Potential Use Cases

This section outlines some potential use cases for the future of the NuNet Platform.

Decentralized genetic-algorithm framework

As a first step toward supporting a rich variety of AI processing algorithms, NuNet will enable to the coordination of decentralized and distributed optimization processes using Genetic Algorithms (GA) and Genetic Programming (GP). These algorithms are chosen because:

  • They are particularly well suited for a processing infrastructure consisting of a large number of processing units with widely varying capability (including some with the very weak capability), loosely and erratically connected together;

  • They are applicable to a wide variety of AI problems and optimization problems, applicable to a variety of practical and scientific domains.

This NuNet based GA/GP framework will support any genotype (solution space) and fitness function (objective function) fulfilling certain APIs, as is commonly done in OO genetic algorithm frameworks.

A specialized, simplified version of many aspects of the broader NuNet ecosystem would apply in this context:

  • AI developers will contribute AI plug-ins to improve the GA/GP framework, e.g. specialized mutation and crossover operators, or EDA (Estimation of Distribution Algorithm) modeling tools or fitness estimation methods

  • Applied AI developers will write code mapping specific types of practical real-world problems (e.g. predicting financial or climatological time series, designing certain types of machinery, extracting concepts from text, mining patterns in tabular or graph data, clustering data vectors, classifying genomic data, etc.) into GA/GP problems

  • App developers will write NuNet apps using this AI code to solve specific problems, e.g. predicting aspects of climate change, recognizing patterns in ecommerce data, learning classification rules from genomic data about human disease, etc.)

  • Users will be able to choose from among these apps, running multiple apps on their devices at various points in time, and in many cases receiving tokens as reward for their contribution of resources

The NuNet-based GA/GP toolset will support commercial services, in which apps provide value to customers who then pay for their services, with their payment ultimately resulting in tokens flowing to the NuNet resource providers. Payment may be made directly in NuNet tokens or in other tokens or (e.g. fiat) currencies via conversion gateways.

As NuNet framework will support staking resources, fiat, and currency for socially beneficial services as defined and voted for by the community members, it will enable to use of GA/GP to provide AI analysis for the common good, e.g. data analysis toward climatology or medicine. It will allow running computation loads of certain socially beneficial but not necessarily commercially profitable and adequately funded projects on the framework. Additionally, NuNet may decide to provide bonus tokens to organizations using AI tools and computing resources for common good, so that they can adequately compensate resource owners; in this sense, a GA/GP application would be used to experiment with tokenomic as well as algorithmic methodologies of the post-monetary economy. Is this a potential case?

Bringing compute to data

Any computing, especially state-of-the-art AI and Machine Learning models, require huge amounts of data to process. Often the data is highly sensitive and/or protected by IP rights. Such data leaving the firewalled premises of its owner is often impossible and leads to heavy inefficiencies in the data economy, where best AI / ML algorithms are unavailable for the high-quality data and vice versa.

Thanks to the functionality of mobile computing processes, NuNet will be able to post any computing process (therefore any AI / ML algorithm) to any NuNet-enabled data source behind the farewell and ensure that no data will cross the firewall boundary. This will greatly increase the ability to use the best available ML/AI algorithms for high-quality data. NuNet is planning the cooperation and technical exploration with Ocean Protocol and other DAIA members for developing a general framework of mobile computing data and applying it to concrete business cases.

Dynamic data aggregation

Mobile computing processes may also allow the building of distributed and decentralized databases and streaming data sources that combine a number of independent actors without requiring to fully expose proprietary data of these sources to third parties. For example, NuNet would provide a containerized program (local agent) for installation on the premises of a proprietary database owner (e.g. hospital of a pharma company). A container would accept external queries, pass them to internal database engines, collect answers, anonymize/secure them and send data to a NuNet-enabled AI engine where data from many local agents would be aggregated and processed. Alternatively, NuNet agents may expose browser plugins as micro-sources of streaming data from individual users, opening a myriad of use cases involving human computation and anonymized data collection (e.g. sentiment, opinions, fake news, text highlighting, etc.). Further, local agents may do part of pre-processing, accounting, and payments. As a bonus, the mechanism would ensure that if relevant data changes on a local database, it becomes immediately available for all agents in the network.

Federated machine learning

The concept of Federated Learning was first introduced by Google in 2017 as collaborative machine learning without central data. Models are trained either from a starting point or from scratch using millions of distributed computing devices and their data. This approach is radically different from traditional, centralized training where the data and computational power are owned, operated, and controlled by a single entity.

In the decentralized learning approach, the device downloads a model from a resource consumer then uses its own data to train this model, and just sends the result of this training step back. Leveraging the data and computing of the devices to be combined to update a model in the cloud. The updates are encrypted and not identifiable - the data remains anonymous and not traceable back to its origin.

This kind of machine learning system is a perfect use case for NuNet and its goals. To realize this quickly we will leverage existing federated training frameworks and include the NuNet adapter in them. This will make it possible to leverage NuNet and all its participating devices for commercial and common training of AI models.

The models can either be single-party, multi-party, or a common system. An example of a single-party system would be a music recommendation service where the model is adapted using individual choices but after a certain number of choices and model updates, the result is communicated back. Here the funding of the resources distributed to the participants would be invested by a single party.

A multi-party system could be a fraud detection system of multiple financial partners that are sharing resulting models but don’t want to expose their individual data sets. Another use case of this would be an autonomous driving model that is shared by multiple cars from multiple vendors but updated on the data of all of them. The funding of the compensational resources is done by the involved parties.

A common system could be a climate prediction model shared and updated by all humans interested in stopping climate change. Another common models could be a medical system to predict and prevent diseases trained and shared among all humankind. The participants can donate their resources to achieve a greater good without having to spend money but instead leveraging their mobile phones and power but most importantly their time and data.

