Related papers: Memorization in In-Context Learning

Memorization in In-Context Learning

URL: http://arxiv.org/abs/2408.11546v2
Date: Sun, 27 Oct 2024 18:04:58 GMT
Title: Memorization in In-Context Learning
Authors: Shahriar Golchin, Mihai Surdeanu, Steven Bethard, Eduardo Blanco, Ellen Riloff,
Abstract summary: In-context learning (ICL) has proven to be an effective strategy for improving the performance of large language models (LLMs) with no additional training. This study is the first to show how ICL surfaces memorized training data and to explore the correlation between this memorization and performance.
Score: 42.218016081867376
License: http://creativecommons.org/licenses/by/4.0/
Abstract: In-context learning (ICL) has proven to be an effective strategy for improving the performance of large language models (LLMs) with no additional training. However, the exact mechanism behind this performance improvement remains unclear. This study is the first to show how ICL surfaces memorized training data and to explore the correlation between this memorization and performance on downstream tasks across various ICL regimes: zero-shot, few-shot, and many-shot. Our most notable findings include: (1) ICL significantly surfaces memorization compared to zero-shot learning in most cases; (2) demonstrations, without their labels, are the most effective element in surfacing memorization; (3) ICL improves performance when the surfaced memorization in few-shot regimes reaches a high level (about 40%); and (4) there is a very strong correlation between performance and memorization in ICL when it outperforms zero-shot learning. Overall, our study uncovers memorization as a new factor impacting ICL, raising an important question: to what extent do LLMs truly generalize from demonstrations in ICL, and how much of their success is due to memorization?

Related papers

Instruction Tuning Vs. In-Context Learning: Revisiting Large Language Models in Few-Shot Computational Social Science [0.1499944454332829]
We evaluate the classification performance of large language models (LLMs) using in-context learning (ICL) and instruction tuning (IT) ICL offers a rapid alternative for task adaptation by learning from examples without explicit gradient updates. Our research highlights the significant advantages of ICL in handling CSS tasks in few-shot settings.
arXiv Detail & Related papers (2024-09-23T02:43:08Z)
ICLEval: Evaluating In-Context Learning Ability of Large Language Models [68.7494310749199]
In-Context Learning (ICL) is a critical capability of Large Language Models (LLMs) as it empowers them to comprehend and reason across interconnected inputs. Existing evaluation frameworks primarily focus on language abilities and knowledge, often overlooking the assessment of ICL ability. We introduce the ICLEval benchmark to evaluate the ICL abilities of LLMs, which encompasses two key sub-abilities: exact copying and rule learning.
arXiv Detail & Related papers (2024-06-21T08:06:10Z)
Investigating the Pre-Training Dynamics of In-Context Learning: Task Recognition vs. Task Learning [99.05401042153214]
In-context learning (ICL) is potentially attributed to two major abilities: task recognition (TR) and task learning (TL) We take the first step by examining the pre-training dynamics of the emergence of ICL. We propose a simple yet effective method to better integrate these two abilities for ICL at inference time.
arXiv Detail & Related papers (2024-06-20T06:37:47Z)
Many-Shot In-Context Learning [58.395589302800566]
Large language models (LLMs) excel at few-shot in-context learning (ICL) We observe significant performance gains across a wide variety of generative and discriminative tasks. Unlike few-shot learning, many-shot learning is effective at overriding pretraining biases.
arXiv Detail & Related papers (2024-04-17T02:49:26Z)
DimCL: Dimensional Contrastive Learning For Improving Self-Supervised Learning [40.25324481491231]
This paper proposes a strategy of performing contrastive learning along the dimensional direction instead of along the batch direction. DimCL aims to enhance the feature diversity, and it can serve as a regularizer to prior SSL frameworks.
arXiv Detail & Related papers (2023-09-21T05:12:55Z)
Investigating the Learning Behaviour of In-context Learning: A Comparison with Supervised Learning [67.25698169440818]
Large language models (LLMs) have shown remarkable capacity for in-context learning (ICL) We train the same LLMs with the same demonstration examples via ICL and supervised learning (SL), respectively, and investigate their performance under label perturbations. First, we find that gold labels have significant impacts on the downstream in-context performance, especially for large language models. Second, when comparing with SL, we show empirically that ICL is less sensitive to label perturbations than SL, and ICL gradually attains comparable performance to SL as the model size increases.
arXiv Detail & Related papers (2023-07-28T09:03:19Z)
In-Context Learning Learns Label Relationships but Is Not Conventional Learning [60.891931501449726]
There is currently no consensus about how in-context learning (ICL) ability of Large Language Models works. We provide novel insights into how ICL leverages label information, revealing both capabilities and limitations. Our experiments show that ICL predictions almost always depend on in-context labels and that ICL can learn truly novel tasks in-context.
arXiv Detail & Related papers (2023-07-23T16:54:41Z)
Decoupled Contrastive Learning [23.25775900388382]
We identify a noticeable negative-positive-coupling (NPC) effect in the widely used cross-entropy (InfoNCE) loss. By properly addressing the NPC effect, we reach a decoupled contrastive learning (DCL) objective function. Our approach achieves $66.9%$ ImageNet top-1 accuracy using batch size 256 within 200 epochs pre-training, outperforming its baseline SimCLR by $5.1%$.
arXiv Detail & Related papers (2021-10-13T16:38:43Z)

This list is automatically generated from the titles and abstracts of the papers in this site.