# Winograd Numerical Stability Improvement on ImageNet Synthetic CNN Workloads V2

Source: [Winograd Numerical Stability Improvement on ImageNet Synthetic CNN Workloads V2](<https://www.sota2.com/research/sota/winograd-numerical-stability-improvement-on-imagenet-synthetic-cnn-workloads-v2>)

Collection index: [llms.txt](<https://www.sota2.com/research/llms.txt>)

Data updated: 2026-02-26T05:39:44.998Z

Evaluation of model stability and numerical performance, specifically focusing on Winograd stability improvements, measured by condition number reduction and precision metrics on the ImageNet Synthetic CNN Workloads V2 dataset.

A research result applies to its stated dataset, split, metric and evaluation configuration. Compare like-for-like settings, preserve the metric's unit and direction, and distinguish paper-reported results from independent reproduction. The best result recorded here is not a claim of exhaustive, current global coverage.

| Scope | Recorded value |
| --- | --- |
| Task | Winograd Numerical Stability Improvement |
| Dataset (including any stated split) | ImageNet Synthetic CNN Workloads V2 |
| Metric | Condition Number Reduction (κ) |
| Unit | SCALAR |
| Direction | Higher is better |
| Results loaded | 3 of 3 |
| Coverage | All results currently returned for this benchmark |

Evaluation settings may differ between results. The original papers and recorded configurations are needed to establish comparability. A shared metric name alone does not establish an identical protocol.

## Metrics

- [Condition Number Reduction (κ)](<https://www.sota2.com/research/sota/winograd-numerical-stability-improvement-on-imagenet-synthetic-cnn-workloads-v2.md?metric=Condition%20Number%20Reduction%20(%CE%BA)>): Higher is better; unit: SCALAR.

- [Error Reduction](<https://www.sota2.com/research/sota/winograd-numerical-stability-improvement-on-imagenet-synthetic-cnn-workloads-v2.md?metric=Error%20Reduction>): Higher is better; unit: SCALAR.

- [Top-1 Accuracy Improvement](<https://www.sota2.com/research/sota/winograd-numerical-stability-improvement-on-imagenet-synthetic-cnn-workloads-v2.md?metric=Top-1%20Accuracy%20Improvement>): Higher is better; unit: PERCENT.

- [Precision/Exactness](<https://www.sota2.com/research/sota/winograd-numerical-stability-improvement-on-imagenet-synthetic-cnn-workloads-v2.md?metric=Precision%2FExactness>): Higher is better; unit: PERCENT.

## Recorded results

| Method | Condition Number Reduction (κ) | Result date | Report paper |
| --- | --- | --- | --- |
| NOVA | 415 | Not recorded | 694a21f29eebe296e72aa5a6 |

### NOVA

Condition Number Reduction (κ): 415. Result ID: 69994cac5698665ec4b6de53.

[Reporting paper](<https://www.sota2.com/research/paper/694a21f29eebe296e72aa5a6.md>)

[Method origin paper](<https://www.sota2.com/research/paper/694a21f29eebe296e72aa5a6.md>)

Recorded evaluation configuration: {"Point Selection (Pts)":true,"Basis Selection (Basis)":false,"Drop-in Compatibility":true}. This record does not assert independent reproduction.

Page 1 of 1. This is an index page, not the full collection.