컴퓨터 비전 논문에서 많이 나오는 표현 정리

모든 major paper을 통합한 표현이고 명사, 동사, 문법적 표현순으로 정리됨

Nouns:

1. Tensor - A mathematical object that can be thought of as a generalization of a vector or matrix. In deep learning, tensors are used to represent data, such as images or other forms of input, as well as the weights and biases of neural networks.
2. Embedding - A technique that maps categorical variables or discrete data, such as words or images, into a continuous vector space. It is commonly used in natural language processing (NLP) and computer vision tasks to represent data in a way that can be easily processed by neural networks.
3. Kernel - In the context of deep learning, a kernel is a small matrix that is applied to an image or feature map to perform convolutional operations. Convolutional kernels are used to extract features from input data, such as edges or textures, and are a fundamental component of many deep learning models.
4. Preprocessing - The process of transforming raw input data, such as images or audio, into a format that can be easily processed by a deep learning model. Preprocessing techniques may include normalization, resizing, or other types of data augmentation.
5. Regularization - A technique used to prevent overfitting in deep learning models by adding additional constraints or penalties to the model's weights or parameters. Common forms of regularization include L1 and L2 regularization, which add a penalty to the sum of the absolute or squared values of the weights.
6. Activation - In deep learning, activation refers to the non-linear function that is applied to the output of a neural network layer. Common activation functions include ReLU, sigmoid, and tanh, and they are used to introduce non-linearities into the model and enable it to learn complex relationships in the data.
7. Gradient - In deep learning, the gradient refers to the derivative of the loss function with respect to the model's parameters. It is used in the backpropagation algorithm to update the model's weights and improve its performance.
8. Epoch - An epoch is a complete iteration through the entire training dataset during model training. The number of epochs is a hyperparameter that can be tuned to optimize model performance.
9. Inference - The process of using a trained deep learning model to make predictions on new data. Inference involves passing the new data through the model and generating output predictions.
10. Backpropagation - A technique used to compute the gradient of the loss function with respect to the model's parameters by propagating errors backward through the network. Backpropagation is used to update the model's weights and improve its performance during training.
11. Augmentation - The process of generating new training data by applying random transformations or distortions to the original data. Augmentation is commonly used in deep learning to increase the size of the training dataset and improve model performance.
12. Overfitting - A phenomenon that occurs when a deep learning model learns to fit the training data too closely, resulting in poor performance on new, unseen data. Overfitting can be prevented using techniques such as regularization or early stopping.
13. Dropout - A regularization technique used to prevent overfitting by randomly dropping out units in a neural network during training. Dropout forces the network to learn more robust features that are less sensitive to the input data.
14. Encoder - A component of an autoencoder deep learning model that maps the input data to a lower-dimensional representation. The encoder is used to compress the input data and extract the most important features.
15. Decoder - The component of an autoencoder that maps the compressed data back to its original dimensions. The decoder is used to reconstruct the original input data from the lower-dimensional representation created by the encoder.

 

 

Verbs:

1. Converge - In deep learning, this refers to the point at which the model has reached its optimal state, and the loss function is no longer decreasing. It means the model is no longer improving, and further training is unnecessary.
2. Regularize - This refers to the process of adding constraints or penalties to the model's parameters to prevent overfitting and improve generalization performance.
3. Propagate - This refers to the forward and backward flow of information through a deep learning model. During forward propagation, input data is passed through the model to generate output predictions, while during backpropagation, errors are propagated backwards through the network to update the model's weights.
4. Normalize - A technique used to preprocess data by scaling the input data to have zero mean and unit variance. Normalization can help improve model performance and reduce the risk of vanishing or exploding gradients.
5. Instantiate - This refers to the process of creating an instance of a class in object-oriented programming. In the context of deep learning, it is used to create neural network models with specific configurations of layers and parameters.
6. Discriminate - In deep learning, this refers to the process of distinguishing between different classes or categories of data. Discrimination is a fundamental task in computer vision, where models are trained to recognize and classify different types of objects or features.
7. Regularize - A technique used to prevent overfitting in deep learning models by adding additional constraints or penalties to the model's weights or parameters. Common forms of regularization include L1 and L2 regularization, which add a penalty to the sum of the absolute or squared values of the weights.
8. Synthesize - This refers to the process of generating new data or images using deep learning models. In computer vision, generative models such as GANs can be used to synthesize new images that resemble the training data.
9. Deconvolve - A process used to reconstruct the input data from a feature map or compressed representation. Deconvolution is used in some types of deep learning models, such as autoencoders, to reconstruct the original input data from a lower-dimensional representation.
10. Initialize - In deep learning, this refers to the process of setting the initial values of the model's weights and biases. Proper initialization is critical to achieving good performance in deep learning models.

