Floating Topic

Learning algorithms

Training set

Test set

Pre-processed dataset

Initial Dataset

Data Cleaning>Data curation>Remove redundant features

Exploratory analysis

Self-Organizing Maps (SOM) SOMs are neural network models that learn to represent high-dimensional data in a lower-dimensional space, typically a two-dimensional grid.

Principal Component Analysis (PCA) PCA aims to reduce the dimensionality of a dataset while preserving the most significant information. It identifies the directions of maximum variance in the data and projects the data onto these principal components.

1. Q-Learning: A model-free algorithm where the agent learns a Q-value function representing the expected cumulative reward for taking an action in a given state. 2. Deep Q Network (DQN): Extends Q-learning by using deep neural networks to approximate the Q-value function, allowing for more complex and high-dimensional state spaces. 3. Policy Gradient Methods: Learn a policy directly, which is a mapping from states to actions, optimizing for the expected cumulative reward. Examples include REINFORCE and Proximal Policy Optimization (PPO). 4. Actor-Critic Methods: Combine aspects of both value-based (critic) and policy-based (actor) methods. The actor suggests actions, and the critic evaluates the suggested actions, providing feedback to update the policy. 5. Deep Deterministic Policy Gradients (DDPG): An off-policy algorithm suitable for continuous action spaces, combining DQN with policy gradients. 6. Monte Carlo Methods: Estimate value functions by sampling sequences of states, actions, and rewards, updating the value function based on the observed returns. 7. Temporal Difference (TD) Learning: Combines elements of Monte Carlo methods and dynamic programming, updating value functions based on the difference between estimated and observed values at each time step. 8. Asynchronous Advantage Actor-Critic (A3C): Utilizes multiple agents running asynchronously to explore and learn more efficiently, and combines actor-critic methods with parallelization. 9. Trust Region Policy Optimization (TRPO): A policy optimization algorithm that aims to improve stability and avoid large policy updates. 10. SARSA (State-Action-Reward-State-Action): Another model-free algorithm similar to Q-learning, but updates the policy based on the current action and the next state's action.

1. Logistic Regression: Suitable for binary classification problems, logistic regression models the probability of an instance belonging to a particular class. 2. Decision Trees: Hierarchical tree structures where each node represents a decision based on input features, leading to the classification of instances. 3. Random Forest: Ensemble method that builds multiple decision trees and combines their predictions to improve accuracy and reduce overfitting. 4. Support Vector Machines (SVM): Constructs a hyperplane to separate instances into different classes, maximizing the margin between classes. 5. K-Nearest Neighbors (KNN): Classifies instances based on the majority class among their k-nearest neighbors in the feature space. 6. Naive Bayes: Based on Bayes' theorem, this probabilistic algorithm assumes independence between features and is effective for text classification. 7. Neural Networks: Deep learning models composed of interconnected layers that learn complex patterns and representations from data. 8. Gradient Boosting: Builds a series of weak learners sequentially, with each one correcting errors made by the previous ones. 9. Linear Discriminant Analysis (LDA): Finds linear combinations of features that best separate classes, making it particularly useful for high-dimensional data. 10. Quadratic Discriminant Analysis (QDA): Similar to LDA but without assuming equal covariance among classes, allowing for more flexibility in modeling.

Kernel Density Estimation (KDE): Utilizes a kernel function to smooth data points and estimate the probability density function. Gaussian Mixture Models (GMM): Represents the distribution as a mixture of multiple Gaussian distributions, each with its own weight, mean, and covariance parameters. Parzen Windows: Similar to KDE but uses fixed windows to estimate density, typically in higher-dimensional spaces. Histogram-based Methods: Divides the data space into bins and estimates density based on the number of points within each bin. Nearest Neighbor Methods: Estimates density at a point based on the number of data points within a specified distance. Autoencoders: Neural network-based approaches that learn a compressed representation of the input data, allowing density estimation in the encoded space. Isolation Forests: A tree-based method that isolates anomalies by recursively partitioning the data. Self-Organizing Maps (SOM): Neural network-based algorithm that produces a low-dimensional representation of input data, which can be used for density estimation.

1. Decision Trees: Tree-like models that make decisions based on input features. 2. Support Vector Machines (SVM): Classifies data points by finding the hyperplane that best separates different classes. 3. K-Nearest Neighbors (KNN): Assigns a class label based on the majority class among its k nearest neighbors. 4. Naive Bayes: Utilizes Bayes' theorem to predict the probability of a sample belonging to a particular class. 5. Logistic Regression: Predicts the probability of an instance belonging to a particular class using a logistic function. 6. Random Forest: Ensemble method that constructs a multitude of decision trees and merges their predictions. 7. Gradient Boosting: Builds a series of weak learners (usually decision trees) and combines their predictions. 8. Neural Networks: Deep learning models composed of interconnected nodes organized in layers for complex pattern recognition.

1. Principal Component Analysis (PCA): Linear method that transforms data into a new coordinate system to capture maximum variance. 2. t-Distributed Stochastic Neighbor Embedding (t-SNE): Non-linear method that preserves local similarities, commonly used for visualizing high-dimensional data in lower dimensions. 3. Linear Discriminant Analysis (LDA): Supervised method that focuses on maximizing the separation between classes while reducing dimensionality. 4. Autoencoders: Neural network-based approach that learns a compressed, distributed representation of the input data. 5. Isomap (Isometric Mapping): Non-linear method that preserves geodesic distances between all points, useful for capturing intrinsic manifold structure. 6. Locally Linear Embedding (LLE): Non-linear method that seeks to preserve local relationships between data points. 7. Random Projection: Projects data into a lower-dimensional space using random matrices while preserving pairwise distances.

