Related papers: Causal Estimation of Memorisation Profiles

While Large Language Models (LLMs) achieve remarkable performance through training on massive datasets, they can exhibit concerning behaviors such as verbatim reproduction of training data rather than true generalization. This memorization…

Modern language models are typically trained over subword sequences, but ultimately define probabilities over character-strings. Ideally, the choice of the tokeniser -- which maps character-strings to subwords -- should not affect the…

In imitation learning, an agent learns how to behave in an environment with an unknown cost function by mimicking expert demonstrations. Existing imitation learning algorithms typically involve solving a sequence of planning or…

Recent research demonstrated that training large language models involves memorization of a significant fraction of training data. Such memorization can lead to privacy violations when training on sensitive user data and thus motivates the…

Existing accounts of grokking explain the phenomena in terms of mechanistic frameworks such as circuit efficiency or lazy-to-rich transitions. However, despite a known dependence between grokking and model size, how model capacity shapes…

Large language models are shown to present privacy risks through memorization of training data, and several recent works have studied such risks for the pre-training phase. Little attention, however, has been given to the fine-tuning phase…

Deep learning models have a propensity for fitting the entire training set even with random labels, which requires memorization of every training sample. In this paper, we explore the memorization effect in adversarial training (AT) for…

We present a new abductive, probabilistic theory of plan recognition. This model differs from previous plan recognition theories in being centered around a model of plan execution: most previous methods have been based on plans as formal…

The learning curve expresses the error rate of a predictive modeling procedure as a function of the sample size of the training dataset. It typically is a decreasing, convex function with a positive limiting value. An estimate of the…

In recent years, model collapse has become a critical issue in language model training, making it essential to understand the underlying mechanisms driving this phenomenon. In this paper, we investigate recursive parametric model training…

Neural scaling laws have garnered significant interest due to their ability to predict model performance as a function of increasing parameters, data, and compute. In this work, we propose a simple statistical ansatz based on memorization…

Can we localize the weights and mechanisms used by a language model to memorize and recite entire paragraphs of its training data? In this paper, we show that while memorization is spread across multiple layers and model components,…

Predictive models are often introduced to decision-making tasks under the rationale that they improve performance over an existing decision-making policy. However, it is challenging to compare predictive performance against an existing…

Numerous recent works target to extend effective context length for language models and various methods, tasks and benchmarks exist to measure model's effective memorization length. However, through thorough investigations, we find…

Driven by vast and diverse textual data, large language models (LLMs) have demonstrated impressive performance across numerous natural language processing (NLP) tasks. Yet, a critical question persists: does their generalization arise from…

When we encounter a new person or place, we may easily encode it into our memories, or we may quickly forget it. Recent work finds that this likelihood of encoding a given entity - memorability - is highly consistent across viewers and…

Reasoning is a core capability of large language models, yet how multi-step reasoning is learned and executed remains unclear. We study this question in a controlled cellular-automata (1dCA) framework that excludes memorisation by using…

Deep learning algorithms are well-known to have a propensity for fitting the training data very well and often fit even outliers and mislabeled data points. Such fitting requires memorization of training data labels, a phenomenon that has…

Large language models memorize parts of their training data. Memorizing short snippets and facts is required to answer questions about the world and to be fluent in any language. But models have also been shown to reproduce long verbatim…

The statistical study of human memory requires large-scale experiments, involving many stimuli conditions and test subjects. While this approach has proven to be quite fruitful for meaningless material such as random lists of words,…