Explore our core focus areas and recent scientific outputs in AI and decentralized infrastructure.
A training-free cryptographic signing layer for audio: an Ed25519 signature hidden in phase and magnitude, verified with the public key alone. 97.5% to 98.3% verification across eight attack conditions at mean PESQ 3.02. Sponsored and led by Phi Lab Foundation.
View details →A training-free video watermark that removes the shared secret. The signer embeds a complete Ed25519 signature in the chroma phase spectrum; anyone holding the 32-byte public key verifies offline. Paper on arXiv, open-source JavaScript library, and a browser demonstration.
View details →Neighbouring crowns stop short of touching, leaving channels of sky no single tree owns. A stand-level simulation where the gap is never specified — it settles as an equilibrium between growth and contact loss. Includes an interactive showcase.
View details →Generic AI-image detectors collapse out-of-distribution (worst-group AUROC 0.526), while a face-focused detector stays robust across unseen generators and corruptions.
View details →Benchmarking demographic representation bias in state-of-the-art image generation models, with mitigation via prompt engineering and fine-tuning.
View details →Adapting vision-language models to quantify visual patterns in children's drawings, supporting a non-clinical screening workflow for developmental psychology.
View details →A benchmark of aligned real–generated pairs with frequency-space evaluation for open-set, cross-architecture robustness testing of AI-image detectors.
View details →Proof-of-Work powered DDNS integrating IPFS storage, zero-trust signatures, and 15-second record propagation.
View details →Multi-branch frequency–spatial fusion model reaching 97.8% accuracy on diffusion-generated image forensics.
View details →Adaptive mid-frequency phase embedding to authenticate audio in the era of AI voice synthesis.
View details →Multi-agent system for automated fact-checking using nine specialized agents, achieving 87% macro-F1 score with evidence-grounded reasoning.
View details →