Analyzing limits for in-context learning
We examine limits of in-context learning (ICL) in transformer models trained from scratch, focusing on function approximation tasks as a controlled setting to uncover fundamental behaviors. While we show empirically that transformer models can generalize, approximating unseen classes of polynomial (non linear) functions, they cannot generalize beyond certain values. We provide both empirical and mathematical arguments explaining that these limitations stem from architectural components, namely layer normalization and the attention scoring function, softmax. Together, our findings reveal structural constraints on ICL that are often masked in more complex NLP tasks but that need to be understood to improve robustness and interpretability in transformer-based models.
Code (0)
등록된 구현이 없습니다.
Tasks
In-Context LearningMethods 이 논문이 사용한 방법론
Similar Papers 제목 키워드 기반
Divide-and-conquer methods for big data analysis
In the context of big data analysis, the divide-and-conquer methodology refers to a multiple-step process: first splitting a data set into several smaller ones; then analyzing each set separately; finally combining resul…
Analyzing Bias in Swiss Federal Supreme Court Judgments Using Facebook's Holistic Bias Dataset: Implications for Language Model Training
Natural Language Processing (NLP) is vital for computers to process and respond accurately to human language. However, biases in training data can introduce unfairness, especially in predicting legal judgment. This study…
Decision MakingLanguage ModelingLanguage ModellingTest-Time Adaptation via Many-Shot Prompting: Benefits, Limits, and Pitfalls
Test-time adaptation enables large language models (LLMs) to modify their behavior at inference without updating model parameters. A common approach is many-shot prompting, where large numbers of in-context learning (ICL…
Test-time AdaptationIndexing Context-Sensitive Reachability
Many context-sensitive data flow analyses can be formulated as a variant of the all-pairs Dyck-CFL reachability problem, which, in general, is of sub-cubic time complexity and quadratic space complexity. Such high comple…
On the Limits of Test-Time Compute: Sequential Reward Filtering for Better Inference
Test-time compute (TTC) has become an increasingly prominent paradigm for enhancing large language models (LLMs). Despite the empirical success of methods such as best-of-$n$ (BoN) sampling and sequential revision, their…