【人工智能】英文学习材料03（每日一句）
🌻个人主页：相洋同学
🥇学习在于行动、总结和坚持，共勉！
目录
Chain Rule (链式法则)
Dimensionality Reduction (降维)
Long Short-Term Memory (LSTM) (长短期记忆网络)
Gradient Explosion (梯度爆炸)
Gradient Vanishing (梯度消失)
Dropout (Dropout)
Seq2Seq (Seq2Seq)
One-Hot Encoding (One-Hot 编码)
Self-Attention Mechanism (自注意力机制)
Multi-Head Attention Mechanism (多头注意力机制)
Chain Rule (链式法则)
The Chain Rule represents an essential principle in calculus for determining derivatives of composite functions. It formulates that when you have two functions where one follows or operates on another, their combined rate of change (derivative) can be found by multiplying their individual rates (derivatives).
essential
Dimensionality Reduction (降维)
Dimensionality reduction is a technique that involves identifying key components within high-dimensional data to represent it more efficiently with fewer variables. This method is commonly applied in machine learning and statistical analysis for model simplification, enhancing computational efficiency, and eliminating unnecessary noise within datasets.
- 涉及（概念、指的是）
- 随机变项（随机变量）
- 主要变数（主要变量）
- 统计学（统计学）
- 简化过程（简化方法）
Long Short-Term Memory (LSTM) (长短期记忆网络)
Long Short-Term Memory (LSTM) networks are a unique category within Recurrent Neural Networks (RNNs), distinguished by their capacity to learn from long-term dependencies. The LSTM architecture is engineered to circumvent the challenges associated with maintaining information over extended periods, effectively enabling sustained memory capabilities.
- long-range dependencies (长程依赖)
- long-term dependency challenge (长期依赖挑战)
- periodic intervals (周期间隔)
Gradient Explosion (梯度爆炸)
Gradient explosion is a phenomenon encountered during the training of deep neural networks, characterized by excessively large gradients of the loss function. This leads to weight updates that are significantly excessive, resulting in deviations from optimal values and ultimately causing instability in the training process and divergence.
- overshoot (超过)
- optimal values (最优值)
- unstable (不稳定)
- divergence (发散)
Gradient Vanishing (梯度消失)
Gradient vanishing is a challenge commonly faced during the training of deep neural networks, where the gradients of the loss function become extremely small, significantly impeding the training process or halting it entirely as the network weights remain unable to update in a meaningful manner.
- encountered (遇到)
- significantly (显著地)
- altogether (完全)
- meaningful way (有意义的方式)
Dropout (Dropout)
Dropout acts as a regularization technique applied during neural network training to mitigate overfitting. By randomly deactivating a subset of neurons while being trained, dropout encourages the network to develop resilient features without relying on specific neuron subsets.
- The regularization technique is a widely adopted method for preventing overfitting in machine learning models.
- The prevent mechanism ensures that the model does not overemphasize noise during training.
- The omission of certain parameters helps maintain the model's generalization capability.
- The subset selection process focuses on identifying the most informative features.
- Robust features are crucial for maintaining model performance across diverse datasets.
- Dependent variables are analyzed to understand their relationships with target outcomes.
- A single set of data is used to train and validate each model instance.
Seq2Seq (Seq2Seq)
The Seq2Seq model, also known as the Sequence to Sequence model, operates within the domain of machine learning by converting a series of tokens, such as words within a phrase, into another sequential output. This architecture is notably employed in tasks like machine translation, where an input sentence in one language is translated into a corresponding output sentence in another language.
- Sequence to Sequence (序列到序列)
- transforms into (转换为)
- sequence of items (物品系列)
- components of the input (输入的组成部分)
- becomes a new output sequence（成为新的输出系列）
One-Hot Encoding (One-Hot 编码)
One-Hot Encoding represents a method for converting classification variables into an input format that enhances the performance of ML algorithms in prediction tasks. Each category is represented by a vector consisting of one element with the value 1 and all others as 0.
- categorial variables (类别变量)
- transformed (转换)
- machine learning algorithms (机器学习算法)
- expresses (表示)
- class (类别)
- item (元素)
Self-Attention Mechanism (自注意力机制)
The Self-Attention Mechanism enables a model to assess the significance of various portions of the input data in distinct ways. It serves as a fundamental element within Transformer models, allowing them to dynamically prioritize which aspects of the input to focus on as they process information.
权衡核心组件以实现动态地基于优先级的数据分析处理
Multi-Head Attention Mechanism (多头注意力机制)
The Multi-Head Attention Mechanism acts as a key component within Transformer models, enabling them to access diverse subspace representations located at different positions. By simultaneously executing multiple self-attention operations, it significantly improves the model's capacity to concentrate on diverse input features at once.
- pay attention to (关注)
- representation subspaces (表示子空间)
- locations (位置)
- perform (执行)
- self-attention processes (自注意力操作)
- parallel (并行)
- improve (增强)
- different aspects (各个方面)
- at the same time (同时)
以上
君子坐而论道，少年起而行之，共勉
