EIP-7911 Scaling Ethereum with Perceptron Tree ZKPs (Scaling Ethereum with a Perceptron Tree ZKP)
No call decision · Scaling & data · EIP status Draft · created 2025-03-19
A proposal that uses a perceptron tree, combining decision trees with perceptron neural networks, to measure relationships across multiple transactions and compress them into a single Zero-Knowledge Proof, reducing gas fees and on-chain verification burden.
Why it is needed
As blockchain transaction volume has grown sharply, both processing capacity limits and privacy issues have surfaced together. Existing zero-knowledge proof approaches are effective at confirming that each individual transaction is valid, but they cannot analyze the relationships between transactions while preserving privacy. ZK-Rollups bundle multiple transactions off-chain and submit a single proof to the Ethereum mainnet to increase throughput and lower costs, but current bundling methods verify transactions individually and do not exploit patterns or relationships between them. Existing privacy solutions also generate a separate zero-knowledge proof per transaction, failing to use the full potential of zero-knowledge proofs. This proposal's perceptron tree model aims to overcome these limits by measuring similarity across multiple transactions and compressing it into a single unified proof.
What changes
Today
- ZK-Rollups bundle multiple transactions together, but within a bundle each transaction is still verified individually.
- Privacy solutions generate a separate zero-knowledge proof for each transaction.
- Patterns and relationships between transactions are not exploited in the proof structure.
After EIP-7911
- A perceptron tree measures similarity across multiple transactions and compresses it into a single unified proof.
- A smart contract uses a single proof to verify both a transaction's classification path and that it meets the similarity threshold.
- Validity is verified without disclosing a transaction's specific contents.
Who is affected
- Users: Bundling multiple transactions into a single proof reduces the gas fee burden.
- L2/rollup developers: They can optionally adopt this new compression method that leverages relationships between transactions in the proof.
- Smart contract developers: They would implement a verification contract that checks the submitted proof's root commitment, classification path, and similarity threshold.
- Client teams: Because the proposal operates at the smart contract level, no change to the protocol or consensus algorithm is required.
In depth
Building the perceptron tree
A recursive tree is built from an input transaction dataset T = {x1, x2, ..., xn}. Each node classifies transaction data using a perceptron with a linear function f(x) = step(Wa.a + Wb.b + theta), where the step function returns 1 if z >= 0 and 0 if z < 0. The tree is built by recursively splitting nodes according to conditions such as node purity (isPure) or maximum depth (maxDepth).
Two similarity calculations
Relationships between transactions are measured in two ways. Path similarity is the proportion of path shared by two transactions within the tree, sim_path(xi, xj) = d/D (d is the shared path length, D is the tree's maximum depth). For example, in a tree of depth 3, if x1 and x2 share only the root node (d=1), the similarity is 1/3 ≈ 0.33. Vector space similarity uses cosine similarity computed from each transaction's feature vector, sim_cos(xi, xj) = (xi.xj)/(||xi|| ||xj||). For example, if x1 = [0.7, 0.8] and x2 = [0.9, 0.4], the similarity is 0.91.
Five steps of proof generation
The proof is generated in five steps. First, a perceptron tree is built from transaction set T, training the perceptron at each node. Next, for each node n, a commitment Cn = H(Wn, bn, Cleft, Cright) is computed using the node's weights Wn, bias bn, and the hashes of its left and right child nodes. Then, a path proof is generated showing a specific transaction xi's classification path within the tree, and it is proven that the similarity between two transactions xi and xj exceeds a predetermined threshold theta. Finally, the final zero-knowledge proof Pi = {Croot, path(xi), sim(xi, xj)} is constructed, combining the root commitment, path proof, and similarity proof.
On-chain Verification Procedure
The smart contract checks three things in the submitted proof Π. First, it checks whether the root commitment C_root matches the pre-registered tree root hash. Second, it verifies that transaction xᵢ is correctly classified along the submitted path path(xᵢ). Third, it checks whether the similarity of the two transactions meets the predetermined threshold θ.
A Design That Requires No Protocol Change
Existing zk-SNARKs and zk-STARKs only verify individual transactions and do not exploit similarity between transactions, which can create inefficiency. This model structures transactions into a perceptron tree, compressing multiple transactions into a single proof and allowing a perceptron-based decision structure to adapt to varied transaction patterns. Because this proposal operates at the smart contract level, it requires no change to the existing Ethereum protocol or consensus algorithm, and it can be implemented alongside existing transaction verification methods, allowing optional adoption without a network upgrade. However, the source categorizes this proposal as Core.
Security considerations
The source lists four benefits. It protects privacy by verifying validity without revealing the specific content of a transaction. It detects data tampering through a commitment value based on the tree structure and weights. It prevents replay attacks by including a unique transaction ID in the proof. It keeps verification computation inside the smart contract simple and efficient, minimizing gas fees.
Current status: Draft
This EIP is a Standards Track (Core) proposal created on March 19, 2025, with status Draft. Since it is still at the draft stage, the details of the specification may change or the proposal itself may not proceed. There is no record of a decision to include it in a fork at a core developer call.