# Engineering Monosemanticity in Toy Models

arXiv: https://arxiv.org/abs/2211.09169 · HTML: https://mlsanity.com/paper/2211.09169/

## Key facts

- Cited by 13 papers in the map (data through 14 Oct 2026).
- Builds on 2 of its 11 references that are in the map.
- First posted on arXiv 16 Nov 2022.

## Builds on (most cited first, 2 of 2)

- [Toy Models of Superposition](https://mlsanity.com/paper/2209.10652/) (2022)
- [Polysemanticity and Capacity in Neural Networks](https://mlsanity.com/paper/2210.01892/) (2022)

## Built on by (most cited first, 13 of 13)

- [Mechanistic Interpretability for AI Safety -- A Review](https://mlsanity.com/paper/2404.14082/) (2024)
- [Finding Neurons in a Haystack: Case Studies with Sparse Probing](https://mlsanity.com/paper/2305.01610/) (2023)
- [Disentangling Neuron Representations with Concept Vectors](https://mlsanity.com/paper/2304.09707/) (2023)
- [Enhancing Neural Network Interpretability with Feature-Aligned Sparse Autoencoders](https://mlsanity.com/paper/2411.01220/) (2024)
- [Mixture of Experts Made Intrinsically Interpretable](https://mlsanity.com/paper/2503.07639/) (2025)
- [Diversity-driven Data Selection for Language Model Tuning through Sparse Autoencoder](https://mlsanity.com/paper/2502.14050/) (2025)
- [Measuring and Guiding Monosemanticity](https://mlsanity.com/paper/2506.19382/) (2025)
- [Causality is Key for Interpretability Claims to Generalise](https://mlsanity.com/paper/2602.16698/) (2026)
- [Wasserstein Distances, Neuronal Entanglement, and Sparsity](https://mlsanity.com/paper/2405.15756/) (2024)
- [Unboxing the Black Box: Mechanistic Interpretability for Algorithmic Understanding of Neural Networks](https://mlsanity.com/paper/2511.19265/) (2025)
- [Expand Neurons, Not Parameters](https://mlsanity.com/paper/2510.04500/) (2025)
- [Negative Pre-activations Differentiate Syntax](https://mlsanity.com/paper/2509.24198/) (2025)
- [Towards Integrated Alignment](https://mlsanity.com/paper/2508.06592/) (2025)

## Abstract (arXiv)

In some neural networks, individual neurons correspond to natural ``features'' in the input. Such monosemantic neurons are of great help in interpretability studies, as they can be cleanly understood. In this work we report preliminary attempts to engineer monosemanticity in toy models. We find that models can be made more monosemantic without increasing the loss by just changing which local minimum the training process finds. More monosemantic loss minima have moderate negative biases, and we are able to use this fact to engineer highly monosemantic models. We are able to mechanistically interpret these models, including the residual polysemantic neurons, and uncover a simple yet surprising algorithm. Finally, we find that providing models with more neurons per layer makes the models more monosemantic, albeit at increased computational cost. These findings point to a number of new questions and avenues for engineering monosemanticity, which we intend to study these in future work.