supervised nlp

ML
unsupervised deep learning

reinforcement
 r v i s e d
pe
1. Supervised Learning Algorithms:
su 2. Linear Regression
3. Logistic Regression
4. Decision Trees
5. Random Forest
6. Gradient Boosting Machines (GBM)
7. Support Vector Machines (SVM)
8. Naive Bayes
9. k-Nearest Neighbors (k-NN)
10. Neural Networks (Multi-layer Perceptron)
11. Gaussian Processes
12. Ensemble Methods (e.g., AdaBoost)
 r v i s e d
n s u p e 1. K-Means Clustering
u 2. Hierarchical Clustering
3. DBSCAN (Density-Based Spatial Clustering
of Applications with Noise)
4. Mean Shift
5. Expectation-Maximization (EM) Algorithm
6. Principal Component Analysis (PCA)
7. Independent Component Analysis (ICA)
8. Autoencoders
9. Generative Adversarial Networks (GANs)
10. Self-Organizing Maps (SOM)
 m e n t
f o r c e
re i n 1. Q-Learning
2. Deep Q-Networks (DQN)
3. Policy Gradient Methods
4. Actor-Critic Methods
5. Monte Carlo Methods
6. Temporal Difference (TD)
Learning
7.
 e
de ingp 1. Feedforward Neural Networks

l e a r n 2. Recurrent Neural Networks (RNNs)

3. Long Short-Term Memory (LSTM)
Networks
4. Convolutional Neural Networks
(CNNs)
5. Generative Adversarial Networks
(GANs)
6. Transformer Models (BERT, GPT)
 DEEP LEARNING
1. Autoencoders:
Variational Autoencoders (VAEs): Train VAEs to learn a probabilistic representation of normal
data and generate reconstructions. Anomalies are identified based on high reconstruction
errors.
Denoising Autoencoders: Train autoencoders to reconstruct clean data from noisy input.
Anomalies result in larger reconstruction errors due to their deviation from normal patterns.
Sparse Autoencoders: Use autoencoders with sparsity constraints to learn compact
representations of normal data. Outliers or anomalies often fail to be reconstructed accurately.
2. Recurrent Neural Networks (RNNs):
LSTM (Long Short-Term Memory) Networks: Apply LSTMs to model sequential data and detect
anomalies based on deviations from learned temporal patterns.
GRU (Gated Recurrent Unit) Networks: Use GRUs for sequence modeling and anomaly detection in
time-series or sequential data.
3.
 GENERATIVE ADVERSARIAL NETWORKS (GANS):
ANOMALY DETECTION GANS: TRAIN GANS TO GENERATE REALISTIC
SAMPLES OF NORMAL DATA AND DETECT ANOMALIES BASED ON THE
DISCRIMINATOR'S ABILITY TO DISTINGUISH BETWEEN REAL AND
dl GENERATED DATA.
CONDITIONAL GANS: CONDITION GANS ON NORMAL DATA AND
GENERATE SAMPLES THAT DEVIATE SIGNIFICANTLY FROM THE
LEARNED DISTRIBUTION, INDICATING ANOMALIES.
DEEP NEURAL NETWORKS (DNNS):
DEEP FEEDFORWARD NETWORKS: UTILIZE DEEP FEEDFORWARD
NETWORKS (MULTI-LAYER PERCEPTRONS) TO LEARN COMPLEX
MAPPINGS FROM INPUT FEATURES TO ANOMALY SCORES.
DEEP CONVOLUTIONAL NETWORKS (CNNS): APPLY CNNS FOR IMAGE-
BASED ANOMALY DETECTION BY LEARNING HIERARCHICAL
REPRESENTATIONS OF NORMAL VISUAL PATTERNS.
Supervised learning
1. Logistic Regression:
Approach: Train a logistic regression classifier to learn
a decision boundary that separates normal instances
(labeled as 0) from anomalies (labeled as 1).
Usage: Suitable for binary classification tasks where
anomalies are defined based on labeled data.
2. Support Vector Machines (SVM):
Approach: Train an SVM with a nonlinear kernel to
learn a hyperplane that maximizes the margin
between normal and anomalous instances.
Usage: Effective for separating complex patterns and
handling non-linear decision boundaries.
 1. Decision Trees:
Approach: Build decision trees that recursively split the feature space to classify
instances as normal or anomalous based on labeled data.
Usage: Useful for capturing non-linear relationships and interactions between
features.
1. Random Forest:
Approach: Ensemble of decision trees where each tree is trained on a subset of
features and instances. Anomalies can be identified based on outlier scores or
ensemble voting.
Usage: Robust and scalable method for detecting anomalies in high-dimensional
data.
1. Gradient Boosting Machines (GBM):
Approach: Train a gradient boosting model (e.g., XGBoost, LightGBM) to iteratively
improve the classification of normal and anomalous instances based on labeled
data.
Usage: Provides high accuracy and is effective in capturing complex relationships in
the data.
 1. Neural Networks (e.g., Multi-layer Perceptron):
Approach: Train a neural network with multiple layers to learn a mapping from
input features to anomaly labels.
Usage: Deep learning models can automatically learn hierarchical representations
and feature transformations for anomaly detection tasks.
1. Naive Bayes:
Approach: Use a Naive Bayes classifier to estimate the probability of an instance
being normal or anomalous based on labeled data and conditional independence
assumptions.
Usage: Simple and efficient method for probabilistic anomaly detection tasks.
1. Ensemble Methods:
Approach: Combine multiple supervised learning models (e.g., SVMs, decision trees)
using ensemble techniques such as bagging or boosting to improve anomaly
detection performance.
Usage: Enhances robustness and generalization by leveraging the diversity of
individual models.
unsupervised learning
 1. Density-Based Methods:
Density Estimation: Use techniques like Gaussian Mixture Models
(GMM) or Kernel Density Estimation (KDE) to model the
probability distribution of normal data.
Anomaly Detection: Identify instances with low probability
densities (outliers) as anomalies.
1. Clustering Algorithms:
2. K-Means Clustering: Group data points into clusters and
identify instances that do not belong to any cluster or
belong to small clusters as anomalies.
3. DBSCAN (Density-Based Spatial Clustering of Applications
with Noise): Detect outliers as data points that are not
part of any dense region in the feature space.
 1. Autoencoders:
Neural Network-based Approach: Train autoencoder models to
reconstruct normal data and measure reconstruction errors.
Anomaly Detection: Identify instances with high reconstruction
errors as anomalies, indicating deviations from learned normal
patterns.
1. Isolation Forest:
Tree-based Ensemble Method: Use isolation forest algorithms to
isolate anomalies efficiently by randomly partitioning the
feature space.
Anomaly Detection: Measure the average path length in isolation
trees to identify instances that require fewer partitions,
indicating anomalies.
 1. One-Class SVM (Support Vector Machines):
Boundary Learning: Train SVM models on normal data to learn a
boundary that encapsulates the normal behavior.
Anomaly Detection: Classify instances lying outside the learned
boundary as anomalies.
1. Statistical Approaches:
Z-Score or Standard Deviation: Calculate statistical measures
(e.g., Z-Score) to identify instances that deviate significantly
from the mean or standard deviation of normal data.
Anomaly Detection: Set thresholds based on statistical
properties to classify outliers as anomalie