Difference between revisions of "PRIMO.ai"
m (→Lean Forward Thinking) |
(→Unsupervised: Probabilistic/Generative) |
||
| Line 56: | Line 56: | ||
==== Unsupervised: Probabilistic/Generative ==== | ==== Unsupervised: Probabilistic/Generative ==== | ||
A probability distribution over a set of classes for each input sample. Unlabeled data is classified as (1) conditional probability of the target Y, or (2) conditional probability of the observable X given a target Y | A probability distribution over a set of classes for each input sample. Unlabeled data is classified as (1) conditional probability of the target Y, or (2) conditional probability of the observable X given a target Y | ||
| + | *[[K-Means]] | ||
*[[Restricted Boltzmann Machine (RBM)]] | *[[Restricted Boltzmann Machine (RBM)]] | ||
*[[Deep Belief Network (DBN)]] | *[[Deep Belief Network (DBN)]] | ||
*[[Variational Autoencoder (VAE)]] | *[[Variational Autoencoder (VAE)]] | ||
| + | |||
==== Competitive ==== | ==== Competitive ==== | ||
*[[Generative Adversarial Network (GAN)]] | *[[Generative Adversarial Network (GAN)]] | ||
Revision as of 11:38, 27 May 2018
Contents
Overview
Background
AI Breakthroughs
AI Fun
How to...
Forward Thinking
Models
Supervised
Labeled (desired solution) data is fed into the algorithm
- Support Vector Machine (SVM)
- Hopfield Network (HN)
- Energy-based Model (EBN)
- Naive Bayes
- Markov Model (Chain, Discrete Time, Continuous Tme, Hidden)
- Perceptron (P)
- Feed Forward Neural Network (FF or FFNN)
- Artificial Neural Network (ANN)
- Deep Neural Network (DNN)
Convolutional
Deonvolutional
Sequence
- Long Short-Term Memory (LSTM), Gated Recurrent Unit (GRU), and Recurrent Neural Network (RNN)
- Attention Model
- Sequence to Sequence (Seq2Seq)
- (Tree) Recursive Neural (Tensor) Network (RNTN)
- Neural Turing Machine
Unsupervised
Some uses of Unsupervised Learning are (1) data compression, (2) classification, (3) clustering, and (4) outlier detection
Unsupervised: Probabilistic/Generative
A probability distribution over a set of classes for each input sample. Unlabeled data is classified as (1) conditional probability of the target Y, or (2) conditional probability of the observable X given a target Y
Competitive
Unsupervised: Non-Probabilistic
Unlabeled data is fed into the algorithm with the algorithm seperating the feature space and return the class associated with the space where a sample originates from.
Reinforcement
An 'agent' algorithm receives a delayed reward in the next time step to evaluate its previous action; in combination with Neural Networks it is capable of solving more complex tasks
- Markov Decision Process (MDP)
- Deep Reinforcement Learning (DRL)
- Deep Q Learning (DQN)
- Neural Coreference
- State-Action-Reward-State-Action (SARSA)
- Deep Deterministic Policy Gradient (DDPG)
- Trust Region Policy Optimization (TRPO)
- Proximal Policy Optimization (PPO)
Hierarchical
Frameworks
TensorFlow
- TensorFlow Overview & Tutorials
- TensorFlow.js
- TensorBoard
- TensorFlow Playground
- TensorFlow Serving
- Related...
Techniques
Mathematical Background
Datasets & Information Analysis
Algorithms
Bag of Tricks
- Activation Functions
- Optimizers
- Pooling
- Hyperparameters
- Visualization
- Transfer Learning
- Competitions
Coding
Platforms: Machine Learning as a Service (MLaaS)
Amazon AWS
- AWS with TensorFlow
- AmazonML
- Deep Learning Amazon Machine Image (DLAMI)
- DeepLens - deep learning enabled video camera
Google Cloud AI
- Google Cloud AI With TensorFlow
- ML Engine
- Prediction API
- Google Developers Codelabs
- Google AI Experiments
- Cloud Vision API - drag & drop picture on webpage
Kaggle
Microsoft Azure
Research & Development
- Self Learning Artificial Intelligence - AutoML
- Explainable Artificial Intelligence
- Differentiable Neural Computer (DNC)
- Genetic Algorithms
- Natural Language Inference (NLI) and Recognizing Textual Entailment (RTE)
- 3D Simulation Environments
- Connecting Brains
- Architectures
- Other Challenges
