Interpreting the Allora White Paper: A Self-Improving Decentralized AI Network

Written by: Xiaobai Navigation Coderworld

Memes are rampant in the current market.AI The track entered a short break.

However, with Nvidia's performance soaring and more coming in the second half of the yearAIIndustry events, crypto AI projects are still worthy of attention.

There is a new trend emerging - zkML (zero knowledgeMachine Learning)and AI Combination of AgentsThe former ensures privacy andSafetyAt the same time, verify the correctness of AI calculation results; the latter through intelligentcontractand decentralized networks to automate task execution and decision making.

Some old crypto projects will take advantage of this new trend to adjust their business direction and try to gain more value in the new cycle.

Allora Network is one of them.

Yesterday, Allora Officially announced its latest technical white paperPositioning itself as a "self-improving decentralized AI network" also means that the project's business is moving closer to narrative hotspots.

At the same time, the project also announced its points incentive plan in May, which is worth paying attention to for both the fur-pulling party and the Alpha hunter.

In an already crowded AI field, what is the uniqueness of Allora? Considering the complexity of its technical white paper, we have interpreted and analyzed it to present the key value points and project profile to everyone in a more popular way.

The old problem of AI resource monopoly

According to the Allora white paper, the project mainlyXiaobai NavigationIt is aimed at an old problem in the current AI field: computing power, algorithms, and data are concentrated in the hands of a few giants, and resource monopoly is not conducive to achieving the optimal state of machine learning (ML).

Allora believes that the key to building optimal machine intelligence lies in maximizing the number of connections in the network, allowing different data sets and algorithms to be freely combined in the network to obtain the most relevant insights.

Therefore, we need a form of swarm intelligence that can connect large data sets and inference algorithms.

In short, in existing crypto AI projects, the cooperation between different models is not good enough, and there are problems with the incentive method. The models are either isolated or not closely and effectively connected, resulting in unsatisfactory reasoning results.

Vitalik also mentioned before that "a higher-level mechanism is needed to judge the performance of different AIs so that AI can participate as a player."

Allora's goal is to enable nodes in decentralized AI networks to collaborate better through better incentive structures; at the same time, to introduce more intelligent ways to identify contextual details to improve the effectiveness of machine learning models, thereby achieving more efficient intelligent reasoning and judgment.

Allora: Introducing context awareness and differentiated incentives to improve model performance

Specifically, how does Allora achieve a "better decentralized AI network"?

The key highlight isContext-aware and differentiated incentive structures.These innovations enable the network to provide the best possible inference results in any setting, while providing fair rewards to each participant for their unique contribution.

But these two words sound a bit mysterious, so let’s first take a look at the participants in the Allora network.

Participants in the Allora network include workers, evaluators, and consumers, each with specific responsibilities and roles:

1. Workers: Provide AI inference results and predict the loss value of other workers' inference results.

2. Reputers: Evaluate the quality of the inference results and prediction loss values provided by the workers.

3. Consumers: They request inference results from the network and pay a fee.

As shown in the figure, the three main participants in the Allora network interact through a coordinator (Topic Coordinator):

• consumerRequest inference results from the network and pay a fee to obtain those results.

• WorkerProvides inference results and predicts the loss value of other workers' inference results. The coordinator synthesizes this information to generate more accurate inference results.

• EvaluatorBased on the inference results and predicted loss values provided by the workers, real data is used for evaluation to ensure the fairness of the evaluation, and rewards are obtained based on their consensus with other evaluators.

Through the design of these three roles, an efficient decentralized machine intelligence network is realized, achieving the goals of optimizing resource utilization and improving inference accuracy. In essence, it is a design that achieves self-improvement and fair rewards through role division and incentive mechanisms.

After understanding these three types of roles, it will be easier to look at Allora's context-awareness and differentiated incentive design.

Inferring the synthesis mechanism

Allora’s inference synthesis mechanism is the key to its decentralized machine intelligence. It is achieved through the following steps:

1. Inference Task: Each worker generates inference results using its own dataset and model.

2. Forecasting Task: Each worker predicts the loss value of other workers' inference results. These predicted loss values represent the expected performance of the workers under the current conditions.

3. Context-Aware Inference：The network uses the prediction loss values provided by the workers to generate a context-aware prediction inference result through weighted average. These weighted averages take into account the accuracy of historical and contextual dependencies.

4. Network Inference: The final network inference is generated by combining the worker's inference results and the context-aware predicted inference results.

The key to this mechanism is that it not only examines the historical accuracy of the model like other crypto projects, but also takes into account the current context, thereby achieving the best inference combination and improving the overall intelligence level of the network.

Differentiated reward mechanism

At the same time, Allora introduces a differentiated reward mechanism to ensure that each participant’s contribution is fairly recognized:

1. Worker Rewards: Distribute rewards based on their contribution to inference and prediction tasks, incentivizing them to provide high-quality data and predictions.

2. Evaluator Rewards: Rewards are distributed based on their proximity to consensus and stake, ensuring accuracy and fairness in evaluation.

