Image Multi Classification

The task of assigning multiple labels to an entire image, indicating what object or scene is present in the image.

Popular Datasets

1. CIFAR-10

• Description: 60,000 32x32 color images in 10 classes, with 6,000 images per class.
• URL: CIFAR-10

2. CIFAR-100

• Description: 60,000 32x32 color images in 100 classes, with 600 images per class.
• URL: CIFAR-100

3. ImageNet

• Description: Over 14 million images categorized into over 20,000 classes.
• URL: ImageNet

4. MNIST

• Description: 70,000 28x28 grayscale images of handwritten digits in 10 classes.
• URL: MNIST

5. Fashion-MNIST

• Description: 70,000 28x28 grayscale images of 10 fashion categories.
• URL: Fashion-MNIST

Popular Models

1. Convolutional Neural Networks (CNNs)

• Examples: LeNet, AlexNet, VGG, ResNet, DenseNet

2. Residual Networks (ResNet)

• Description: Introduces skip connections to prevent vanishing gradients.
• URL: ResNet

3. DenseNet

• Description: Connects each layer to every other layer in a feed-forward fashion.
• URL: DenseNet

4. Inception Networks (GoogLeNet)

• Description: Uses multiple types of convolutions in parallel to capture different features.
• URL: Inception

5. EfficientNet

• Description: Scales up model size while balancing network depth, width, and resolution.
• URL: EfficientNet

Hyperparameters

1. Learning Rate

• Description: Controls how much to change the model in response to the estimated error each time the model weights are updated.

2. Batch Size

• Description: The number of training examples utilized in one iteration.

3. Number of Epochs

• Description: The number of complete passes through the training dataset.

4. Optimizer

• Examples: SGD, Adam, RMSprop

5. Weight Initialization

• Examples: Xavier, He initialization

6. Dropout Rate

• Description: Fraction of the input units to drop.

7. Learning Rate Decay

• Description: Reduces the learning rate as training progresses.

Popular Loss Functions

1. Cross-Entropy Loss

• Description: Measures the performance of a classification model whose output is a probability value between 0 and 1.

2. Hinge Loss

• Description: Used for “maximum-margin” classification, mostly for SVMs.

3. Kullback-Leibler Divergence Loss

• Description: Measures how one probability distribution diverges from a second, expected probability distribution.

Popular Evaluation Metrics

1. Accuracy

• Description: The ratio of correctly predicted observation to the total observations.

2. Precision

• Description: The ratio of correctly predicted positive observations to the total predicted positives.

3. Recall (Sensitivity)

• Description: The ratio of correctly predicted positive observations to the all observations in actual class.

4. F1 Score

• Description: The weighted average of Precision and Recall.

5. Confusion Matrix

• Description: A table used to describe the performance of a classification model.

6. ROC-AUC

• Description: A graphical plot that illustrates the diagnostic ability of a binary classifier system as its discrimination threshold is varied.

Other Important Topics

1. Data Augmentation

• Description: Techniques used to increase the diversity of your training data without actually collecting new data.
• Examples: Rotation, Translation, Scaling, Flipping, Adding Noise

2. Transfer Learning

• Description: Reusing a pre-trained model on a new problem.
• Example: Fine-tuning a model pre-trained on ImageNet for a new classification task.

3. Regularization Techniques

• Examples: L1/L2 regularization, Dropout, Early Stopping

4. Ensemble Methods

• Description: Combining the predictions of multiple models to improve generalizability and robustness.
• Examples: Bagging, Boosting, Stacking

5. Fine-Tuning

• Description: Adjusting the parameters of an existing model to better fit the new data.

6. Hyperparameter Tuning

• Techniques: Grid Search, Random Search, Bayesian Optimization

7. Model Interpretability

• Techniques: SHAP values, LIME