English
Related papers

Related papers: Superposition Yields Robust Neural Scaling

200 papers

Large language models (LLMs) continue to face challenges in reliably solving reasoning tasks, particularly those that require precise rule following, as often found in mathematical reasoning. This paper introduces a novel neurosymbolic…

Machine Learning · Computer Science 2025-11-19 Varun Dhanraj , Chris Eliasmith

Larger models learn tasks smaller models do not. What drives this phenomenon? We develop a simple phenomenological argument that power-law scaling already suggests that a larger model will be able to learn a part of the data distribution…

Representation learning that leverages large-scale labelled datasets, is central to recent progress in machine learning. Access to task relevant labels at scale is often scarce or expensive, motivating the need to learn from unlabelled…

Machine Learning · Computer Science 2022-02-14 Arna Ghosh , Arnab Kumar Mondal , Kumar Krishna Agrawal , Blake Richards

Data is the cornerstone of large language models (LLMs), but not all data is useful for model learning. Carefully selected data can better elicit the capabilities of LLMs with much less computational overhead. Most methods concentrate on…

Machine Learning · Computer Science 2024-07-12 Mingjia Yin , Chuhan Wu , Yufei Wang , Hao Wang , Wei Guo , Yasheng Wang , Yong Liu , Ruiming Tang , Defu Lian , Enhong Chen

Recent advances in large language models (LLMs) have been largely driven by scaling laws for individual models, which predict performance improvements as model parameters and data volume increase. However, the capabilities of any single LLM…

Machine Learning · Computer Science 2026-01-29 Dakuan Lu , Jiaqi Zhang , Cheng Yuan , Jiawei Shao , Xuelong Li

While scaling laws for Large Language Models (LLMs) traditionally focus on proxy metrics like pretraining loss, predicting downstream task performance has been considered unreliable. This paper challenges that view by proposing a direct…

Machine Learning · Computer Science 2025-12-10 Jakub Krajewski , Amitis Shidani , Dan Busbridge , Sam Wiseman , Jason Ramapuram

The superposition hypothesis states that single neurons may participate in representing multiple features in order for the neural network to represent more features than it has neurons. In neuroscience and AI, representational alignment…

Machine Learning · Computer Science 2025-11-14 André Longon , David Klindt , Meenakshi Khosla

Large Language Models (LLMs) have demonstrated remarkable capabilities, yet their scalability raises a critical question: Have we reached the scaling ceiling? This paper addresses this pivotal question by developing a unified theoretical…

Machine Learning · Computer Science 2024-12-24 Charles Luo

How do neural language models acquire a language's structure when trained for next-token prediction? We address this question by deriving theoretical scaling laws for neural network performance on synthetic datasets generated by the Random…

Machine Learning · Computer Science 2025-05-13 Francesco Cagnetta , Alessandro Favero , Antonio Sclocchi , Matthieu Wyart

The impressive linguistic abilities of large language models (LLMs) have recommended them as models of human sentence processing, with some conjecturing a positive 'quality-power' relationship (Wilcox et al., 2023), in which language…

Computation and Language · Computer Science 2025-05-20 Yi-Chien Lin , Hongao Zhu , William Schuler

Neural scaling laws describe how model performance improves as a power law with size, but existing work focuses on models above 100M parameters. The sub-20M regime -- where TinyML and edge AI operate -- remains unexamined. We train 90…

Machine Learning · Computer Science 2026-03-10 Mohammed Alnemari , Rizwan Qureshi , Nader Begrazadah

Recent theories suggest that Neural Scaling Laws arise whenever the task is linearly decomposed into power-law distributed units. Alternatively, scaling laws also emerge when data exhibit a hierarchically compositional structure, as is…

Machine Learning · Statistics 2025-05-13 Francesco Cagnetta , Hyunmo Kang , Matthieu Wyart

On a variety of tasks, the performance of neural networks predictably improves with training time, dataset size and model size across many orders of magnitude. This phenomenon is known as a neural scaling law. Of fundamental importance is…

Machine Learning · Statistics 2024-06-25 Blake Bordelon , Alexander Atanasov , Cengiz Pehlevan

Neural scaling laws approximate a language model's loss as a power-law function of parameter count $N$ and token count $D$. Following Chinchilla-style compute-optimal training, many studies fit scaling laws from runs performed under a fixed…

Machine Learning · Computer Science 2026-05-14 Joshua Shay Kricheli , Alexander Lawrence Reid , Soumajyoti Sarkar , Venkata Gandikota , Paulo Shakarian

Scaling law principles indicate a power-law correlation between loss and variables such as model size, dataset size, and computational resources utilized during training. These principles play a vital role in optimizing various aspects of…

Machine Learning · Computer Science 2024-04-08 Hui Su , Zhi Tian , Xiaoyu Shen , Xunliang Cai

Scaling laws aim to accurately predict model performance across different scales. Existing scaling-law studies almost exclusively rely on cross-entropy as the evaluation metric. However, cross-entropy provides only a partial view of…

Machine Learning · Computer Science 2025-10-24 Baoqing Yue , Jinyuan Zhou , Zixi Wei , Jingtao Zhan , Qingyao Ai , Yiqun Liu

Large language models (LLMs) are a promising venue for natural language understanding and generation. However, current LLMs are far from reliable: they are prone to generating non-factual information and, more crucially, to contradicting…

Computation and Language · Computer Science 2024-09-24 Diego Calanzone , Stefano Teso , Antonio Vergari

Prior work has shown that large language models (LLMs) often converge to accurate input embedding for numbers, based on sinusoidal representations. In this work, we quantify that these representations are in fact strikingly systematic, to…

We find that the cross-entropy loss curves of neural language models empirically adhere to a scaling law with learning rate (LR) annealing over training steps: $$L(s) = L_0 + A\cdot S_1^{-\alpha} - C\cdot S_2,$$ where $L(s)$ is the…

Computation and Language · Computer Science 2024-10-28 Howe Tissue , Venus Wang , Lu Wang

Superalignment, where humans act as weak supervisors for superhuman models, has become a crucial problem with the rapid development of Large Language Models (LLMs). Recent work has preliminarily studied this problem by using weak models to…

Computation and Language · Computer Science 2025-03-03 Wenkai Yang , Shiqi Shen , Guangyao Shen , Wei Yao , Yong Liu , Zhi Gong , Yankai Lin , Ji-Rong Wen