---
title: An exactly solvable model for emergence and scaling laws in the multitask sparse parity problem
url: https://www.ml-quant.com/papers/arxiv/2404.17563/
site: ML-Quant (https://www.ml-quant.com)
updated: 2026-09-26
license: Summaries CC BY 4.0; links go to the original sources
index: https://www.ml-quant.com/llms.txt
identifier: arXiv:2404.17563
source_url: https://arxiv.org/abs/2404.17563
featured: 2024-10-17
citations: 19
topic: ML & AI Methods
---


# An exactly solvable model for emergence and scaling laws in the multitask sparse parity problem

A new framework represents each new ability in deep learning models as a basis function, providing analytic expressions for the emergence of new skills and scaling laws of the loss with various factors.

- Source: https://arxiv.org/abs/2404.17563
- Identifier: arXiv:2404.17563
- Released: 2024-04-26
- First featured: Quant Letter No. 70 (2024-10-17): https://www.ml-quant.com/issues/2024-10-17/
- Citations (Semantic Scholar): 19
- Published in: Neural Information Processing Systems
- Topic: ML & AI Methods

## Related

- [Transformers are SSMs: Generalized Models and Efficient Algorithms Through Structured State Space Duality](https://www.ml-quant.com/papers/arxiv/2405.21060/): The research identifies a link between state-space models and Transformers in deep learning, leading to the creation of a faster language modeling architecture, Mamba-2.
- [SOAP: Improving and Stabilizing Shampoo using Adam](https://www.ml-quant.com/papers/arxiv/2409.11321/): A new algorithm, SOAP, enhances the computational efficiency of the Shampoo preconditioning method in deep learning tasks, reducing iterations and time, with an online implementation available.
- [Edge Directionality Improves Learning on Heterophilic Graphs](https://www.ml-quant.com/papers/arxiv/2305.10498/): The study presents Directed Graph Neural Network (Dir-GNN), a new deep learning framework for directed graphs that surpasses traditional models in heterophilic benchmarks.
- [TabR: Tabular Deep Learning Meets Nearest Neighbors](https://www.ml-quant.com/papers/arxiv/2307.14338/): Tabular DL Meets Nearest Neighbors: TabR, a new deep learning model for tabular data, outperforms existing models by using a k-Nearest-Neighbors-like component for better predictions.
- [Modular Duality in Deep Learning](https://www.ml-quant.com/papers/arxiv/2410.21265/): The article presents a new theory of modular dualization for general neural networks, providing a theoretical basis for fast and scalable training algorithms, potentially leading to a new generation of optimizers for neural architectures.
- [Mechanistic Design and Scaling of Hybrid Architectures](https://www.ml-quant.com/papers/arxiv/2403.17844/): The creation of deep learning architectures can be streamlined using an end-to-end design pipeline, which uses synthetic tasks to predict scaling laws and find optimal architectures.