发表机构
Monash University（莫纳什大学）
机构由 AI 辅助整理，请以论文原文为准。AI 中文总结
研究贝叶斯滤波变压器（BFT）在预测时内化的潜在任务先验和后验，现有预测空间比较方法脆弱。提出用预测蒙特卡罗（PMC）作为通用解释工具，通过下一个token生成近似潜在任务的隐式先验和后验，并应用于三个任务族验证，直接在潜在空间回答解释性问题。
AI 中文摘要
贝叶斯滤波变压器（BFT）是在分两步生成的序列上训练的变压器：首先从先验中抽取一个潜在任务，然后根据该任务抽取观测值。在自回归对数损失下训练，理想情况下，BFT的下一个token预测是由该先验和条件律诱导的贝叶斯后验预测分布（PPD）。实际上，训练后的BFT只是这个理想PPD的近似，这就产生了一个解释性问题：训练后的BFT实际上内化了关于潜在任务的什么先验和后验？现有工作通过将训练后的BFT的预测与各种“参考”后验的预测进行比较来回答这个问题。这种预测空间比较很脆弱。我们使用预测蒙特卡罗（PMC）作为任何BFT的通用可解释性工具，仅通过下一个token生成，PMC就能返回潜在任务上隐式先验和后验的近似值，直接在潜在空间中回答解释性问题。我们将PMC应用于跨越0-马尔可夫和1-马尔可夫可交换性的三个风格化任务族。之前在这些设置中报告的现象在潜在空间中仍然可见。代码可在这个https URL获取。
英文摘要
A Bayes-filtered transformer (BFT) is a transformer trained on sequences that are generated in two steps: first a latent task is drawn from a prior, then observations are drawn conditional on that task. Trained under autoregressive log loss, the BFT's next-token prediction, in the idealized limit, is the Bayesian posterior predictive distribution (PPD) induced by that prior and that conditional law. In practice the trained BFT is only an approximation of this ideal PPD, raising an interpretive question: what prior and posterior over the latent task has the trained BFT actually internalized? Existing work answers this question by comparing the trained BFT's predictions against the predictions of various "reference" posteriors, each standing in for a different candidate algorithm or computation the BFT might be implementing. This prediction-space comparison is fragile: different posteriors can share the same posterior-mean predictions. We use predictive Monte Carlo (PMC) as a general interpretability tool for any BFT: using only next-token generation, PMC returns an approximation to the implicit prior and posterior over the latent task, answering the interpretive question directly in latent space. We apply PMC to three stylized task families spanning 0-Markov and 1-Markov exchangeability. The phenomena previously reported in these settings remain visible in latent space. Code is available at https://github.com/afiq-aswadi/bft-pmc
