“their assertion that ILP can't address noisy or ambiguous data is plainly wrong”
Learning Explanatory Rules from Noisy Data
可微分归纳逻辑编程:用反向传播学规则,既抗噪又能接神经网络
论文提出可微分归纳逻辑(Differentiable Inductive Logic)框架,既能解决传统归纳逻辑编程(ILP)擅长的任务,又对训练数据中的噪声和错误具有传统 ILP 难以应对的鲁棒性。由于它通过针对似然目标的反向传播训练,还可与神经网络相连,处理像原始像素这类含糊的非符号数据,同时保持数据效率和神经网络自身难以达到的泛化能力。
HN 上的人常把它作为 DeepMind 神经 ILP、神经符号混合推理的例子来引用,也用它讨论 ILP 凭借背景知识降低样本复杂度;有评论者认为其“ILP 无法处理噪声数据”的断言明显有误。
学习路径
读这篇之前,按顺序读书架上的 4 篇 · 参考文献来自 OpenAlex,顺序和要点由 Claude Code 排出这条路径先看神经网络学习算法和程序的早期尝试,再看可微解释器与程序归纳语言,理解 dILP 要在抗噪和可解释之间解决什么问题。
- 1
- 2
- 3
把程序草图做成可微解释器,和 dILP 同属可微程序归纳一类。
读程序草图中空位由数据训练的思路,以及可微 Forth 机器的结构。
- 4
比较各种程序归纳的推断算法,其中包含基于梯度的路线。
读TerpreT 把程序归纳写成概率编程的方式,以及不同后端的比较结论。
- 5Learning Explanatory Rules from Noisy Data
目标论文:用可微分的归纳逻辑编程从噪声数据中学规则。
读可微推理的构造、用反向传播学习子句权重,以及抗噪实验。
HN 书评
推荐在前,同类按时间 · 每篇最多保留 20 条,统计用全部提及“For a slightly more impartial opinion, here's a DeepMind paper that performs neural ILP: [LINK]”
“An example of such an approach taken with success is given at the end: Learning Explanatory Rules from Noisy Data”
“As a source of this claim I always quote this DeepMind paper where Metagol is compared to the authors' own system”
“This is one of the theories learned by the δILP system for the less-than relation between images of numbers, taken from the pdf version of the paper [1].”
“For an example, see this paper by a couple of people at DeepMind:”
看全部 8 条 →
“Learning Explanatory Rules from Noisy Data by DeepMind, which also uses metarules.”
“So the work described in the paper basically benefits from ILP's ability to use background knowlede to reduce sample complexity.”
- Learning Explanatory Rules from Noisy Data [pdf] ↗stablemap · 2017-11-15 · 3 分
- Learning Explanatory Rules from Noisy Data [with Neural Networks] ↗YeGoblynQueenne · 2018-02-16 · 4 分