Inferix Decentralized GPU
  • Getting Started
    • Overview
    • $IFX
    • Resources
    • Brand Kit
    • Frequently asked questions (FAQs)
  • Inferix Whitepaper
    • Introduction
      • Rendering network using crowdsourced GPU
      • Rendering verification problem
    • High-level description of ANGV
      • Noise generation
      • Noise verification
      • Thread model
    • Implementation of ANGV
      • Structure of noise
      • Noise insertion
        • Geometric constraints
        • Distortion region
      • Adaptive noise spreading
      • Verification key generation
      • Noise verification
      • Threat analysis
        • Attacks on verification keys
        • Attacks on noises
        • Attacks on verifiers
      • Performance evaluation
      • Integration
    • Decentralized visual computing
      • Client Apps plugin
      • Client API and SDK
      • Manager node
      • Worker node
      • Decentralized storage
        • Data categories
        • Multi-level 3D polygon data
        • Polygon digester
        • Decentralized storage
        • Decentralized cache
      • Data security with FHE and TEE
        • Verifier data security enhancement with FHE
        • Worker and Manager data security enhancement with FHE
    • Decentralized federated AI
      • Federated learning with TensorOpera
      • Meta LLaMA
      • Stable Diffusion
      • Other AI models
      • Inferix AI
    • Economic model
      • GPU compute market for visual computing and federated AI
      • Inferix vision
      • $IFX token
      • Burn-Mint-Work token issuance model
      • Inferix bench and IBME
        • IB and IBM
        • IBME
      • Price simulation
      • Token metrics and allocation
        • Token allocation
        • Token vesting
      • Governance
      • Node staking and rewards
        • Worker
        • Verifier
        • Manager
        • Penalty pool
      • Node sale and guaranteed node buyback
        • Node sales
        • Guaranteed Node Buyback
    • Future development
      • PoR and NFT minting for graphics creative assets
      • ZKP and PoR communication
      • Inferix RemotePC
      • Rendering professional network
    • References
    • Appendix A: Proofs
    • Appendix B: Price simulation details
    • Appendix C: Hardware requirements for nodes
    • Appendix D: Performance evaluation data
  • Worker Node Guide
    • What is Worker Node
      • How do the Worker Node work
      • Worker Node Rewards
      • How to run Worker Node
      • What is the Worker Node License (NFT)
    • Worker Node Sales
      • Guide to Purchase Worker Nodes
      • Worker Node Sale Timeline
      • Node Supply, Price, Tiers and Purchase Caps
      • Guaranteed Node Buyback
      • How to get Node Whitelisted?
      • Smart Contract Addresses
      • User Discounts & Referral Program
      • Worker Node Purchase FAQ
      • ABKK Collaboration FAQ
  • Verifier Node Guide
    • What is Verifier Node
      • How do the Verifier Node work
      • Verifier Node Rewards
      • How to run Verifier Node
      • What is the Verifier Node License (NFT)
    • Verifier Node Sales
      • Guide to Purchase Verifier Nodes
      • Verifier Node Sale Timeline
      • Node Supply, Price, Tiers and Purchase Caps
      • Guaranteed Node Buyback
      • How to get Node Whitelisted?
      • Smart Contract Addresses
      • User Discounts & Referral Program
      • Verifier Node Purchase FAQ
      • Aethir Node Winners FAQ
  • Inferix MVP
    • Tutorial: MVP for designers & GPU owners
    • PoR MVP
  • Inferix Testnet 2 on Solana & IoTeX [ENDED]
    • Adding GPUs to the Network
      • For GPU providers
      • For GPU providers without funds
      • For users without GPUs
      • For Inferix Node Holders
    • Renting GPU Devices
    • User Revenue Calculation
      • Worker Rewards
      • Rental Revenue
      • Viewing Revenue
      • Claiming Rewards
    • GPU Staking & Unstaking
      • Staking Requirements
      • Unstaking GPUs
    • Guide to get tIFX tokens
    • Why choose Inferix DePIN GPU Solutions?
  • Inferix Testnet 1 on IoTeX [ENDED]
    • Inferix GPU Solutions
    • Adding GPUs to the Network
    • Renting GPU Devices
    • User Revenue Calculation
    • GPU Staking
    • Multiple options to participate in the Staking & Mining Program
    • Special airdrop for Inferix Node Holders! 🎉
    • Guide to get tIFX tokens
    • FAQ
  • Inferix Explorer
  • Team & Achievements
    • Our Story
    • Team
    • Member of Cohort 1 DePINSurf
    • Achievements
  • Community & Events
    • Events
    • Inferix Campaign: "ALLIANCE" (ENDED)
  • Terms of Service
    • Privacy Policy
    • Airdrop Terms of Service
Powered by GitBook
On this page
  1. Inferix Whitepaper
  2. Implementation of ANGV