3. Overall Reward Distribution：The reward mechanism not only encourages participants to actively contribute, but also avoids excessive concentration of a single participant through decentralized design.

Some solutions currently in use on Allora:

• AI Price Prediction:Providing accurate, real-time asset price information critical to advanced financial primitives.

• AI-powered vault: Enables developers to implement advanced DeFi strategies and increase earnings potential.

• AI Risk Modeling: Allows the protocol to build moreSafetysystems to deal with external risks.

• AnyML: Provides easy integration of any machine learning model so that anyone (not just machine learning engineers) can build more powerful products using decentralized AI.

Tokeneconomy

The Allora network uses its nativeTokenALLO来促进网络参与者之间的价值交换。ALLO代币的具体用途包括：

1. Purchase inference results: Users can use ALLO tokens to purchase inference results generated by the network. Allora adopts a "what are you willing to pay" (PWYW) model, allowing users to independently decide the ALLO fee to pay for inference.

2. Pay the participation fee: ALLO tokens can be used to pay for creating topics or participating in the network (as a worker, evaluator or network validator). Participation fees are variable.

3. pledge: Evaluators and network validators can use ALLO tokens for staking, and other token holders can also delegate their tokens to evaluators or network validators. Evaluators, validators and their delegators who stake will receive ALLO rewards.

4. Reward Payment: The network pays rewards to participants using ALLO tokens. For workers, these rewards are proportional to their unique contribution to the network's accuracy. For reviewers and network validators, these rewards are proportional to their stake and consensus.

Token Value

The token economics in the Allora Network are designed to ensure the intrinsic value and stability of the tokens:

1. Fee income: All network fees will be added to the network treasury to pay for rewards. This means that in practice, the network treasury will decay more slowly than a simple exponential decay, thus maintaining a high APY

2. Token Recovery: The fees collected for network usage are first used to pay rewards and then to mint new tokens. This means that depending on market dynamics, the circulation of ALLO can increase (corresponding to inflation) or decrease (corresponding to deflation)

3. Smooth issuance mechanism: By applying an exponential moving average, the issuance of tokens is smoothed, thereby avoiding a sharp drop in APY when the main tokens are unlocked, ensuring that token holders continue to stake their tokens.

However, the white paper does not mention the token sale date and details. For more information, please follow its social media.

The resources behind Allora

The above content actually does not mention the zkML technology mentioned at the beginning of the article, and it seems that Allora has nothing to do with this technology.

But behind Allora, the old project Upshot is the core contributor to Allora development.

Upshot enhances Allora’s capabilities by deploying its flagship price prediction model on the network, which provides AI-driven price information for over 400 million assets. The most accurate predictions from this model have historically shown confidence levels of 95-99%.

In addition, the output of this model can be obtained through zkPredictor (The largest on-chain zkML application to date) is provided to enable applications to use the output in a cryptographically verifiable manner.

At the same time, Upshot also received US$22 million in financing led by Polychain, Framework, CoinFund and Blockchain Capital in 2022. The direction at that time was to use technology to perform real-time NFT asset evaluation. With the rise of AI, the track has also changed, but the technology accumulated before has also been applied to the new Allora.

Roadmap and Testnet Incentives

According to previous information from Allora's official blog, the launch of the project is divided into three phases:

• Testnet Phase 1: Mid-February 2024

• Testnet Phase 2: Mid-March 2024

• Mainnet: Early second quarter of 2024

At this point in time, the project progress seems to have been delayed, but it is still in the stage before the mainnet launch.

In order to build momentum and get more people to use it, Allora also launched its first phase of the testnet incentive plan on May 17. Participating in on-chain and off-chain activities can also earn points in order to gain expectations for more subsequent airdrops.

Activities that earn points include:

On-chain activities

1. Create themes: Identify and define specific problems or areas of interest within the network to engage other participants in developing and delivering solutions.

2. Introduce machine learning models: Add machine learning models to the network for others to use.

3. Use Allora-powered applications: Engage with applications and services that leverage Allora's machine intelligence capabilities

Off-chain activities

1. CommunityGet involved: Follow Allora on Twitter and join the Discord and Telegram groups.

2. participateCommunityActivities: Participate in selectedCommunityEvents and activities to support the Allora Network.

Currently, activities that are easy for ordinary users to participate in can be found on the Galxe activity page. Interested players canClick here to participate

In general, Allora is an encryption project with certain technological innovations, background resources, and capability reuse. It can follow the trend in the transformation of AI hot spots and maximize its capabilities to expand new business directions, at least ensuring that it will not fall behind in the new attention war.

As for how high the upper limit is, one is to wait for the AI wind to blow again, and the other is to depend on more subsequent operational gameplay of the project.

The article comes from the Internet:Interpreting the Allora White Paper: A Self-Improving Decentralized AI Network

Related recommendation: Outlier Ventures: The rise of decentralized social media

This article will take a deep dive into several major crypto-decentralized social media competitors, their respective features, and architecture. Written by: Lorenzo Sicilia, Head of Engineering at Outlier Ventures Translated by: xiaozou, Golden Finance Outlier Ventures has noticed the healthy development of some decentralized social networks…