Original project plan: https://hackmd.io/qdYZCxweQmix8ezdh7l-Aw?view Changed milestone As we’ve deepened our understanding of the field, we are refining the second milestone to focus on a specific FHE-based private proof delegation protocol and define concrete deliverables around it. Additionally, we are adjusting the project timeline due to an unexpected reduction in team size. Milestone : Implementation and Benchmarking of FHE-SNARK Milestone Description The third milestone (June–August 2025) focuses on implementing and analyzing core components of the FHE-SNARK protocol described in ePrint 2025/302. Our primary goal is to uncover and characterize the practical costs and trade-offs inherent in this protocol — especially those that are not precisely captured in the original theoretical analysis. This includes:

What is committed oblivious transfer(COT)? COT is oblivious transfer with commitments to sender's messages, chooser's choice bit, and chooser's received message. Verifiable OT(VOT) is almost the same with COT but it doesn't require the commitment to received message. COT can produce commitment to the chooser's output value which can be used by the surrounding protocol later e.g. in zkSNARK. Any VOT protocol can be transformed into COT by committing to the output message $s_b$ and prove its authenticity. It can prevent selective failure attack. COT for single bits COT for single bits was first introduced in 1995 [1]. OT sender Alice is committed to bits $a_0$, $a_1$. OT receiver Bob is committed to bit $b$. At the end of the protocol, Bob is committed to $a_b$ and knows nothing about $a_{\bar b}$ and Alice learns nothing about $b$. We will not cover this protocol in detail in this doc. Please refer to [2].

Why Privacy Preserving Machine Learning (PPML) Protect sensitive data during ML process Prevent sending sensitive data directly to MLaaS server Prevent leaking model and its parameters LLM memorize training examples Public language model fine-tuned on private data can be misused to recover private information Infer if specific user is in the training set etc In different ML steps

Recap: What is ZKP? Given relation $R$, statement $x$ and witness $w$, Prove $R(x,w) == 1$ without leaking knowledge about $w$ What is IZK? -> Interactive Zero-Knowledge proofs Pros No trusted setup Fast prover

2PC with preprocessing GPU acceleration. different protocol for GPU and CPU Order of magnitude better latency, communication than CrypTen Standard 2PC security (semi-honest?) SIGMA is based on Orca. Orca only focuses on convolutional neural network that supports simple non-linearities like relu. SIGMA propose FSS based protocol for non-linear operations like GeLU, Softmax, and layer-normalization. Description of FSS

HardnessとProgrammable Cryptography Hardnessとはなにか?💎 ハードネス (硬度?確度?) スクリーンショット 2023-11-27 170227 How likely the thing will happen. どれくらいの可能性である事象が起こる(あるいはある状態になっている)かを表す尺度

Prerequisite Knowledge Cryptographic Hash Function Secret Sharing TechniquesAdditive Sharing Polynomial Interpolation (Shamir) Agenda General Introduction to MPC Introduction to Garbled Circuit Introduction to MPZ Let's play with MPZ

Group1 Social graphs/Reputation System Eigen Trust Trust scoring algorithm in a peer-to-peer network. Each peer has local trust scores of the peers it knows. (trust scores are calculated locally and normalized) Using MPC, eigen trust scores can be calculated without disclosing the local trust scores Challenges The number of peers in the network grows, number of communication/computations grows

What is this: 2-party computation that let 2 parties to compute agreed upon function with their private data forcing them to take some pre-defined on-chain action depending on the result of the computation. By using garbled circuit and some zkps, we can construct such scheme as publicly verifiable 2-party computation. MPC Discussion What is MPC (Secure multiparty computation)? Quick review. MPC is a cryptographic schemes that enable multiple parties to jointly compute a function on their private inputs without revealing any information about their inputs to the other parties. Parties: P_1, ..., P_n Secret Inputs: x_1, ..., x_n Evaluate: f(x_1, x_2, ..., x_n)

Purpose of this exchange: What users can do? Private order matching : match order without revealing info to the counter party using PeekABook Private peer finding : find advertiser without revealing info to intermediaries using blind-find Atomic swap : swap a pair with the advertiser on agreed pair price without revealing info to other parties using ZKOPRU Todos [ ] Design a way to establish p2p connection after searching a peer using blind-find network [ ] Refine js-smp to amount negotiation(maybe advertiser just can accept if taker's amount is smaller or equal to the advertiser's amount)