深度学习基础 Probabilistic Graphical Models | Statistical and Algorithmic Foundations of Deep Learning

深度学习基础 Probabilistic Graphical Models | Statistical and Algorithmic Foundations of Deep Learning
目录

Probabilistic Graphical Models

Statistical and Algorithmic Foundations of Deep Learning

Author: Eric Xing

01 An overview of DL components

Historical remarks: early days of neural networks

我们知道生物神经元是这样的:
深度学习基础 Probabilistic Graphical Models | Statistical and Algorithmic Foundations of Deep Learning
上游细胞通过轴突(Axon)将神经递质传送给下游细胞的树突。 人工智能受到该原理的启发,是按照下图来构造人工神经元(或者是感知器)的。
深度学习基础 Probabilistic Graphical Models | Statistical and Algorithmic Foundations of Deep Learning
类似的,生物神经网络 —— > 人工神经网络
![在这里插入图片描述](https://www.icode9.com/i/ll/?i=2020051209264072.png?,type_ZmFuZ3poZW5naGVpdGk,shadow_10,text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L05HVWV2ZXIxNQ==,size_16,color_FFFFFF,t_70Reverse-mode automatic differentiation (aka backpropagation)

Reverse-mode automatic differentiation (aka backpropagation)

下面我们来看看具体的感知器学习算法。
深度学习基础 Probabilistic Graphical Models | Statistical and Algorithmic Foundations of Deep Learning
假设这是一个回归问题x->y,\(y = f(x)+\eta\)$, 则目标函数为
深度学习基础 Probabilistic Graphical Models | Statistical and Algorithmic Foundations of Deep Learning
为了求出该函数的解,我们需要对其求导,具体的:
深度学习基础 Probabilistic Graphical Models | Statistical and Algorithmic Foundations of Deep Learning
其中
深度学习基础 Probabilistic Graphical Models | Statistical and Algorithmic Foundations of Deep Learning

由此\(w\)的更新公式为:
深度学习基础 Probabilistic Graphical Models | Statistical and Algorithmic Foundations of Deep Learning

下面我们来说说神经网络模型:
深度学习基础 Probabilistic Graphical Models | Statistical and Algorithmic Foundations of Deep Learning
其中,隐藏单元没有目标。

人工神经网络不过是可以由计算图表示的复杂功能组成。
深度学习基础 Probabilistic Graphical Models | Statistical and Algorithmic Foundations of Deep Learning
通过应用链式规则并使用反向累积,我们得到:
深度学习基础 Probabilistic Graphical Models | Statistical and Algorithmic Foundations of Deep Learning
该算法通常称为反向传播。 如果某些功能是随机的怎么办?使用随机反向传播!现代软件包可以自动执行此操作(稍后再介绍)

Modern building blocks: units, layers, activations functions, loss functions, etc.

常用激活函数:

  • Linear and ReLU
  • Sigmoid and tanh
  • Etc.

网络层:

  • Fully connected
  • Convolutional & pooling
  • Recurrent
  • ResNets
  • Etc.
    -深度学习基础 Probabilistic Graphical Models | Statistical and Algorithmic Foundations of Deep Learning
    深度学习基础 Probabilistic Graphical Models | Statistical and Algorithmic Foundations of Deep Learning

也就是说基本构成要素的可以任意组合,如果有多种损失功能的话,可以实现多目标预测和转移学习等。 只要有足够的数据,更深的架构就会不断改进。

Feature learning
成功学习中间表示[Lee et al ICML 2009,Lee et al NIPS 2009]
深度学习基础 Probabilistic Graphical Models | Statistical and Algorithmic Foundations of Deep Learning
表示学习:网络学习越来越多的抽象数据表示形式,这些数据被“解开”,即可以进行线性分离。

02 Similarities and differences between GMs and NNs

Graphical models vs. computational graphs

Graphical models:

