---
kind: concept
slug: binary-neural-network
title: Binary neural network
summary: >-
  Networks whose weights and activations are approximated by sums of several binary terms, so that convolutions become bitwise operations. ABC-Net reports 65.0% ImageNet Top-1 for ResNet-18 with five weight and five activation bases, against 69.3% full precision and 51.2% for XNOR-Net. Each layer costs M×N binary convolutions, and no speed is measured.
tags: [quantization, binary-networks, efficient-inference, fpga]
papers: ["1711.11294", "2112.06126", "2505.08992"]
related: [weight-quantization, quantization-aware-training, post-training-quantization, batchnorm-folding, asymmetric-quantization, low-precision-arithmetic]
status: draft
updated_at: 2026-10-25
---

## Overview

A binary network keeps its weights and activations as values of +1 or −1, so a
multiply becomes a cheap bitwise operation. With binary values, a dot product
reduces to XNOR and popcount operations [arxiv:2112.06126 §6.1 eq. 14]. The gap
to full precision is the problem: a single-bit scheme loses much accuracy on
large tasks. ABC-Net [Towards Accurate Binary Convolutional Neural Network](/papers/1711.11294)
closes part of that gap by writing each layer's real weights as a weighted sum
of M binary filters, and its activations as a weighted sum of N binary
activations [arxiv:1711.11294 §3.1.1, §3.2]. The convolution then becomes M×N
binary convolutions, each scaled by trained factors [arxiv:1711.11294 eq. 12].
The network is trained with full-precision gradients through a straight-through
estimator, starting from a full-precision pretrained model [arxiv:1711.11294
§3.1.1, §3.3].

The invariant across the variants is the test-time form: binary bases and
binary activations, combined with a small number of real scale factors and
shifts [arxiv:1711.11294 §3]. Two surveys add the earlier single-bit schemes
and an FPGA case study. [arxiv:2112.06126 §6.1] covers the single-bit weight
and sign schemes, and [arxiv:2505.08992 §V.A] describes one binarised design
on an FPGA. Both surveys copy their numbers from the works they cite, so their
results are secondhand. Channel-wise weight bases are described but not tested
[arxiv:1711.11294 §3.1.2].

## Variants

### Per-layer weight bases with binary activations

M weight bases shared by the whole layer, and N binary activations with trainable
shifts [arxiv:1711.11294 §3.1.1, §3.2]. This is the setting in which the paper
reports its headline number: ResNet-18 reaches 65.0% Top-1 at M = N = 5, against
69.3% full precision [arxiv:1711.11294 tab. 2]. Card:
[Towards Accurate Binary Convolutional Neural Network](/papers/1711.11294).

### Binary weights with full-precision activations

The same weight approximation, but the activations are left at full precision.
At M = 5 on ResNet-18 this reaches 68.3% Top-1 [arxiv:1711.11294 tab. 1]. The
activations then stay real-valued, so the bitwise argument in the paper
applies to the weight side only [arxiv:1711.11294 §5.2]. The paper reports no
speed measurement for either variant.

### Single basis and scale (BWN and XNOR-Net)

With M = 1, each layer keeps one binary basis and one real scale, which the
ABC-Net paper says reduces to the BWN scheme [arxiv:1711.11294 §3.1.2]. The
survey gives the scale as the mean absolute weight of the matrix
[arxiv:2112.06126 eq. 15] [arxiv:2112.06126 §6.1]. On ImageNet with ResNet-18,
the survey lists BWN at 60.8% Top-1 with full-precision activations and XNOR-Net
at 51.2% with binary activations [arxiv:2112.06126 tab. 4]. ABC-Net's own table
lists the same two numbers [arxiv:1711.11294 tab. 3]. Its M = 1 run with
full-precision activations reaches 62.8% [arxiv:1711.11294 tab. 1], which is
not the same row (see Contradictions).