Adaptive noise spreading

PreviousDistortion regionNextVerification key generation

Last updated 8 months ago

As discussed above, each atomic wi∈Ww_i \in Wwi​∈W has its distortion contribution at the spatial region kik_iki​ on the rendered image. Since the rendering process contains multiple options to improve the quality of the output (noise reduction, anti-aliasing, etc.), it is severe if the region kik_iki​(s) fall into perceptually insignificant regions of the image because the deliberate distortions raised by wiw_iwi​(s) would be eliminated by the rendering enhancement. Hence, distortions are preferred to be placed in human perceptually significant regions. However, to keep the compromise between the fidelity of the rendered frames and the robustness of noise verification, the strength of the distortion of each wiw_iwi​ must be tuned so that its deviation from locally enclosed regions is within a predetermined bound.

We handle this problem using the adaptive noise spreading , , . Research on the human visual perception agrees that the important information of images is located at high energy and low frequency spectral regions . Then before embedding the object vector Ω\OmegaΩ, we render the scene GGG at some low settings to get a sampling instance of the image. Next, we proceed both the spatial and spectral analysis on this instance to get perceptually significant spatial regions, called preferred regions. The atomic noises wiw_iwi​(s) will be placed in these regions. Simultaneously, we tune the expectation value μ\muμ used in generating atomic noises so that the energy (statistically given in proposition below) of high frequencies of the noises are sufficiently higher than the threshold used in the .

Proposition 2. (Convergence of energies)

Let {Xi,Yi∣i∈N}\left\{ \mathcal{X}_i, \mathcal{Y}_i \mid i \in \mathbb{N} \right\}{Xi​,Yi​∣i∈N} be a set of independent and identically distributed normal random variables Xi∼Yi∼N(μ,σ2)\mathcal{X}_i \sim \mathcal{Y}_i \sim \mathcal{N}\left(\mu, \sigma^2\right)Xi​∼Yi​∼N(μ,σ2). Let [XiYi]t[X_i \quad Y_i]^t[Xi​Yi​]t be a sample of the random vector [XiYi]t[\mathcal{X}_i \quad \mathcal{Y}_i]^t[Xi​Yi​]t and wiw_iwi​ be the signal (x,y)↦Ae2iπ(xXi+yYi)\left(x,y\right) \mapsto Ae^{2i\pi\left(\frac{x}{X_i} + \frac{y}{Y_i}\right)}(x,y)↦Ae2iπ(Xi​x​+Yi​y​) for any i∈Ni \in \mathbb{N}i∈N. Then the average of discrete Fourier transforms F‾n=1n∑i=1nFi\overline{F}_n = \frac{1}{n}\sum\limits_{i=1}^{n} F_iFn​=n1​i=1∑n​Fi​ converges:

F‾n(u,v)→n→∞a.sAM×N(1−e2iπMμ)(1−e2iπNμ)(1−e2iπ(1μ+uM))(1−e2iπ(1μ+vN))\overline{F}_n \left(u,v\right) \xrightarrow[n \to \infty]{a.s} \frac{A}{M \times N} \frac{\left(1 - e^{2i\pi \frac{M}{\mu}}\right) \left(1 - e^{2i\pi \frac{N}{\mu}}\right)}{\left(1 - e^{2i\pi\left(\frac{1}{\mu} + \frac{u}{M}\right)}\right) \left(1 - e^{2i\pi\left(\frac{1}{\mu} + \frac{v}{N}\right)}\right)}Fn​(u,v)a.sn→∞​M×NA​(1−e2iπ(μ1​+Mu​))(1−e2iπ(μ1​+Nv​))(1−e2iπμM​)(1−e2iπμN​)​

for all 0≤u<M, 0≤v<N0 \leq u < M, \ 0 \leq v < N0≤u<M, 0≤v<N.

Proof. Direct application of the continuous mapping theorem. For details, see .

It is worth noting that the rendering work of GGG generally contains multiple tasks, each requires to render multiple image frames. But the number of reference instances used for spectral analysis is much smaller, practically less than 1% the total number of rendered frames. While keeping the robustness of ANGV, we can adjust this ratio be even smaller by increasing the size nnn of the noise vector WWW.

noise verification
Appendix A
[7]
[16]
[17]
[18]
[10]