  • 用于以图形形式编码有意义的知识和相关的不确定性的表示形式
  • 学习和推理基于经过充分研究(依赖于结构)的技术(例如EM,消息传递,VI,MCMC等)的丰富工具箱
  • 图形代表模型
    深度学习基础 Probabilistic Graphical Models | Statistical and Algorithmic Foundations of Deep Learning
    Utility of the graph
  • 一种用于从局部结构综合全局损失函数的工具(潜在功能,特征功能等)
  • 一种设计合理有效的推理算法的工具(总和,均值场等)
  • 激发近似和惩罚的工具(结构化MF,树近似等)
  • 用于监视理论和经验行为以及推理准确性的工具

Utility of the loss function

  • 学习算法和模型质量的主要衡量指标

深度学习基础 Probabilistic Graphical Models | Statistical and Algorithmic Foundations of Deep Learning
深度学习基础 Probabilistic Graphical Models | Statistical and Algorithmic Foundations of Deep Learning
深度学习基础 Probabilistic Graphical Models | Statistical and Algorithmic Foundations of Deep Learning
Deep neural networks :

  • 学习有助于最终指标上的计算和性能的表示形式(中间表示形式不保证一定有意义)
  • 学习主要基于梯度下降法(aka反向传播);推论通常是微不足道的,并通过“向前传递”完成
  • 图形代表计算

Utility of the network

  • 概念上综合复杂决策假设的工具(分阶段的投影和聚合)
  • 用于组织计算操作的工具(潜在状态的分阶段更新)
  • 用于设计加工步骤和计算模块的工具(逐层并行化)
  • 在评估DL推理算法方面没有明显的用途

深度学习基础 Probabilistic Graphical Models | Statistical and Algorithmic Foundations of Deep Learning

到目前为止,图形模型是概率分布的表示,而神经网络是函数近似器(无概率含义)。有些神经网络实际上是图形模型(即单位/神经元代表随机变量):

  • 玻尔兹曼机器Boltzmann machines (Hinton&Sejnowsky,1983)
  • 受限制的玻尔兹曼机器Restricted Boltzmann machines(Smolensky,1986)
  • Sigmoid信念网络的学习和推理Learning and Inference in sigmoid belief networks(Neal,1992)
  • 深度信念网络中的快速学习Fast learning in deep belief networks(Hinton,Osindero,Teh,2006年)
  • 深度玻尔兹曼机器Deep Boltzmann machines(Salakhutdinov和Hinton,2009年)

接下来我们会逐一介绍他们。

I: Restricted Boltzmann Machines
受限玻尔兹曼机器,缩写为RBM。 RBM是用二部图(bi-partite graph)表示的马尔可夫随机场,图的一层/部分中的所有节点都连接到另一层中的所有节点; 没有层间连接。
深度学习基础 Probabilistic Graphical Models | Statistical and Algorithmic Foundations of Deep Learning
联合分布为:
深度学习基础 Probabilistic Graphical Models | Statistical and Algorithmic Foundations of Deep Learning
单个数据点的对数似然度(不可观察的边际被边缘化):
深度学习基础 Probabilistic Graphical Models | Statistical and Algorithmic Foundations of Deep Learning
对数似然比的梯度 模型参数:
深度学习基础 Probabilistic Graphical Models | Statistical and Algorithmic Foundations of Deep Learning
对数似然比的梯度 参数(替代形式):
深度学习基础 Probabilistic Graphical Models | Statistical and Algorithmic Foundations of Deep Learning
两种期望都可以通过抽样来近似, 从后部采样是准确的(RBM在给定的h上分解)。 通过MCMC从关节进行采样(例如,吉布斯采样)

在神经网络文献中:

  • 计算第一项称为钳位/唤醒/正相(网络是“清醒的”,因为它取决于可见变量)
  • 计算第二项称为非固定/睡眠/*/负相(该网络“处于睡眠状态”,因为它对关节的可见变量进行了采样;比喻,它梦见了可见的输入)