### Sign and stochastic binarization without scales (BinaryConnect, BinaryNet)

BinaryConnect draws each weight as +1 with a probability set by a hard sigmoid of
the real weight. BinaryNet takes the sign of the weight instead
[arxiv:2112.06126 §6.1 eq. 11, eq. 12, eq. 13]. The survey notes that BinaryConnect
keeps real-valued master weights for the gradient updates [arxiv:2112.06126
§6.1]. The survey's CIFAR-10 table lists BinaryConnect at 91.73% with binary weights
and float32 activations, and BNN at 88.6% with 1-bit weights and activations, on a
CNN [arxiv:2112.06126 tab. 3]. These are CIFAR-10 numbers and cannot be compared
with the ImageNet rows above.

### Binarised streaming pipelines on an FPGA (FINN)

FINN is a case study in the FPGA survey. It runs a binarised network as a
layer-pipelined stream, with weights in LUT arrays and 1-bit activations passed
between stages [arxiv:2505.08992 §V.A]. The survey quotes FINN on a Zynq-7020 at
1-bit weights and activations, about 5 TOPS counted in binary operations, around
50k LUTs (90% of the device) and no DSP use. Energy efficiency is not reported
[arxiv:2505.08992 tab. 2]. Binary operations are not comparable with the TOPS
figures of other designs in the same table. This case uses 1-bit weights and
activations, so it does not run the multi-basis form of ABC-Net.

## When to use which

| Situation | Use | Why |
|---|---|---|
| Accuracy must stay near full precision and the budget allows several binary convolutions per layer | M = N = 5, binary activations | 65.0% vs 69.3% on ResNet-18 [arxiv:1711.11294 tab. 2]. Each layer costs M×N binary convolutions [arxiv:1711.11294 eq. 12] |
| Activations are costly to binarize, or accuracy is more important than the activation speed-up | Binary weights, full-precision activations | 68.3% vs 69.3% at M = 5 [arxiv:1711.11294 tab. 1]. The paper's speed argument covers the weights only [arxiv:1711.11294 §5.2] |
| Tight compute or memory budget with a large accuracy loss acceptable | M = N = 1 | 42.7% on ResNet-18, lower than the 51.2% the paper lists for XNOR-Net [arxiv:1711.11294 tab. 2, tab. 3] |
| Need the smallest change from full precision and a single scale per layer is enough | Single basis and scale (M = 1) | The M = 1 case of the ABC-Net form [arxiv:1711.11294 §3.1.2]. Sign-and-scale is the form the survey describes for XNOR-Net [arxiv:2112.06126 eq. 15] |
| Deeper ResNet, large M and N | M = N = 5 | ResNet-50 reaches 70.1% vs 76.1% full precision, a 6.0-point gap against 4.3 for ResNet-18 [arxiv:1711.11294 tab. 2] |
| Target is an FPGA and the design can be a streaming pipeline of 1-bit layers | FINN-style binarised pipeline | Case study with 1-bit weights and activations, LUT-based weights and no DSP use [arxiv:2505.08992 §V.A, tab. 2] |
| Need a measured speed or energy figure | Not from these sources | No timings, hardware results or energy numbers are reported for ABC-Net [arxiv:1711.11294 §5.2]. The FINN energy figure is not reported [arxiv:2505.08992 tab. 2] |
| Training budget is tight | Per-layer bases | The channel-wise variant adds training computation [arxiv:1711.11294 §3.1.2] |
| No official reference code | Re-implement from the equations, starting with M = 1 | The ABC-Net card found no official code [arxiv:1711.11294 §4]. Third-party code was not used |

The paper's accuracy numbers are from ImageNet with ResNet-18, ResNet-34 and
ResNet-50 [arxiv:1711.11294 tab. 2]. A VGG-like network on a forest-trail
task is reported separately, with a single run per model [arxiv:1711.11294
tab. S5]. The survey's CIFAR-10 results cover different networks and are not
ranked against ImageNet results [arxiv:2112.06126 tab. 3]. Results for other
tasks or architectures are not claimed.