This kind of training adapts the model and possibly the application using it on the device to the data of the individual user - thus resulting in a hyper-personalized model where the input of the individual becomes an update for the model of the whole.

Health data pre-processing and sharing

The health wearables device market is booming with double-digit annual growth rates, expected to reach 450 million shipments by 202225 and amount to $60B markets by 2023. Health wearables become parts of patients’ treatment plans, insurance companies’ policies, and individual lifestyles. The amount of data collected globally by these devices increases even at higher rates.

However, data privacy and consent are continuing to pose significant barriers to the realization of the myriad opportunities offered by individuals’ health-related data in the domains of patient monitoring, longevity therapeutics, predictive, preventive, personalized, and participatory medicine, and medical research in general. Providing continuous aggregation, processing, and mining for data collected via multiple devices from different vendors is an unresolved engineering and management problem. Unlocking this potential is paramount for human society and civilization on many levels, solutions towards which contain tremendous social, business, and personal value.

Data sharing among individual health wearables are currently very limited -- each wearable provides only certain types of sensors and information but not other, i.e. heart rate, step count, blood pressure, sleep cycle, etc. This information is usually stored in a private cloud, which is accessible and shareable only among users of the same provider or device manufacturer. Medical research and precision medicine, however, are based on the integration of all these types of data which, currently is based on sharing databases at the provider level. Furthermore, predictive, preventive, personalized, and participatory medicine (P4) needs the integration of other types of data, including genome sequencing, electronic medical record, and more.

The core architecture of NuNet provides a platform enabling effective management and enaction of decisions regarding data sharing, processing, storage, and anonymization, where these decisions may happen at the level of the individual human or device. Specifically, in the NuNet approach:

  1. The Fitness data of each device is recorded and stored in a local database or cloud as the application of a device provider allows;

  2. NuNet adapter is installed on every individual device (e.g. smartphone or tablet) where fitness data is stored. The adapter exposes NuNet APIs for resource description, traceability and provenance, resource and data discovery, and others;

  3. Using these APIs each device announces the availability of certain types of data to other NuNet-enabled devices; Using the same APIs, healthcare service providers and data aggregators announce their services;

  4. Healthcare, personalized and precision medicine providers, longevity therapeutics, and medical researchers use NuNet-enabled devices to find data sources, sign contracts, provide micropayments for personal data usage and offer their services in terms of personalized advice and analysis;

  5. Health wearable users can search for service providers and additional knowledge

    that can be retrieved from their data. Alternatively, they receive notifications about offers from service providers which they may accept or reject at any moment, retrieving their personal data;

  6. NuNet enables data pre-processing and anonymization at a user's device by decentralizing data processing workflow and installing parts of it to devices where data is. In this way, it ensures that private data never leaves a device in the first place.

  7. Service providers, which use proprietary or open-source algorithms for data analysis and aggregation, leverage the decentralized computing framework of NuNet to bid for free computing resources available on mobile devices which may or may not correspond to the ones that provide data.

  8. Individual data and compute resource providers establish formal digital relations with data aggregators and compute resource users via the smart contract mechanism. Smart contracts can involve any type of barter (data for analysis, compute resources for data) exchange or micropayments. Canceling a smart contract ensures that the data is not accessed by third parties anymore.

Secure data exchange in decentralized systems

The complex problem of ensuring the security of IoT ecosystems is the biggest obstacle to large-scale IoT adoption and integration into business models. Furthermore, data privacy, provenance, and high granularity access management, while being instrumental for unlocking the potential of the data economy, hit new levels of complexity in decentralized IoT systems where β€œfirewalls” have to be distributed across a large number of devices, most of which are too low powered to run full operating systems or an Internet protocol stacks. IoT security systems have to be decentralized by design, without a single trust layer or a trusted party. The ability to customize and integrate blockchain and state-of-the-art trusted computing technologies in decentralized computing workflows on the NuNet framework allows solution providers to address many obstacles to IoT adoption by design on a case-by-case basis.

Flexible decentralized computations at the edge

NuNet leverages the computing frameworks of its partners by allowing to build flexible and radically decentralized computation graphs spanning IoT devices of different capacities and owned by different economic players and community members - simple or advanced sensors, robotic microcontrollers, embedded systems, virtual machines on the edge, fog and cloud. It enables to the design of efficient and fast data and AI workflows for dynamic IoT environments where huge amounts of streaming data can be processed as close to the edge as required by the business model and capacities of the particular system. In the future, NuNet and SingularityNET are planning to partner for implementing technologies required for the automatic adaptability for balancing computing loads in IoT networks in real-time.

On-demand computing resources for layer 2 technologies

A somewhat unexpected but potentially highly valuable application of the NuNet framework is its ability to provide on-demand computing resources to applications built on blockchain layer 2 technologies. For example, decentralized exchanges (DEX) are suffering the same scalability and speed issues as the underlying blockchains and are orders of magnitude slower than their centralized counterparts. While DEXes solve the scalability and speed issues of real-time matching by different designs from centralized exchanges, speed and scalability will remain important factors for the development of the DeFi industry. Upcoming layer 2 technologies (e.g. Ethereum Plasma or Cardano Hydra) will allow offloading computing from the main blockchain to side chains and in this way increase its scalability. Exchanges are known to experience surges of high-frequency trading and low-latency are races that are quite rare and require a lot of spare computer power to be provisioned by centralized exchanges 27. Decentralized exchanges, however, may allow high-speed traders to acquire speed and computing power from peer-to-peer networks, such as NuNet, on-demand. NuNet’s in-built dynamic pricing of computing resources may provide economic mechanisms for computing resources to be allocated where there is a need.

Last updated