Components

Sumary: GUT-AI Initiative is subdivived into a series of components (i.e. subprojects) in order to build a user-friendly Open-Data, Open-Source, Decentralized ecosystem with support from GUT-AI Foundation). This is a list of all the components of this Ecosystem.

It is important to note that each component definition intentionally does not include how to be implemented, but only what to be implemented. The reason is that there should be no constraints or limits on the 'how' since new advances in Technology can potentially bring new opportunities to improve the 'how' a specific component is implemented. The 'why' each component is necessary is explained in the Summary above and also in the Vision and Mission of GUT-AI. The aim of GUT-AI initiative is not to reinvent the wheel. If a tool or solution for something already exits, then an integration can be created for that tool or solution.

Table of Contents

List of component identifiers
Meta Component
Five-Layer Architecture
Layer 1 components
- Component C1.1: Distributed Smart Grids
- Component C1.2: GUT-AI DCP
Layer 2 components
Layer 3 components
Layer 4 components
Layer 5 components

List of component identifiers

See identifiers for the list of identifiers of all components.

Meta Component

The Meta component is a component about all the other components. Its repository is the current one.

Five-Layer Architecture

The following Five-Layer Architecture is used for the GUT-AI Protocol:

Layer 1 - Foundation Layer (Physical and Data Link)
Layer 2 - Infrastructure Layer (IaaS)
Layer 3 - Platform Layer (PaaS)
Layer 4 - Intelligence Layer (AIaaS)
Layer 5 - Innovation Layer (SaaS)

Abbreviations:
IaaS = Infrastructure as a Service
PaaS = Platform as a Service
AIaaS = AI as a Service
SaaS = Software as a Service
FD = Fully Decentralized
SD = Semi-Decentralized

Layer 1 components

Component C1.1: Distributed Smart Grids

Description: Bring Distributed Smart Grids into production in real life through Blockchain and AI solutions (GUT-AI DCP) powered by energy storage. Also use AI to improve Distributed Smart Grids.

Aims:

No hierarchical, centralized (electricity or communication) authority for the residential and commercial consumers (i.e. non-industrial)
Use of grid-connected microgrid for both electricity and communication
Use of specific hardware and devices
- Distributed Energy Sources (e.g. photovoltaic panels)
- Distributed Electricity and Energy Storage (e.g. batteries)
- Smart Grid connectivity equipment (e.g. physical servers, dish antenna)
Use of stored energy for the demands of the GUT-AI DCP through in-house physical servers in order to reduce waste of the generated energy
Use of Communication for proactive and online diagnosis of transient faults and prognosis of potential blackouts
Use of Real-Time Pricing through a Decentralized Exchange (DEX) for power markets
Distributed Computer Network for Communication, DSM and Real-Time Pricing
Support for GUT-AI DCP and other decentralized cloud providers
Support for Internet of Things (IoT)
Support for interoperable electric vehicles
Support for conventional (dieasel and petrol) vehicles
Support for Near-Zero Energy Buildings (NZEBs)
Support for Aeroponics, Hydroponics and Aquaponics for near-zero energy farming
Support for conventional agronomics and livestock farming
Maximum freedom and liberty to each household on how to operate their own household as part of the whole ecosystem

Component C1.2: GUT-AI DCP

Description: Create a dedicated Decentralized Cloud Proivder (DCP) related to GUT-AI for the information storage needs. Also use AI to improve DCP.

Aims:

No hierarchical, centralized authority (i.e. similar to blockchain)
Hosting
Databases (SQL and NoSQL)
Data Warehouses
Data Lakes
Anything else that a conventional, centralized Cloud Provider can offer

Layer 2 components

Component C2.1: GUT-AI Marketplace

Description: Create a dedicated Marketplace for products (data, software apps) and services (Contractors and Freelancers) related to GUT-AI. Each digital product will be a module, which will be interoperable and integrable with any other module (just like pieces of a puzzle or building blocks).

