# Improving Discrete Optimisation Via Decoupled Straight-Through Estimator

arXiv: https://arxiv.org/abs/2410.13331 · HTML: https://mlsanity.com/paper/2410.13331/

## Key facts

- Cited by 5 papers in the map (data through 14 Oct 2026).
- Builds on 12 of its 17 references that are in the map.
- First posted on arXiv 17 Oct 2024.

## Builds on (most cited first, 12 of 12)

- [Neural Discrete Representation Learning](https://mlsanity.com/paper/1711.00937/) (2017)
- [Estimating or Propagating Gradients Through Stochastic Neurons for Conditional Computation](https://mlsanity.com/paper/1308.3432/) (2013)
- [The Concrete Distribution: A Continuous Relaxation of Discrete Random Variables](https://mlsanity.com/paper/1611.00712/) (2016)
- [MaskGIT: Masked Generative Image Transformer](https://mlsanity.com/paper/2202.04200/) (2022)
- [Visual Autoregressive Modeling: Scalable Image Generation via Next-Scale Prediction](https://mlsanity.com/paper/2404.02905/) (2024)
- [Mastering Atari with Discrete World Models](https://mlsanity.com/paper/2010.02193/) (2020)
- [InfoGAN: Interpretable Representation Learning by Information Maximizing Generative Adversarial Nets](https://mlsanity.com/paper/1606.03657/) (2016)
- [End-to-end Optimized Image Compression](https://mlsanity.com/paper/1611.01704/) (2016)
- [Transformers are Sample-Efficient World Models](https://mlsanity.com/paper/2209.00588/) (2022)
- [Deep Differentiable Logic Gate Networks](https://mlsanity.com/paper/2210.08277/) (2022)
- [Discrete-Valued Neural Communication](https://mlsanity.com/paper/2107.02367/) (2021)
- [Reintroducing Straight-Through Estimators as Principled Methods for Stochastic Binary Networks](https://mlsanity.com/paper/2006.06880/) (2020)

## Built on by (most cited first, 5 of 5)

- [Beyond Surrogate Gradients: Fully Differentiable Token Pruning for Vision-Language Models](https://mlsanity.com/paper/2605.28051/) (2026)
- [Voxify3D: Pixel Art Meets Volumetric Rendering](https://mlsanity.com/paper/2512.07834/) (2025)
- [NeuroSymActive: Differentiable Neural-Symbolic Reasoning with Active Exploration for Knowledge Graph Question Answering](https://mlsanity.com/paper/2602.15353/) (2026)
- [Conjunctive Prompt Attacks in Multi-Agent LLM Systems](https://mlsanity.com/paper/2604.16543/) (2026)
- [GMA-SAWGAN-GP: A Novel Data Generative Framework to Enhance IDS Detection Performance](https://mlsanity.com/paper/2603.28838/) (2026)

## Abstract (arXiv)

The Straight-Through Estimator (STE) is the dominant method for training neural networks with discrete variables, enabling gradient-based optimisation by routing gradients through a differentiable surrogate. However, existing STE variants conflate two fundamentally distinct concerns: forward-pass stochasticity, which controls exploration and latent space utilisation, and backward-pass gradient dispersion i.e how learning signals are distributed across categories. We show that these concerns are qualitatively different and that tying them to a single temperature parameter leaves significant performance gains untapped. We propose Decoupled Straight-Through (Decoupled ST), a minimal modification that introduces separate temperatures for the forward pass (τf) and the backward pass (τb). This simple change enables independent tuning of exploration and gradient dispersion. Across three diverse tasks (Stochastic Binary Networks, Categorical Autoencoders, and Differentiable Logic Gate Networks), Decoupled ST consistently outperforms Identity STE, Softmax STE, and Straight-Through Gumbel-Softmax. Crucially, optimal (τf, τb) configurations lie far off the diagonal τf = τb, confirming that the two concerns do require different answers and that single-temperature methods are fundamentally constrained.