通过随机梯度下降(SGD)优化给定数据的模型对数似然来完成学习, 第二项(负相)的估计严重依赖于马尔可夫链的混合特性,这经常导致收敛缓慢并且需要额外的计算。

II: Sigmoid Belief Networks
深度学习基础 Probabilistic Graphical Models | Statistical and Algorithmic Foundations of Deep Learning
Sigimoid信念网是简单的贝叶斯网络,其二进制变量的条件概率由Sigmoid函数表示:
深度学习基础 Probabilistic Graphical Models | Statistical and Algorithmic Foundations of Deep Learning
贝叶斯网络表现出一种称为“解释效应”的现象:如果A与C相关,则B与C相关的机会减少。 ⇒在给定C的情况下A和B相互关联。
深度学习基础 Probabilistic Graphical Models | Statistical and Algorithmic Foundations of Deep Learning
值得注意的是, 由于“解释效应”,当我们以信念网络中的可见层为条件时,所有隐藏变量都将成为因变量。

Sigmoid Belief Networks as graphical models

尼尔提出了用于学习和推理的蒙特卡洛方法(尼尔,1992年):
深度学习基础 Probabilistic Graphical Models | Statistical and Algorithmic Foundations of Deep Learning
RBMs are infinite belief networks
要对模型参数进行梯度更新,我们需要通过采样计算期望值。

  • 我们可以在第一阶段从后验中精确采样
  • 我们运行吉布斯块抽样,以从联合分布中近似抽取样本
    深度学习基础 Probabilistic Graphical Models | Statistical and Algorithmic Foundations of Deep Learning

条件分布\(p(v| h)\)和\(p(h|v)\)用sigmoid表示, 因此,我们可以将以RBM表示的联合分布中的Gibbs采样视为无限深的Sigmoid信念网络中的自顶向下传播!
深度学习基础 Probabilistic Graphical Models | Statistical and Algorithmic Foundations of Deep Learning
RBM等效于无限深的信念网络。当我们训练RBM时,实际上就是在训练一个无限深的简短网, 只是所有图层的权重都捆绑在一起。如果权重在某种程度上“统一”,我们将获得一个深度信仰网络。

Deep Belief Networks and Boltzmann Machines

III: Deep Belief Nets
深度学习基础 Probabilistic Graphical Models | Statistical and Algorithmic Foundations of Deep Learning
DBN是混合图形模型(链图)。其联合概率分布可表示为:
深度学习基础 Probabilistic Graphical Models | Statistical and Algorithmic Foundations of Deep Learning

其中蕴含的挑战:
由于explaining away effect,因此在DBN中进行精确推断是有问题的
训练分两个阶段进行:

  • 贪婪的预训练+临时微调; 没有适当的联合训练
  • 近似推断为前馈(自下而上)

Layer-wise pre-training

  • 预训练并冻结第一个RBM
  • 在顶部堆叠另一个RBM并对其进行训练
  • 重物2层以上的重物保持绑紧状态
  • 我们重复此过程:预训练和解开

Fine-tuning

  • Pre-training is quite ad-hoc(特别指定) and is unlikely to lead to a good probabilistic model per se
  • However, the layers of representations could perhaps be useful for some other downstream tasks!
  • We can further “fine-tune” a pre-trained DBN for some other task

Setting A: Unsupervised learning (DBN → autoencoder)

  1. Pre-train a stack of RBMs in a greedy layer-wise fashion
  2. “Unroll” the RBMs to create an autoencoder
  3. Fine-tune the parameters by optimizing the reconstruction error(重构误差)

深度学习基础 Probabilistic Graphical Models | Statistical and Algorithmic Foundations of Deep Learning
深度学习基础 Probabilistic Graphical Models | Statistical and Algorithmic Foundations of Deep Learning
深度学习基础 Probabilistic Graphical Models | Statistical and Algorithmic Foundations of Deep Learning

Setting B: Supervised learning (DBN → classifier)

  1. Pre-train a stack of RBMs in a greedy layer-wise fashion
  2. “Unroll” the RBMs to create a feedforward classifier
  3. Fine-tune the parameters by optimizing the reconstruction error

Deep Belief Nets and Boltzmann Machines
深度学习基础 Probabilistic Graphical Models | Statistical and Algorithmic Foundations of Deep Learning
DBMs are fully un-directed models (Markov random fields). Can be trained similarly as RBMs via MCMC (Hinton & Sejnowski, 1983). Use a variational approximation(变分近似) of the data distribution for faster training (Salakhutdinov & Hinton, 2009). Similarly, can be used to initialize other networks for downstream tasks

A few critical points to note about all these models:

