kenricnelson

Project owner

Risk-aware Assessments of AI Algorithms

Photrek will provide risk-aware assessments of ML/AI algorithms. The assessments will use a generalization of information theory that incorporates a measure of risk tolerance. The SingularityNET community will benefit from the service by having access to tools that enhance SingularityNET applications and by the revenue generated by other users.

H+DAO Transhumanist Decentralized Projects

A group of individuals with long involvement in the transhumanist community, and with advocacy of the nonprofit Humanity+, are coming together to create H+DAO. Operating as a distinct organization from Humanity+, but in close partnership with it, H+DAO is a new decentralized autonomous organization. It possesses an innovative structure yet is established with goals strongly akin to Humanity+: To support the ethical and appropriate use of technology to better the human condition.

Universal Sentence Parser

I want to create a service that splits a text into sentences, according to a given dictionary, in a fast and robust way. It can be applied to natural languages and artificial ones.

MUSEVERSE: Music into the metaverse

Play, jam, and perform music in the metaverse. Our vision is to build AI tools for composing and performing music in the backend on SingularityNET.We want to create an AI for analyzing musical instruments from physical space and projecting them into virtual reality. This provides an immersive experience of bridging reality with virtual reality in music production. As the first musical instrument, we have chosen a piano keyboard. Once we have this technology ready, we want to integrate it into these metaverses. 

• Somnium Space
• Victoria VR
• Decentraland
• Sandbox 
• SophiaVerse

Robert Haas

Project owner

Evolutionary search for robust trading strategies

This project uses grammar-guided genetic programming to automatically search for robust and creative trading strategies on user-defined markets, indicators, metrics and external timeseries data from predictive models, fundamental analysis or elsewhere. The main challenge in the design of trading strategies is not the automated search process, although an advanced one will be used here, but rather the design and implemention of a proper backtesting approach to evaluate candidate strategies that avoids common pitfalls such as overfitting to the provided data or optimizing the wrong goal. This challenge will be tackled by in-depth literature research about state-of-the-art statistical evaluation techniques and review of existing backtesting libraries. Consequently, the search itself will be tuned by initially developing simple and then increasingly advanced grammars that define different search spaces of candidate trading strategies.

Both 1) the evaluation method for candidate strategies and 2) the grammars for defining suitable spaces of candidate strategies will be integrated into an open-source software package and distributed via GitHub and PyPI. Based on this package, a service will be developed for SingularityNET's AI platform, where users can provide data plus various preferences to the algorithm and it will return evolved trading strategies that perform well on historical data and have a reasonably high chance to actually perform similarly well on live markets. These strategies can then be used either for manual trading or as part of an automated trading system, which involves further aspects such as portfolio management and position sizing for multiple strategies. When the markets undergo a regime change after some time, the strategies can easily be adapted by making use of the service again.

Robert Haas

Project owner

Grammar-guided genetic programming

Grammar-guided genetic programming (GGGP, G3P) is a set of advanced optimization methods from the field of evolutionary computation (EC). These methods allow to search for optimal programs in any formal language that can be defined with a context-free grammar. This includes most general-purpose programming languages (e.g. Python, C++, Rust, OpenCog's Atomese), domain-specific languages (e.g. HTML, JSON, protobuf) and user-written mini-languages (e.g. geometric shape configurations).

The aim of this proposal is to implement five different G3P algorithms (CFG-GP, GE, piGE, DSGE, WHGE) that represent the state-of-the-art, integrate them in a single open-source software package, create a collection of benchmark problems from literature to compare them rigorously, and then bring the best performing algorithm as general-purpose optimization service to the SingularityNET platform. A proof-of-concept prototype with two algorithms (CFG-GP, GE) has already been created and successfully applied to solve example problems from different domains, which is shown in the attached supplementary material to demonstrate the feasibility of the project. The overall utility of these methods comes from the fact that many real-world tasks can be formulated as optimization problems and G3P methods can tackle a wide range of them. This flexibility comes from the ease of defining a search space with a grammar and because there is no need to implement new solution representations or search operators for each new domain.

Justin Diamond

Project owner

Hetzerk: Decentralized Physics Simulations

Free decentralized physics computations while rewarding computation & data for corporate, academic, and community involvement with SingularityNet, Nunet, and Cardano

Avatar 2 Human Learning: Net Zero and Carbon Negative Emissons

We propose an adaptive AI service API to improve sales and study within virtual worlds by reducing UI and UX friction as natural language processing, behavioral avatar AI, and accessibility intelligence. A partnership between Carbix Corporation and Xyris XR Design provides business acumen, an advanced complex technology, a library of existing meta verse projects, and interactive avatars, giving us a solid foundation for the success of this project. 

kenricnelson

Project owner

Risk-Aware Data Generation for SingularityNet Applications

Photrek will provide a Coupled Variational Autoencoder as a service for the SingularityNet community. This algorithm will enable the learning of risk-aware models that can generate robust, accurate simulations and forecasts.

ZRush – a hybrid control system for robot swarms

Swarm Robotics is an emerging field of adapting the phenomenon of natural swarms to robotics. It is a study of robots that are aimed to mimic natural swarms, like ants and birds, to form a system that is scalable, flexible, and robust. These robots show self-organization, autonomy, cooperation, and coordination amongst themselves. Benefits of swarms over traditional robotics include fault tolerance, scalability, operational efficiency, and ability to operate in dangerous or hazardous environments. 

We have taken inspiration from popular RTS (real-time strategy) games, such as StarCraft and Age of Empires, which have successfully used swarm intelligence and multi-agent AI systems in the past 20+ years.

ZRush aims to extend the same mechanics and user experience to real-world practical applications by designing a control system for swarms of robots that combines autonomous swarm behaviour with human piloting and decision-making. The control system focusses on precision, reliability, and flexibility and aims at overcoming the issues that currently limit the use of robot swarms in real-life scenarios.

Soubir Acharya

Project owner

Simply AI with Trust

Zero2AI aims to select the best usable API-based AI services on different repositories, including SingularityNET's decentralized AI platform, and offer them as curated microservices linked to an NFT. The goal is to make AI accessible to a non-technical audience right on their Smartphone. The proposition includes auto-curation of AI models and bias detection.