*Assignment No.1* BEIT 2019 PATTERN
1. *Deep Learning*:
It's a subset of machine learning that uses neural networks to learn from data.
2. *Machine Learning Packages*:
Python packages like Scikit-Learn, Keras, TensorFlow, and PyTorch support machine learning.
3. *Keras Ecosystem*:
kerasNLP for natural language processing
kerasCV for computer vision
AutoKeras for automated model selection
Model Optimization for optimizing neural networks
kerastuner for hyperparameter tuning
4. *Theano*: Theano was used for building and training neural networks.
5. *PyTorch Tensors*:
PyTorch tensors are like the building blocks for deep learning in PyTorch.
They're similar to arrays but designed to work efficiently with GPUs.
Pyro is Uber's probabilistic programming framework for AI, and Tesla is a GPU programming library from NVIDIA.
*Assignment No.2*
1. *Feedforward Neural Network*:
A basic neural network with data flowing in one direction.
2. *How It Works*:
Input data passes through layers to make predictions.
3. *Real-Time Scenarios*:
Used for tasks like handwriting recognition and stock price prediction.
4. *Components*:
Layers of neurons with weights and activation functions.
5. *Cost Function*:
Measures prediction accuracy.
6. **Mean Square Error (MSE)**:
Common cost function for regression tasks.
7. *Loss Function*:
Measures the error for single examples.
8. *Cross-Entropy Loss*:
Used for classification tasks.
9. *Kernel Concept*:
Kernels aren't directly related to feedforward networks.
10. *MNIST and CIFAR-10*:
Datasets for digit recognition and image classification.
11. *Parameter Setting*:
Configure parameters like batch size and learning rate.
12. *Flattening*:
Converting multi-dimensional data to 1D.
13. *Sigmoid vs. Softmax*:
Activation functions for different tasks.
14. *Optimizer*:
Algorithms that help the model learn.
15. *Epochs*:
Complete passes through the dataset in training.
*Assignment No.3*
1. *Image Classification*:
Assigning labels to images.
2. *Deep Learning for Image Classification*:
Deep learning makes image classification more accurate. Common networks are CNNs, ResNets, and Inception.
3. *CNN*:
Convolutional Neural Network for image analysis.
4. *Convolution Operation and Kernel*:
A kernel slides over an image to extract features.
5. *Kernel on Input Image*:
The kernel multiplies and sums parts of the image.
6. *Types of Convolution*:
Valid and Same Convolution, and multiple Convolution Layers.
7. *Feature Extraction*:
Convolution layers learn features from edges to complex patterns.
*Assignment No.4*
1. *Anomaly Detection*:
Finding unusual patterns in data.
2. *Autoencoders*:
Neural networks for unsupervised learning and feature extraction.
3. *Applications of Autoencoders*:
Used in image denoising, recommendation systems, and anomaly detection.
4. *Anomaly Detection Algorithms*:
Various algorithms like autoencoders, isolation forests, and SVM.
5. *Difference from Novelty Detection*:
Anomaly detection finds both known and unknown anomalies, while novelty detection focuses on unknown ones
.
6. *Autoencoders Blocks*:
Include encoder and decoder, minimizing reconstruction error.
7. *Reconstruction and Reconstruction Errors*:
Reconstructing input and measuring how well it matches.
8. *MinMaxScaler from sklearn*:
Scales data features between 0 and 1.
9. *train_test_split from sklearn*:
Splits data for model evaluation.
10. *Anomaly Scores*:
Indicate data point anomaly.
11. *TensorFlow Dataset*:
Library for machine learning datasets.
12. *ECG Dataset*:
Contains electrocardiogram signals for heart analysis.
13. *Keras Optimizers*:
Algorithms that update model weights.
14. *Keras Layers*:
Dense layers for connections, Dropout layers for preventing overfitting.
15. *ReLU Activation*:
A popular activation function for neural networks.
*Assignment No.5*
1. **NLP (Natural Language Processing)**:
AI field focusing on human language.
2. *Word Embedding*:
Representing words as vectors for AI understanding.
3. *Word2Vec*:
Technique for learning word embeddings.
4. *Applications of Word Embedding*:
Used in sentiment analysis, translation, and more.
5. *CBOW Architecture*:
Predicts a word from its context.
6. *Input and Output of CBOW*:
Context words in, target word out.
7. *Tokenizer*:
Breaks text into words.
8. *Window Size in CBOW*:
Controls the number of context words.
9. *Embedding and Lambda Layers in Keras*:
Used for word embeddings and custom operations.
10. **yield()**:
Pauses functions in Python to iterate over large datasets.
*Assignment No.6*
1. *Transfer Learning*: Adapting a model trained on one task to another.
2. *Pretrained Models*: Ready-made models for various tasks.
3. *PyTorch*: Deep learning library with dynamic graphs.
4. *Advantages of Transfer Learning*: Faster learning, less data needed.
5. *Applications of Transfer Learning*: Used in image, text, and more tasks.
6. *Datasets*: Caltech 101 for object recognition and ImageNet for benchmarking.
7. *Transfer Learning Steps*: Choose model, adjust, load weights, fine-tune.
8. *Data Augmentation*: Creating data variations to prevent overfitting.
9. *Preprocessing*: Prepares data for models.
10. *PyTorch Transforms Module*: Operations for data preprocessing.
11. *Validation Transforms*: Consistent data preparation for validation.
12. *VGG-16 Model*: Effective model for image classification tasks.
0 Comments