  • The primary goal of deep generative models is to represent the distribution of the observable variables. Adding layers of hidden variables allows to represent increasingly more complex distributions.
  • Hidden variables are secondary (auxiliary) elements used to facilitate learning of complex dependencies between the observables.
  • Training of the model is ad-hoc, but what matters is the quality of learned hidden representations.
  • Representations are judged by their usefulness on a downstream task (the probabilistic meaning of the model is often discarded at the end).
  • In contrast, classical graphical models are often concerned with the correctness of learning and inference of all variables

Conclusion

  • DL & GM: the fields are similar in the beginning (structure, energy, etc.), and then diverge to their own signature pipelines
  • DL: most effort is directed to comparing different architectures and their components (models are driven by evaluating empirical performance on a downstream tasks)
  • DL models are good at learning robust hierarchical representations from the data and suitable for simple reasoning (call it “low-level cognition”)
  • GM: the effort is directed towards improving inference accuracy and convergence speed
  • GMs are best for provably correct inference and suitable for high-level complex reasoning tasks (call it “high-level cognition”) 推理任务
  • Convergence of both fields is very promising!

03 Combining DL methods and GMs

Using outputs of NNs as inputs to GMs

Combining sequential NNs and GMs
HMM:隐马尔可夫
深度学习基础 Probabilistic Graphical Models | Statistical and Algorithmic Foundations of Deep Learning
Hybrid NNs + conditional GMs
深度学习基础 Probabilistic Graphical Models | Statistical and Algorithmic Foundations of Deep Learning
In a standard CRF条件随机场, each of the factor cells is a parameter.
In a hybrid model, these values are computed by a neural network.
深度学习基础 Probabilistic Graphical Models | Statistical and Algorithmic Foundations of Deep Learning
深度学习基础 Probabilistic Graphical Models | Statistical and Algorithmic Foundations of Deep Learning

GMs with potential functions represented by NNs q NNs with structured outputs

Using GMs as Prediction Explanations

深度学习基础 Probabilistic Graphical Models | Statistical and Algorithmic Foundations of Deep Learning
!!!! How do we build a powerful predictive model whose predictions we can interpret in terms of semantically meaningful features?

Contextual Explanation Networks (CENs)

深度学习基础 Probabilistic Graphical Models | Statistical and Algorithmic Foundations of Deep Learning

  • The final prediction is made by a linear GM.
  • Each coefficient assigns a weight to a meaningful attribute.
  • Allows us to judge predictions in terms of GMs produced by the context encoder.

CEN: Implementation Details
深度学习基础 Probabilistic Graphical Models | Statistical and Algorithmic Foundations of Deep Learning
Workflow:

  • Maintain a (sparse稀疏) dictionary of GM parameters.
  • Process complex inputs (images, text, time series, etc.) using deep nets; use soft attention to either select or combine models from the dictionary.
    • Use constructed GMs (e.g., CRFs) to make predictions.
    • Inspect GM parameters to understand the reasoning behind predictions.

Results: imagery as context
深度学习基础 Probabilistic Graphical Models | Statistical and Algorithmic Foundations of Deep Learning
Based on the imagery, CEN learns to select different models for urban and rural

Results: classical image & text datasets
深度学习基础 Probabilistic Graphical Models | Statistical and Algorithmic Foundations of Deep Learning
CEN architectures for survival analysis
深度学习基础 Probabilistic Graphical Models | Statistical and Algorithmic Foundations of Deep Learning

04 Bayesian Learning of NNs

Bayesian learning of NN parameters q Deep kernel learning

A neural network as a probabilistic model: Likelihood: \(p(y|x, \theta)\)

  • Categorical distribution for classification ⇒ cross-entropy loss 交叉熵损失
  • Gaussian distribution for regression ⇒ squared loss平方损失
  • Gaussianprior⇒L2regularization
  • Laplaceprior⇒L1regularization

Bayesian learning [MacKay 1992, Neal 1996, de Freitas 2003]
深度学习基础 Probabilistic Graphical Models | Statistical and Algorithmic Foundations of Deep Learning

上一篇:新概念1(144课)


下一篇:What is ISO Unit in SAP?