## Implementation sketch

Python-flavoured pseudocode of the test-time form. The shifts, the number of
bases and the scales are trained; the sign of zero and the least-squares solver
are not stated in the ABC-Net paper [arxiv:1711.11294 §3.1.1, eq. 3]. The
BinaryNet sign maps zero to +1 [arxiv:2112.06126 eq. 13]. That is a different
scheme and does not settle the ABC-Net choice.

```pseudocode
# Step 1: M binary weight bases from shifted, centred weights  [eq. 2]
def weight_bases(W, shifts):
    centred = W - mean(W)
    spread = std(W)
    # sign(0) is not stated for ABC-Net; BinaryNet maps zero to +1  [eq. 13]
    return [sign(centred + u * spread) for u in shifts]

# Step 2: scales alpha by least squares on the fixed bases  [eq. 3]
# Solver not stated in the paper
alpha = least_squares(bases, W)

# Step 3: N binary activations with trainable shifts v_n  [eq. 8, eq. 9]
def binary_acts(R, v):
    # Threshold sits at R = 0.5 - v_n
    return [2 * (clip(R + v_n, 0, 1) >= 0.5) - 1 for v_n in v]

# Step 4: convolution as M x N bitwise convolutions  [eq. 11, eq. 12]
out = sum(alpha[m] * beta[n] * conv(bases[m], acts[n])
          for m in range(M) for n in range(N))
```

Training uses a straight-through gradient through the sign and rounding steps
[arxiv:1711.11294 eq. 7, eq. 10]. The mask given for the activations does not
match the forward threshold, see Contradictions.

## Contradictions

- **Gap to full precision, same paper.** The text gives a Top-1 gap of 0.9 points
  for M = 5 with full-precision activations [arxiv:1711.11294 §4.1.1]. Table 1
  gives 69.3 − 68.3 = 1.0 points [arxiv:1711.11294 tab. 1].
- **"Comparable accuracy" claim.** The introduction says the accuracy is
  comparable to the full-precision model [arxiv:1711.11294 §1]. The tables show
  that for binary weights with full-precision activations only (68.3 vs 69.3),
  and show a 4.3-point gap for binary weights and activations at M = N = 5
  [arxiv:1711.11294 tab. 2, tab. 3].
- **Straight-through mask versus threshold.** The forward threshold is at
  R = 0.5 − v [arxiv:1711.11294 eq. 9]. The gradient mask is written as
  0 ≤ R − v ≤ 1, which is centred on a different point [arxiv:1711.11294 eq. 10].
  The paper does not discuss the difference.
- **BNN row in the comparison table.** The table lists BNN at 42.2% Top-1; the
  paper states this is its own run on ImageNet, not a published figure
  [arxiv:1711.11294 tab. 3, §4.1.3]. The survey's BNN figure of 88.6% is a
  CIFAR-10 result [arxiv:2112.06126 tab. 3]. The two figures are for different
  datasets and are not a disagreement.
- **BWN at M = 1.** ABC-Net says its M = 1 case reduces to BWN [arxiv:1711.11294
  §3.1.2]. Its own table reports M = 1 with full-precision activations at 62.8%
  Top-1 [arxiv:1711.11294 tab. 1]. The survey lists BWN at 60.8% for the same
  network and setting [arxiv:2112.06126 tab. 4]. The paper does not explain the
  gap between the two numbers.

## Papers

- [Towards Accurate Binary Convolutional Neural Network](/papers/1711.11294):
  weight and activation approximation by M binary weight bases and N binary
  activations, with ResNet-18 results on ImageNet.
- [Neural Network Quantization for Efficient Inference: A Survey](/papers/2112.06126):
  survey of quantization schemes, including the binary weight, sign and
  stochastic schemes and their CIFAR-10 and ImageNet numbers.
- [Edge-AI FPGA dataflow and tiling review](/papers/2505.08992):
  survey of FPGA dataflow and tiling, with the FINN binarised streaming
  pipeline as one case study.