为语言几乎无法提供概念间或概念与关系间的真实排序依据。我们提出利用天文真实值，通过AstroPT——一个基于数百万星系图像训练的Transformer——作为校准测试平台。AstroPT是类LLM的模型，在一个预先已知概念难度排序及概念间关系的领域内训练。我们在检查点、层、模型规模和目标选择上探测冻结的表征，发现星系属性以固定顺序出现，该顺序与它们的已知难度一致：几乎直接写入像素的量（如波段星等）在训练早期即可解码，且在网络浅层；而基于多波段/光谱的量及推断量（如红移和特定恒星形成率）则出现较晚，且在网络深层。该顺序对我们测试的训练目标具有不变性，且随模型容量在幅度上变化，但序列不变。我们的线性探测方向进一步恢复了星系属性间的已知物理结构。我们的发现表明，天文学为校准原本盲目应用于LLMs的机械可解释性方法提供了受控沙盒。
英文摘要
Interpretability research increasingly asks when concepts emerge during training and whether linear probes recover real structure, but in language models these claims are hard to validate because language offers little ground-truth ordering of concepts or relationships among them. We propose the use of astronomical ground truth through AstroPT, a transformer trained on millions of galaxy images, as a calibration testbed. AstroPT is an LLM-like model trained within a domain where the difficulty ordering of concepts and the relations among them are known in advance. Probing frozen representations across checkpoints, layers, model sizes, and objective choices, we find that galaxy properties emerge in a fixed order that tracks their known difficulty---quantities written almost directly into the pixels (band magnitude) become decodable early in training and shallow in the network, while multiband/spectra based and inferred quantities (such as redshift and specific star formation rate) emerge later and deeper. This order is invariant to our tested training objectives, and scales in magnitude but not in sequence with capacity. Our linear probe directions further recover the known physical structure among galaxy properties. Our findings suggest that astronomy offers a controlled sandbox for calibrating mechanistic interpretability methods we otherwise apply to LLMs blind.
发表机构
- IUCAA（印度大学间天文和天体物理中心）
- AstroAI
- Washington University in St. Louis（圣路易斯华盛顿大学）
机构由 AI 辅助整理，请以论文原文为准。
