AFTD Auto-Formalizing Theoretical Domains

ReLUHiddenLayer

structure

A hidden layer of width k consists of a k-by-k rational weight matrix and a rational bias vector of length k.

Statement

structure ReLUHiddenLayer (k : ℕ) where
  weight : Fin k → Fin k → ℚ
  bias : Fin k → ℚ
Defined in
arXiv:2610.06092, Section 2.1
Verified
20 Oct 2026
Axioms
none listed
Used by 18
  • op75_final_encoding_templateThe final target layer encoding is a fixed prefix followed by the negative and positive target coefficient encodings.
  • ReLUNetworkA feedforward ReLU neural network from the real line to the real line with hidden layers of width k, specified by rational affine input…
  • op75_compact_block_effectFor an input in the unit interval and a nonnegative accumulator, the seven-layer block for a nonnegative integer weight transforms the…
  • op75_network_evalFor a unit-interval input, the explicit subset-sum network evaluates to the final orbit coordinate plus the absolute difference between…
  • relu_network_width4_liftingWidth-4 lifting theorem (Theorem 5.8 of arXiv:2610.06092): For every 3-CNF formula $f$ with variable indices in $\mathbb{N}$, there exists…
  • op75_block_encoding_templateThe seven-layer block encoding is a fixed prefix, the positive coefficient encoding, a fixed middle, the negative coefficient encoding,…
  • evalNetworkEvaluation of a feedforward ReLU neural network of hidden layer width $k$ on a real input $x$, which forms an initial width-$k$ vector by…
  • evalHiddenLayerEvaluation of a hidden layer of width $k$ on an input vector $x \in \mathbb{R}^k$, computing for each neuron $i$ the rectified linear unit…
  • op75_compact_blockSeven width-three hidden layers for one nonnegative integer weight in the one-pass subset-sum construction.
  • op75_compact_networkThe width-three network obtained by concatenating a seven-layer block for each integer weight, followed by the final target-checking layer…
  • relu_network3_encode_injectiveSanity check of the encoding: distinct width-3 ReLU networks have distinct Bool encodings.
  • op75_compact_finalA width-three layer that forwards the orbit coordinate and computes the two positive parts of the accumulated sum minus the target.
  • op75_fold_blocksFor a nonnegative initial accumulator and an initial point in the unit interval, folding the seven-layer blocks gives the final orbit…
  • relu_network_eval_continuousFor every natural number $k$ and every feedforward ReLU neural network net of width $k$ with rational weights and biases, the evaluation…
  • relu_network3_encodeEncoding of a width-3 ReLU network as a Bool word, of length linear in the total bit length of its rational entries: input weights, input…
  • op_75_relu_width3_liftingOpen Problem OP-75 (arXiv:2610.06092, Remark 5.15): For every 3-CNF formula $f$ with variable indices in $\mathbb{N}$, there exists a…
  • op75_compact_layerA width-three hidden layer formed from a rational three-by-three matrix and rational bias vector.
  • relu_encode_hidden_layer3Encoding of a width-3 ReLU hidden layer as a Bool word: its 3 by 3 weight matrix row by row, then its bias vector.

Read back from the Lean

A structure ReLUHiddenLayer parameterized by a natural number k : ℕ, with two fields: 1. weight: a map of type Fin k → Fin k → ℚ (a $k \times k$ matrix of rational weights), 2. bias: a map of type Fin k → ℚ (a length-$k$ vector of rational biases), with a default value assigned via sorry.

Written by a model that saw only the Lean, never the English above. If the two disagree, that is worth a challenge.

Lean source view module on GitHub

/-- A hidden layer of width k in a ReLU neural network with rational weights and biases. -/
structure ReLUHiddenLayer (k : ℕ) where
  weight : Fin k → Fin k → ℚ
  bias : Fin k → ℚ
See it in the graphDiscuss this resultChallenge it