1. K-Means Clustering: Divides data into k clusters based on similarity. 2. Hierarchical Clustering: Builds a tree of clusters to represent the hierarchy of relationships. 3. DBSCAN (Density-Based Spatial Clustering of Applications with Noise): Identifies clusters based on dense regions in the data space. 4. Mean Shift: Shifts centroids towards the mean of the data points, iteratively refining cluster assignments. 5. Agglomerative Clustering: Starts with individual data points as clusters and merges them based on similarity. 6. Spectral Clustering: Uses eigenvectors of a similarity matrix to partition data into clusters. 7. Affinity Propagation: Allows data points to "vote" on the most representative exemplar.

<Scikit learn, Tensorflow, Pytorch, GMM>

<PCA, t-SNE, UMAP, LLE, Factor analysis, MATLAB>

<RapidMIne, Weka, Orange, KNIME, Scilit learn, pandas, MLxtend>

TensorFlow, Matplotlib, skikit learn, PyTorch

<K Means clustering, Mean shift clustrering, DBSCAN clustering, Agglomerative, hierachical clustering, Gaussian mixture>

<Linear regression, Neural network regression, Decision tree regression, Lasso regression, ridge regression, ie: >

<Naive bayes classifier, Decision trees, Random forest, SVM, KNN>

Reinforcement learning involves training an agent to make decisions in an environment to maximize rewards. It learns through trial and error, receiving feedback in the form of rewards or penalties. Reinforcement learning has been applied in areas like robotics, game playing, and autonomous vehicles

Unsupervised learning is used when the data is unlabeled. The goal is to find patterns or groupings in the data without any prior knowledge. Clustering algorithms, such as k-means and hierarchical clustering, are commonly used in unsupervised learning.

This technique involves training a model using labeled data to make predictions or classify new data points. Examples include linear regression, decision trees, and support vector machines.

Density estimation

basket analysis

Decision Making

Dimension Reduction

Clustering

Regression

Classification

Model Monitoring

Model Deployment

Train Model

Data Preprocessing and EDA

Gathering Data

Text-to-speech (TTS)

Automatic Speech Recognition or ASR

Template-Based NLG: This approach involves using predefined templates or rules to generate text based on structured data. It is useful for generating simple, repetitive texts with fixed structures, such as weather reports or financial summaries. Rule-Based NLG: Rule-based NLG involves using a set of linguistic rules and templates to generate text. These rules capture syntactic and semantic patterns and allow for more flexibility in generating varied and contextually appropriate output. It is commonly used in chatbots and automated report generation. Statistical NLG: Statistical NLG utilizes machine learning algorithms and statistical models to generate text. These models learn patterns and generate text based on probabilities derived from large datasets. It is often used for tasks like text summarization, machine translation, and content generation. Neural NLG: Neural NLG employs deep learning techniques, particularly recurrent neural networks (RNNs) and transformers, to generate text. These models learn from large amounts of data and can capture complex patterns and generate more fluent and coherent text. Neural NLG is widely used in applications like language translation, chatbots, and text generation for creative purposes. Hybrid Approaches: Some NLG systems combine multiple approaches, leveraging the strengths of each. For example, a system may use rule-based techniques for controlling the structure and content of the generated text, while employing statistical or neural models to generate more fluent and natural-sounding language.

Text Preprocessing: This involves tasks such as tokenization (splitting text into individual words or tokens), stemming (reducing words to their base form), and part-of-speech tagging (assigning grammatical tags to words). Named Entity Recognition (NER): NER identifies and classifies named entities in text, such as names of people, organizations, locations, dates, and other specific entities. Sentiment Analysis: This component determines the sentiment or emotional tone expressed in a piece of text, categorizing it as positive, negative, or neutral. Intent Recognition: Intent recognition aims to understand the purpose or intention behind a user's input. It involves identifying the user's goal or desired action from their text or speech. Entity Resolution: Entity resolution resolves references to entities in text by linking them to their corresponding real-world entities. For example, linking pronouns like "he" or "it" to the actual person or object they refer to. Language Modeling: Language modeling focuses on understanding the structure and context of language. It involves tasks such as grammar analysis, syntactic parsing, and semantic role labeling. Dialogue Management: Dialogue management deals with maintaining a conversational context and managing the flow of conversation between a user and a system. It includes tasks like turn-taking, maintaining state, and generating appropriate responses.

focuses on the generation of human-like natural language text or speech by machines.

subfield of Natural Language Processing (NLP) that focuses on the comprehension and interpretation of human language by machines.

Natural Language Generation (NLG)

Natural Language Understanding (NLU)

The machine reply with the audio file

Data to audio conversion occurs

The machine processes the data

Audio - text conversion takes place

The machine capture the audio

Human Talk to machine

Natural Language Processing, is like teaching computers to understand and use human language. It helps machines read, interpret, and respond to what people say or write, making communication between humans and computers more natural.

Feature extraction - Feature extraction in machine learning is typically performed by collaboration of data scientists, machine learning engineers, or individuals with expertise in the field of data analysis and Algorithms

Feature extraction+Classification - the models themselves are capable of learning intricate features automatically during the training process

A specialized form of machine learning using neural networks with many layers (deep neural networks) to analyze and process complex data. (advanced)

An essential subset of AI that involves statisticical analysis, data science and the development of algorithms allowing systems to learn from data and improve performance over time. (traditional)

Reinforcement Learning: .

Unsupervised Learning

Supervised Learning:

Deep learning

machine learning

NLP