Aims:

Open Data (e.g. datasets, pre-trained models) as modules
Proprietary Data (e.g. datasets, pre-trained models) as modules
Centralized and decentralized SaaS modules developed by third parties
Centralized and decentralized PaaS modules developed by third parties
Centralized and decentralized IaaS modules developed by third parties
Marketplace for marketplaces by third parties for physical products (e.g. computers, physical servers, robots, photovoltaic panels)
Contractors and Freelancers (e.g. Data Scientists, Data Engineers, Machine Learning Engineers, Blockchain Developers)
Decentralized Exchange (DEX)
Support for conventional (credit and debit card) payments
Support for crypto payments

Component C2.2: Automated Data Preparation

Description: Perform Automated Data Preparation using AI.

Aims:

Data Collection
Data Synthesis / Data Simulation / Adversarial Learning
Data Fusion and Data Integration
Data Wrangling / Data Munging
Data Scraping
Data Sampling
Data Cleaning

Component C2.3: CI/CD

Description: Perform Continuous Integreation/Continuous Delivery (CI/CD) for all ML systems and also all associated systems. Also use AI to improve CI/CD (AIOps).

Aims:

Reproducibility
Replicability
Code Version Control
Data Version Control (for both datasets and pretrained models)
Automatic Configurations (with default, but adjustable values)
Machine Resource Management
Governance and Regulatory Compliance (e.g. GDPR, HIPAA, ISOs)
Monitoring and Reporting
Diagnostics
Testing and Quality Assurance (for both code and data)
User of containers (e.g. Docker)
User of orchestration (e.g. Kubernetes)
Use of microservices
Support for embedded devices and IoT devices
Support for Tensor Computation Libraries (e.g. TensorFlow, PyTorch, MXNet, JAX)
Support for Asynchronous Communication (e.g. ActiveMQ, RabbitMQ, Apache Kafka)
Support for Synchronous Communication (e.g. REST, GraphQL)
Support for Databases (SQL and NoSQL), Data Warehouses and Data Lakes
Support for Data Workflow Management (e.g. Airflow, Kubeflow, MLflow)
Support for High-Performance Model Serving (e.g. KServe, Seldon Core, BentoML)
Direct integration to Top 10 centralized IaaS cloud providers
Direct integration to Top 10 decentralized IaaS cloud providers
Direct integration to GUT-AI Marketplace and other marketplaces
Webhooks and API for direct integration to IaaS, PaaS, SaaS providers
Automation, MLOps, DataOps, MoodelOps, DevOps
Information Security, SecDevOps, DevSecOps
Anything else reducing the technical debt

Component C2.4: DX

Description: Enhance Developer Experience (DX) to make it developer-friendly for almost anyone who can write code at any level.

Aims:

Separation of concerns
Publication of optional standards and good practise guidelines
User-friendly User Interface (UI) and Dashboards
User-friendly configurations (e.g. using yaml and json)
Anything else reduing the cultural debt or improving the DX

Layer 3 components

Component C3.1: AutoDS

Description: Perform Automated Data Science (AutoDS) by combining (internal or external) modules together in an adjustable way.

Aims:

Automated Data Preparation
AutoML
Continual Learning
CI/CD
Distributed Systems for ML
Sandbox for experiments

Component C3.2: AutoML

Description: Perform Automated Machine Learning (AutoML).

Aims:

Automated Data Preprocessing
NAS
Hyperparameter Optimization (HPO)

Component C3.3: Automated Data Preprocessing

Description: Perform Automated Data Preprocessing.

Aims:

Automated Feature Selection
Automated Feature Extraction
- Rule-based AI
- Representation Learning (Supervised, Unsupervised, Self-Supervised)
  
  Data Augmentation / Contrastive Learning
  
  Feature Construction / Generative Learning
  
  Adversarial Learning
Representation disentanglement
Representation Transfer
Multimodal Representation Learning
Self-Supervised Learning (for efficient RL Downstream Tasks)

Component C3.4: NAS

Description: Perform Neural Architecture Search (NAS).

Aims:

Automated Model Selection
- Search space
- Architecture Optimization
Automated Model Estimation

Component C3.5: Continual Learning

Description: Perform Continual Learning.

