Skip to content

Latest commit

 

History

History
 
 

Folders and files

NameName
Last commit message
Last commit date

parent directory

..
 
 

README.md

Components

Sumary: The GUT-AI initiative is subdivived into a series of components in order to build a user-friendly Open-Data, Open-Source, Decentralized ecosystem under the umbrella of the GUT-AI Foundation. This is a list of all the components of such an ecosystem.

It i 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 above and also in the Vision and Mission of GUT-AI.

Table of Contents

Layer 1 components

Component C1.1: Distributed Smart Grids

Description: Create Distributed Smart Grids for the energy storage needs of the Blockchain and AI era.

Aims:

  • Smart Grids
  • Distributed Renewables (e.g. PV)
  • Decentralized Electricity and Energy Storage (e.g. batteries)
  • Distributed Computer Network for Communication
  • Support for GUT-AI DCP and other decentralized cloud providers
  • Support for interoperable electric vehicles
  • Support for conventional (dieasel and petrol) vehicles
  • Support for Near-Zero Energy Buildings (NZEBs)

Component C1.2: GUT-AI DCP

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

Aims:

  • Hosting
  • Databases (SQL and NoSQL)
  • Data Warehouses
  • Data Lakes
  • Anything else that a conventional, centralized Cloud Provider can offer

Component C1.3: 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.

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, servers, robots)
  • Contractors and Freelancers (e.g. Data Scientists, Data Engineers, Machine Learning Engineers, Blockchain Developers)

Component C1.4: 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 C1.5: 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 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 Model Serving (e.g. KFServing, 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 C1.6: 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
  • 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 2 components

Component C2.1: AutoDS

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

Aims:

Component C2.2: AutoML

Description: Perform Automated Machine Learning (AutoML).

Aims:

Component C2.3: 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

Component C2.4: NAS

Description: Perform Neural Architecture Search (NAS).

Aims:

  • Automated Model Selection
    • Search space
    • Architecture Optimization
    • Hyperparameter Optimization
  • Automated Model Estimation

Component C2.5: Continual Learning

Description: Perform Continual Learning.

Aims:

  • Automated Model Retraining

Component C2.6: 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)

Component C2.7: Solve memory bottleneck

Description: Solve the issue of memory bottleneck in order to enable the Inference of Deep Learning models in embedded devices.

Aims:

  • Model Compression and Weight Sharing
  • Nodes Pruning and Weight Pruning
  • Quantized Training
  • Huffman Coding
  • Representation disentanglement on the sparse weight matrix
  • Structured Sparsity Learning (StSL)
  • Soft-Weight Sharing
  • Variational Dropout
  • Structured Bayesian Pruning
  • Bayesian Compression
  • Lottery Ticket Hypothesis
  • NAS
  • Start with no connections, and add complexity as needed (e.g. Weighted Agnostic Neural Networks)
  • Bayesian Neural Networks (BNNs)

Layer 3 components

Component C3.1: Automated Scientific Discovery

Description: Perform Automated Scientific Discovery.

Aims: TODO

Component C3.2: MTSU

Description: Perform Multitask Scence Understanding (MTSU) by applying Multitak Learning on Computer Visions tasks from 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

Component C3.3: Grounded CV

Description: Perform Grounded Computer Vision (Grounded CV) by applying Grounded Cognition on Computer Visions tasks from a single mobile robot or a single aerial robot (drone).

Aims:

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

Component C3.4: ASR

Description: Perform Automatic Speech Recognition (ASR).

Aims: TODO

Component C3.5: TTS

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

Aims: TODO

Component C3.6: SER

Description: Perform Speech Emotion Recognition (SER).

Aims: TODO

Component C3.7: MT

Description: Perform Machine Translation (MT).

Aims: TODO

Component C3.8: TOD

Description: Perform Task-Oriented Dialog (TOD).

Aims: TODO

Component C3.9: QA

Description: Perform open-domain Question-Answering (QA).

Aims: TODO

Component C3.10: Grounded QA

Description: Perform Grounded Question-Answering (Grounded QA).

Aims: TODO

Component C3.11: VSPT

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

Aims: TODO

Component C3.12: Multi-Robot Path Planning

Description: Perform Multi-Robot Path Planning.

Aims: TODO

Component C3.13: Multi-Robot Target Detection and Tracking

Description: Perform Multi-Robot Target Detection and Tracking.

Aims: TODO

Component C3.15: Anomaly Detection

Description: Perform Anomaly Detection.

Aims: TODO

Component C3.16: Recommender Engines

Description: Implement Recommender Engines.

Aims: TODO

Layer 4 components

Component C4.1: Automated Protoyping

Description: Perform Automated Protoyping.

Aims:

  • Ideation and Creation

Component C4.2: Automated UX

Description: Perform Automated User Experience (Automated UX) during Product Discovery and Product Development.

Aims:

  • Automated User Research
  • Automated User Validation
  • Automated UX Research

Component C4.3: Automated Marketing

Description: Perform Automated Marketing.

Aims: TODO

Component C4.4: Automated Sales

Description: Perform Automated Sales.

Aims: TODO

Component C4.5: Automated Customer Support

Description: Perform Customer Support.

Aims: TODO

Component C4.6: Automated Governance and Compliance

Description: Perform Automated Governance and Compliance for the Blockchain and AI era.

Aims: TODO

Component C4.7: Portfolio Management

Description: Perform Portfolio Management for the Blockchain and AI era.

Aims: TODO