欢迎光临澳大利亚新华书店网 [登录 | 免费注册]

    • 深度学习基础(影印版)(英文版)
      • 作者:(美)尼基尔·巴杜马
      • 出版社:东南大学
      • ISBN:9787564175177
      • 出版日期:2018/02/01
      • 页数:283
    • 售价:32
  • 内容大纲

        随着神经网络在21世纪重振旗鼓,深度学习已成为极其活跃的研究领域,为现代机器学习开辟了道路。在《深度学习基础(影印版)(英文版)》这本实用书籍中,作者尼基尔·巴杜马提供清晰的解释以引导你学完这个复杂领域的主要概念。
        Google、微软和Facebook等公司正在积极发展内部的深度学习团队。对于我们而言,尽管如此,深度学习仍然是一门非常复杂和难以掌握的课题。如果你熟悉Python,并且具有微积分背景,以及对于机器学习的基本理解,本书将帮助你开启深度学习之旅。
  • 作者介绍

        Nikhil Buduma是Remedy的联合创始人和首席科学家,该公司位于美国旧金山,旨在建立数据驱动为主的健康管理新系统。16岁时,他在圣何塞州立大学管理过一个药物发现实验室,为资源受限的社区研发新颖而低成本的筛查方法。到了19岁,他是国际生物学奥林匹克竞赛的两枚金牌获得者。随后加入MIT,在那里他专注于开发大规模数据系统以影响健康服务、精神健康和医药研究。在MIT,他联合创立了Lean On Me,一家全国性的非营利组织,提供匿名短信热线在大学校园内实现有效的一对一支持,并运用数据来积极影响身心健康。如今,Nikhil通过他的风投基金QVenture Partners投资硬科技和数据公司,还为Milwaukee Brewers篮球队管理一支数据分析团队。
  • 目录

    Preface
    1. The Neural Network
      Building Intelligent Machines
      The Limits of Traditional Computer Programs
      The Mechanics of Machine Learning
      The Neuron
      Expressing Linear Perceptrons as Neurons
      Feed-Forward Neural Networks
      Linear Neurons and Their Limitations
      Sigmoid, Tanh, and ReLU Neurons
      Softmax Output Layers
      Looking Forward
    2. Training Feed-Forward Neural Networks
      The Fast-Food Problem
      Gradient Descent
      The Delta Rule and Learning Rates
      Gradient Descent with Sigmoidal Neurons
      The Backpropagation Algorithm
      Stochastic and Minibatch Gradient Descent
      Test Sets, Validation Sets, and Overfitting
      Preventing Overfitting in Deep Neural Networks
      Summary
    3. Implementing Neural Networks in TensorFIow
      What Is TensorFlow?
      How Does TensorFlow Compare to Alternatives?
      Installing TensorFlow
      Creating and Manipulating TensorFlow Variables
      TensorFlow Operations
      Placeholder Tensors
      Sessions in TensorFlow
      Navigating Variable Scopes and Sharing Variables
      Managing Models over the CPU and GPU
      Specifying the Logistic Regression Model in TensorFlow
      Logging and Training the Logistic Regression Model
      Leveraging TensorBoard to Visualize Computation Graphs and Learning
      Building a Multilayer Model for MNIST in TensorFlow
      Summary
    4. Beyond Gradient Descent
      The Challenges with Gradient Descent
      Local Minima in the Error Surfaces of Deep Networks
      Model Identifiability
      How Pesky Are Spurious Local Minima in Deep Networks?
      Flat Regions in the Error Surface
      When the Gradient Points in the Wrong Direction
      Momentum-Based Optimization
      A Brief View of Second-Order Methods
      Learning Rate Adaptation
        AdaGrad--Accumulating Historical Gradients
        RMSProp--Exponentially Weighted Moving Average of Gradients
        Adam--Combining Momentum and RMSProp

      The Philosophy Behind Optimizer Selection
      Summary
    5. Convolutional Neural Networks
      Neurons in Human Vision
      The Shortcomings of Feature Selection
      Vanilla Deep Neural Networks Don't Scale
      Filters and Feature Maps
      Full Description of the Convolutional Layer
      Max Pooling
      Full Architectural Description of Convolution Networks
      Closing the Loop on MNIST with Convolutional Networks
      Image Preprocessing Pipelines Enable More Robust Models
      Accelerating Training with Batch Normalization
      Building a Convolutional Network for CIFAR-10
      Visualizing Learning in Convolutional Networks
      Leveraging Convolutional Filters to Replicate Artistic Styles
      Learning Convolutional Filters for Other Problem Domains
      Summary
    6. Embedding and Representation Learning
      Learning Lower-Dimensional Representations
      Principal Component Analysis
      Motivating the Autoencoder Architecture
      Implementing an Autoencoder in TensorFlow
      Denoising to Force Robust Representations
      Sparsity in Autoencoders
      When Context Is More Informative than the Input Vector
      The Word2Vec Framework
      Implementing the Skip-Gram Architecture
      Summary
    7. Models for Sequence Analysis
      Analyzing Variable-Length Inputs
      Tackling seq2seq with Neural N-Grams
      Implementing a Part-of-Speech Tagger
      Dependency Parsing and SyntaxNet
      Beam Search and Global Normalization
      A Case for Stateful Deep Learning Models
      Recurrent Neural Networks
      The Challenges with Vanishing Gradients
      Long Short-Term Memory (LSTM) Units
      TensorFlow Primitives for RNN Models
      Implementing a Sentiment Analysis Model
      Solving seq2seq Tasks with Recurrent Neural Networks
      Augmenting Recurrent Networks with Attention
      Dissecting a Neural Translation Network
      Summary
    8. Memory Augmented Neural Networks
      Neural Turing Machines
      Attention-Based Memory Access
      NTM Memory Addressing Mechanisms
      Differentiable Neural Computers

      Interference-Free Writing in DNCs
      DNC Memory Reuse
      Temporal Linking of DNC Writes
      Understanding the DNC Read Head
      The DNC Controller Network
      Visualizing the DNC in Action
      Implementing the DNC in TensorFlow
      Teaching a DNC to Read and Comprehend
      Summary
    9. Deep Reinforcement Learning
      Deep Reinforcement Learning Masters Atari Games
      What Is Reinforcement Learning?
      Markov Decision Processes (MDP)
        Policy
        Future Return
        Discounted Future Return
      Explore Versus Exploit
      Policy Versus Value Learning
        Policy Learning via Policy Gradients
      Pole-Cart with Policy Gradients
        OpenAI Gym
        Creating an Agent
        Building the Model and Optimizer
        Sampling Actions
        Keeping Track of History
        Policy Gradient Main Function
        PGAgent Performance on Pole-Cart
      Q-Learning and Deep Q-Networks
        The Bellman Equation
        Issues with Value Iteration
        Approximating the Q-Function
        Deep Q-Network (DQN)
        Training DQN
        Learning Stability
        Target Q-Network
        Experience Replay
        From Q-Function to Policy
        DQN and the Markov Assumption
        DQN's Solution to the Markov Assumption
        Playing Breakout wth DQN
        Building Our Architecture
        Stacking Frames
        Setting Up Training Operations
        Updating Our Target Q-Network
        Implementing Experience Replay
        DQN Main Loop
        DQNAgent Results on Breakout
      Improving and Moving Beyond DQN
        Deep Recurrent Q-Networks (DRQN)
        Asynchronous Advantage Actor-Critic Agent (A3C)

        UNsupervised REinforcement and Auxiliary Learning (UNREAL)
      Summary
    Index