Aims:

Automated Model Retraining
Intra-Agent Transfer Learning in RL
Causal Learning (to address the Moravec's Paradox)
Memory Bottleneck
Meta Learning
Multitask Learning
Transfer Learning
Few-Shot Learning
Zero-Shot Learning
Continual AutoML

Component C3.6: Distributed Systems for ML

Description: Introduce and perform Distributed Systems that are model-specific for ML and especially for Gradient-Based Optimization methods.

Aims:

Support for generic Distributed Systems (e.g. Horovod, DeepSpeed)
Devise new ML-specific architectures (similar to Petuum V2)
Large-Scale ML

Component C3.7: Memory Bottleneck

Description: Solve the issue of memory bottleneck in order to enable the Inference of Deep Learning models in embedded devices (while also addressing Moravec's Paradox).

Aims:

Model Compression and Weight Sharing
Nodes Pruning and Weight Pruning
Product Quantization (or Vevtor Quantization)
Precision Quantization (or Scalar Quantization)
Huffman Coding
Representation disentanglement on the sparse weight matrix
Structured Sparsity Learning (StSL)
Soft-Weight Sharing
Variational Dropout
Structured Bayesian Pruning
Knowledge distillation
Bayesian Compression
Lottery Ticket Hypothesis
NAS
Start with no connections, and add complexity as needed (e.g. Weighted Agnostic Neural Networks)
Weighted Linear Finite-State Machines (WLFSM)
Bayesian Neural Networks (BNNs)
Automated extraction of compressed knowledge
Automated Indexing, Caching and Searching (of compressed knowledge)
Compressed Feature Extraction (i.e. Compression of Representation Learning Models)
Competitive Learning

Component C3.8: Multi-tool CMS

Description: Multi-tool Content Management System (CMS) for anything external to the company.

Aims:

Headless CMS with independent web front-ends
E-Commerce functionality
Payments functionality
Website builder using drag-and-drop elements
Privacy-aware (GDPR-compliant)
Plugins

Component C3.9: Multi-tool CRM

Description: Multi-tool Customer Relationaship Management (CRM) software for anything internal to the company.

Aims:

Headless CRM with independent web front-ends
Client database & Contact Management
Payments functionality
Quotation, Invoicing, Billing & Accounting
Marketing Automation & Lead Management
Workflow Automation
Customer Support & Helpdesk
Enterprise Management System (ERP)
Inventory
Contract Management
Privacy-aware (GDPR-compliant)
Plugins

Component C3.10: Multi-tool Task Management

Description: Multi-tool Task Management software for building software products.

Aims:

Task Creation and Assignment
Time Tracking and Progress Monitoring
Bug Tracking and Issue Management
Integration with Version Control
Release Planning and Management
Agile boards (Scrum and Kanban)
Roadmaps
Plugins

Component C3.11: NOS

Description: Neural Operating System (NOS) for deploying Deep Learning into production.

Aims:

Integrations with CMS, CRM and Task Management software
Interoperability with main Linux distributions (e.g., Ubuntu, Debian, CentOS)

Layer 4 components

Component C4.1: Automated Scientific Discovery

Description: Perform Automated Scientific Discovery.

Aims:

AutoML
Automated Scientific Discovery using model-based Reinforcement Learning
Automated Scientific Discovery using model-free Reinforcement Learning
Automated Scientific Discovery using Dynamical Systems
Representation disentanglement to find neural state variables
Automated extraction of compressed knowledge
Automated extraction of 'learnable' rules (i.e. 'oscillatory' determinism) in accordance with GUT and TLKA theory
Causal Learning (to address the Moravec's Paradox)
Representation disentanglement
Explainable AI (XAI)
- Counterfactuals
- Factuals

Component C4.2: MTSU

Description: Perform Multitask Scence Understanding (MTSU) by applying Multitak Learning on Computer Visions tasks on a still and immobile camera.

