[FRIAM] Needed: AI taxonomy

[Steve Smith]

On 4/3/23 8:33 PM, Tom Johnson wrote:
> Nicholas Diakopoulos (@ndiakopoulos) tweeted at 2:57 PM on Mon, Apr 
> 03, 2023:
> Given the proliferation of AI models being released these days, do we 
> need a sort of Linnean system (i.e. kingdom, phylum, class, order, 
> family, genus, species) to classify and make sense of the space?
> (https://twitter.com/ndiakopoulos/status/1642994632404369408?t=Mt6Jj5Y5lsMRjbrjrnAosA&s=03 
> <https://twitter.com/ndiakopoulos/status/1642994632404369408?t=Mt6Jj5Y5lsMRjbrjrnAosA&s=03>)
while this may not be as helpful as one would hope, I dove straight into 
the horse's mouth to poke around for hints:
(this is definitely more complete and cogent and probably accurate than 
if I'd winged it based on my own flawed conception)

GPT4:

    there are multiple ways of classifying AI systems. The
    classification schemes can differ based on various factors such as
    the application domain, the learning approach, the level of
    autonomy, and the cognitive architecture.

    For example, based on the application domain, AI systems can be
    classified as healthcare AI systems, finance AI systems, or
    autonomous vehicle AI systems. Based on the learning approach, AI
    systems can be classified as supervised learning systems,
    unsupervised learning systems, or reinforcement learning systems.
    Based on the level of autonomy, AI systems can be classified as
    human-in-the-loop systems, semi-autonomous systems, or fully
    autonomous systems. Based on the cognitive architecture, AI systems
    can be classified as symbolic AI systems, connectionist AI systems,
    or hybrid AI systems.

    Each classification scheme serves a specific purpose and provides a
    different perspective on AI systems. It is important to choose the
    classification scheme that best fits the context of the problem
    being addressed.

and (an elaborated breakdown by "learning approach"). A "proper 
ontology" might involve the intersection/superposition of the 4 
(nominally?) hierarchical 1) Application Domain; 2) Learning Approach; 
3) Level of Autonomy; 4) cognitive architecture (and others yet to be 
considered?). Each of these "cuts" through the space would be what I 
call a "facet" in a "faceted ontology".

While I'm familiar with many of the concepts and terms implied below, I 
can't claim to begin to have my own mental/intuitive arms wrapped around 
the space really... folks here with more grounded knowledge/experience 
might be able untangle this in a more useful/interesting stylization?

    A "learning method" classification scheme (more cladistic than
    linnaean?)

     1.

        *Supervised Learning Models:*

          * Regression models
              o Linear regression
              o Polynomial regression
              o Ridge regression
              o Lasso regression
              o Elastic Net
          * Classification models
              o Logistic regression
              o Decision tree
              o Random forest
              o Naive Bayes
              o K-Nearest Neighbors (KNN)
              o Support Vector Machine (SVM)
              o Neural Networks (including Convolutional NN, Recurrent
                NN, and Transformer)
          * Time series forecasting models
              o Autoregressive Integrated Moving Average (ARIMA)
              o Exponential Smoothing (ETS)
              o Prophet
              o Long Short-Term Memory (LSTM)
     2.

        *Unsupervised Learning Models:*

          * Clustering models
              o K-Means
              o Hierarchical Clustering
              o Density-Based Spatial Clustering of Applications with
                Noise (DBSCAN)
              o Gaussian Mixture Model (GMM)
          * Dimensionality Reduction models
              o Principal Component Analysis (PCA)
              o t-Distributed Stochastic Neighbor Embedding (t-SNE)
              o Linear Discriminant Analysis (LDA)
              o Independent Component Analysis (ICA)
          * Association Rule Learning models
              o Apriori Algorithm
              o Eclat Algorithm
     3.

        *Reinforcement Learning Models:*

          * Value-Based models
              o Q-Learning
              o Deep Q-Network (DQN)
              o Double Deep Q-Network (DDQN)
          * Policy-Based models
              o Policy Gradient
              o Trust Region Policy Optimization (TRPO)
              o Proximal Policy Optimization (PPO)
          * Actor-Critic models
              o Asynchronous Advantage Actor-Critic (A3C)
              o Deep Deterministic Policy Gradient (DDPG)
              o Twin Delayed DDPG (TD3)
              o Soft Actor-Critic (SAC)
     4.

        *Generative Models:*

          * Variational Autoencoders (VAE)
          * Generative Adversarial Networks (GAN)
              o Deep Convolutional GAN (DCGAN)
              o Conditional GAN (CGAN)
              o CycleGAN
          * Auto-Regression models
              o Autoregressive (AR) model
              o Auto-Regressive Moving Average (ARMA) model
              o Auto-Regressive Integrated Moving Average (ARIMA) model
     5.

        *Transfer Learning Models:*

          * Fine-Tuning models
              o BERT
              o GPT
              o ResNet
              o Inception
          * Multi-Task Learning models
              o Deep Neural Decision Forests (DNDF)
              o Multi-Task Neural Networks (MTNN)
          * Domain Adaptation models
              o Adversarial Domain Adaptation
              o Domain-Adversarial Neural Network (DANN)
              o Deep Adaptation Networks (DAN)
     6.

        *Hybrid Models:*

          * Recommender Systems models
              o Collaborative Filtering + Content-Based Filtering
              o Collaborative Filtering + Deep Learning
              o Content-Based Filtering + Deep Learning
              o Hybrid Matrix Factorization
          * Vision-Language models
              o Visual Question Answering (VQA)
              o Image Captioning
              o Visual Dialog
          * Natural Language Processing (NLP) models
              o Text Classification + Named Entity Recognition (NER)
              o Text Classification + Sentiment Analysis
              o Text Classification + Topic Modeling


-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://redfish.com/pipermail/friam_redfish.com/attachments/20230404/eeaf3ce0/attachment-0001.html>