3/ How Polyhedra i-D works:-Verification runs locally on your device-Only a cryptographic proof is shared, never the face-Proof verified on-chain in milliseconds, immutable & non-reversible-Built on zkML + GPU-accelerated ExpanderA new trust primitive for the AI-native internet. ...

Problem Statement - Traditional ID checks are vulnerable to AI-driven forgeries, posing a significant risk to identity verification processes [1] - Centralized biometric databases present ongoing data leak risks, jeopardizing user privacy and security [1] - Regulated industries face challenges in age/identity verification without wanting to store sensitive user data, creating a compliance burden [1] Solution: Polyhedra i-D - Polyhedra i-D utilizes local inference and Zero-Knowledge Proofs (zkML) for face recognition, ensuring user data privacy [2] - The system generates cryptographic proofs instead of uploading raw photos, protecting sensitive information [2] - Service providers receive verification results without accessing underlying user data, enhancing data protection [2] - On-chain verification allows smart contracts to directly verify proofs, enabling decentralized identity management [2] Technology and Roadmap - zkML proves facial verification locally without exposing inputs, enhancing privacy [4] - Expander Prover accelerates proof generation, improving efficiency [4] - zkPyTorch turns PyTorch ML models into ZK circuits, enabling advanced applications [4] - Trusted Execution Environments (TEE) securely store biometrics in line with C2PA standards, ensuring compliance [4] - A web demo is live, with a mobile app release and SDK for third-party integrations planned [4] - Expansion into finance, age verification, and broader digital identity use cases is underway [4] Key Benefits - Users can prove legitimacy without handing over sensitive documents, enhancing privacy and control [4] - Businesses avoid the liability of storing identity data, reducing risk and compliance burdens [4] - Polyhedra i-D establishes a default login layer for the AI-native internet, combining compliance with data protection [4]

AI and Market Opportunity - AI is rapidly creating trillion-dollar markets [1] - The rise of unicorns and billion-dollar compute deals signifies substantial market activity [1] Trust and Verifiability - Verifiability is crucial for ensuring AI systems are real, safe, and trustworthy [1] - Building trust in AI systems is essential before they reshape jobs, the economy, and even human bodies [1] zkML's Role - zkML is designed for a future where verifiability is paramount [1]

Product & Technology - Early stage zkML product is being tested [1]

AI Development & Future - The future of AI is verifiable, indicating a focus on transparency and trust in AI systems [1] - zkGPT prototype and repository are coming soon, suggesting rapid development and open-source initiatives in the AI field [1] - Integration with zkPyTorch is planned, highlighting the importance of compatibility and collaboration within the AI ecosystem [1] - zkML-as-a-service is on the horizon, pointing towards the commercialization and accessibility of AI technologies [1]

Performance Improvement - CUDA 13.0 compatibility fix for Fiat-Shamir [1] - Shared memory optimization achieves 1 TB/s bandwidth [1] - Achieved 9,000 zk proofs/sec on m31ext3 [1] - GPU acceleration for MSM on KZG commitments [1]

Development Progress - Expander merged Ethereum Foundation PR to fix MPI bugs in macOS 15 build [1] - Expander enabled Sumcheck protocol for variable-length polynomials [1] - Expander progressed on Docker service module for zkML (zero-knowledge machine learning) [1]

Technology Advancement - zkML updates aim for faster proving processes [1] - zkML updates target lighter computational requirements [1] - zkML updates emphasize easier deployment, even on personal devices [1]

Performance Improvements - Improved shared memory across multi-threaded processes [1] - Flexible SIMD config for better parallelism [1] - Refined PCS interface and efficient multi-claim merging [1] - Setup, proving, and verification are now cleanly separated [1] Resource Optimization - Lower memory footprint for zkML, for example, VGG requires less than 8GB [1] - Fine-grained CPU resource control and deterministic proofs [1]

Product Update - A major update to Expander, a proving backend, has been released [1] - Expander is built for performance, flexibility, and real-world zkML use [1]