Aims:

Object Detection
Object Recognition
Face Recognition
Image Segmentation (Semantic and Instance)
Image Captioning and Image Categorization
Visual Relationship Detection
Action Classification
Activity Recognition
Pose Estimation
Super-Resolution
Denoising
Image Acquisition and Reconstruction
Image Restoration
Image Generation
Image Registration
Domain Adaptation
Multi-Object Motion Detection and Tracking
Vision-Based Motion Analysis
Vision as Inverse Graphics
Image Synthesis
Video Synthesis

Component C4.3: Grounded CV

Description: Perform Grounded Computer Vision (Grounded CV) by applying Grounded Cognition on Computer Visions tasks on multiple mobile robots or multiple aerial robots (drones) or a combination of them (but using only a single modality, i.e. images or video).

Aims:

MTSU
Simultaneous Localization and Mapping (SLAM).
3D Scene Reconstruction
Surface Reconstruction
Structure from Motion
Feature Matching
Active Tracking
Exploration
Navigation

Component C4.4: ASR

Description: Perform Automatic Speech Recognition (ASR).

Aims:

End-to-End ASR
ASR as Inverse TTS

Component C4.5: TTS

Description: Perform Text-to-Speech (TTS).

Aims:

End-to-End TTS
Multimodal TTS

Component C4.6: SER

Description: Perform Speech Emotion Recognition (SER).

Aims:

Perform Unsupervised Learning to learn a hierarchical model about the number of emotions
Representation disentanglement of lingustic (lexical) and paralinguistic (non-lexical) features
End-to-End SER

Component C4.7: MT

Description: Perform Machine Translation (MT) using Multitask Learning for various languages.

Aims:

End-to-End MT

Component C4.8: TOD

Description: Perform Task-Oriented Dialogue (TOD) using Multitak Learning.

Aims:

Natural Language Understanding (NLU)
- Named-Entity Recognition / Entities Extraction
- Intent Classification / Intent Detection
Dialogue Manager
Natural Language Generation (NLG)

Component C4.9: QA

Description: Perform open-domain Question-Answering (QA), aka Non-Task-Oriented Dialogue.

Aims:

ML-based QA (Corpus-based or Image-based)
- Retrieval-based models (using Utterance selection)
- Generative models
QA as Inverse Question Generation

Component C4.10: VSPT

Description: Perform Visuo-spatial Perpsective-Taking (VSPT).

Aims:

Level 1 (L1) VSPT
Level 2 (L2) VSPT

Component C4.11: Multi-Agent Communication

Description: Implement Multi-Agent Communication.

Aims:

Communication among agents in Deep RL
Interpretation of emergent communication (among heterogenous or homogeneous agents)
Body language
Sign language
Inter-Agent Transfer Learning in RL
- Inverse Reinforcement Learning (IRL)
- Imitation Learning
- Learning from Demonstrations

Component C4.12: Multi-Robot Path Planning

Description: Perform Multi-Robot Path Planning (i.e. visuo-motor abilities).

Aims:

MTSU
Object-Goal Navigation
Collision Avoidance
Path Planning
Multi-Agent Communication

Component C4.13: Multi-Robot Target Detection and Tracking

Description: Perform Multi-Robot Target Detection and Tracking.

Aims:

Target Detection
Target Tracking
Multi-Agent Communication

Component C4.14: Anomaly Detection

Description: Perform Anomaly Detection.

Aims:

Anomaly Detection

Component C4.15: Recommender Engines

Description: Implement Recommender Engines.

Aims:

Recommender Engines

Component C4.16: Grounded QA

Description: Perform Grounded Question-Answering (Grounded QA) by applying Grounded Cognition on QA tasks on multiple mobile robots or multiple aerial robots (drones) or a combination of them using Multimodal Learning (i.e. visuo-linguistic abilities).

Aims:

QA
VSPT
Multi-Agent Communication
Multi-Robot Path Planning

Component C4.17: Grounded NLP

Description: Perform Grounded Natural Language Processing (Grounded NLP) by applying Grounded Cognition on NLP tasks on multiple mobile robots or multiple aerial robots (drones) or a combination of them using Multimodal Learning.