 

Grammatical expressions:

1. Convolutional Neural Network (CNN) - This is a specific type of deep learning model commonly used in computer vision tasks. CNNs are designed to identify patterns in image data by applying filters to the input image.
2. Long Short-Term Memory (LSTM) - This is a type of recurrent neural network (RNN) architecture that is commonly used in natural language processing tasks. LSTMs are designed to model the long-term dependencies in sequential data, such as sentences or speech.
3. Backpropagation - This is a common algorithm used to train deep learning models. During backpropagation, errors are propagated backwards through the network to update the model's weights and improve its performance.
4. Stochastic Gradient Descent (SGD) - This is a popular optimization algorithm used in deep learning to update the weights of a neural network during training. SGD updates the weights using a small subset of the training data, called a mini-batch, to speed up training and reduce memory requirements.
5. Dropout - This is a regularization technique used to prevent overfitting in deep learning models. Dropout randomly drops out (sets to zero) a fraction of the model's neurons during training, forcing the remaining neurons to learn more robust and generalizable features.
6. Batch Normalization - This is a technique used to improve the stability and performance of deep learning models by normalizing the input data to each layer. Batch normalization helps to reduce the effects of internal covariate shift, which is the change in the distribution of input data to a layer during training.
7. Activation Function - This is a function applied to the output of each neuron in a deep learning model to introduce non-linearity and improve the model's expressive power. Common activation functions include ReLU, sigmoid, and tanh.
8. Loss Function - This is a function used to measure the difference between the model's predicted output and the actual output. The loss function is used during training to update the model's weights and improve its performance.
9. Gradient Descent - This is a common optimization algorithm used to update the weights of a neural network during training. Gradient descent calculates the gradient of the loss function with respect to the weights and updates the weights in the direction of the steepest descent to minimize the loss.
10. Hyperparameter - This is a parameter in a deep learning model that is set manually by the user, rather than learned from the data during training. Hyperparameters include things like the learning rate, number of layers, and number of neurons in each layer, and can have a significant impact on the performance of the model.

 

 

2019년 이후 나온 major paper의 표현들

Nouns:

1. GAN (Generative Adversarial Network) - a type of deep learning model that generates new data by training two networks, a generator and a discriminator, to compete against each other.
2. Transformer - a type of neural network architecture used for sequence-to-sequence tasks, such as language translation.
3. Self-Attention - a mechanism in the Transformer architecture that allows the model to attend to different parts of the input sequence.
4. Pretraining - a technique used to improve the performance of deep learning models by training them on a large amount of data before fine-tuning on a specific task.
5. Contrastive Learning - a type of unsupervised learning where the model learns to differentiate between similar and dissimilar examples.
6. Few-Shot Learning - a type of machine learning where the model is trained to learn new concepts with only a few examples.
7. Object Detection - a task in computer vision where the model is trained to identify and locate objects in an image.
8. Semantic Segmentation - a task in computer vision where the model is trained to assign a label to each pixel in an image.
9. Transfer Learning - a technique where a pre-trained model is used as a starting point for a new task.
10. Feature Extraction - a process where the model extracts important features from the input data.

Verbs:

1. Pretrain - to train a model on a large amount of data before fine-tuning on a specific task.
2. Fine-Tune - to train a model on a specific task with a smaller amount of data, after pretraining on a larger dataset.
3. Embed - to represent an input in a high-dimensional vector space.
4. Discriminate - to distinguish between similar and dissimilar examples.
5. Regularize - to prevent overfitting in a model.
6. Augment - to increase the amount of training data by generating new examples.
7. Attend - to focus the model's attention on specific parts of the input sequence.
8. Encode - to convert the input data into a high-dimensional vector representation.
9. Segment - to assign a label to each pixel in an image.
10. Propagate - to update the model's weights using backpropagation.

Grammatical Expressions:

1. End-to-End - a type of learning where the model is trained to map inputs directly to outputs, without intermediate steps.
2. Data Augmentation - a technique used to increase the amount of training data by generating new examples from the existing data.
3. Inference - the process of using a trained model to make predictions on new, unseen data.
4. Cross-Entropy - a loss function commonly used in classification tasks.
5. Dropout - a regularization technique used to prevent overfitting in deep learning models.
6. Gradient Descent - a common optimization algorithm used to update the weights of a neural network during training.
7. Learning Rate - a hyperparameter in a deep learning model that controls the size of weight updates during training.
8. Overfitting - a problem where the model performs well on the training data but poorly on new, unseen data.
9. Convolution - a mathematical operation commonly used in computer vision to extract features from an image.
10. Activation Function - a function applied to the output of each neuron in a deep learning model to introduce non-linearity and improve the model's expressive power.