Snark ChocolateICICLE
Resources
Company

Yuval Domb

Yuval Domb is a multi-disciplinary engineer and chief architect at Ingonyama. He has decades of experience leading research, algorithms, and design teams in various fields such as wireless communication, algorithmic trading, and most recently cryptography. Yuval holds a PhD in Electrical Engineering - Information Theory from Tel Aviv University.

Blogs by 

Yuval Domb

Fast Danksharding using ICICLE

In this blog we provide technical details on our Danksharding CUDA implementation. Using ICICLE, we have implemented a flow that can run entirely in the GPU, with only the relevant outputs to be sent to a host machine and populated on the Ethereum network.

A Mathematical Theory of Danksharding

Danksharding is the new sharding design proposed for the Ethereum 2.0 blockchain, which introduces significant simplifications compared to previous designs. In Danksharding, the Beacon block is a periodic data structure, constructed by a Builder, whose primary concern is to enable Validators to verify the correctness and availability of its data, using constant-time sampling and verification.

Multi-Precision Fast Modular Multiplication

Improving modular multiplication, even by a little bit, can result in significant acceleration…So obviously, we at Ingonyama are obsessed with this problem.

Fast Modular Multiplication

Modular multiplication is arguably the most computationally-dominant arithmetic primitive in any cryptographic system. This note presents an efficient, hardware-friendly algorithm that, to the best of the author’s knowledge, outperforms existing algorithms to date.

Provable Watermark Extraction

zkDL++ is a novel framework designed for provable AI. Leveraging zkDL++, we address a key challenge in generative AI watermarking: Maintaining privacy while ensuring provability. By enhancing the watermarking system developed by Meta, zkDL++ solves the problem of needing to keep watermark extractors private to avoid attacks, offering a more secure solution. Beyond watermarking, zkDL++ proves the integrity of any deep neural network (DNN) with high efficiency.