Challenge: Prior research has focused on optimizing general-purpose large language models to downstream tasks . however, these approaches inherently introduce data dependency, which hinders generalization and reusability.
Approach: They propose an algorithm that localizes the most task-sensitive attention heads and prunes by restricting attention training updates to these heads, thereby reducing alignment costs.
Outcome: The proposed algorithm achieves 2% performance improvement over baselines on three tasks while localizing the most task-sensitive attention heads.

Similar Papers

Mitigating Catastrophic Forgetting in Large Language Models with Forgetting-aware Pruning (2025.emnlp-main)

Copied to clipboard

Challenge: Recent advances in large language models (LLMs) have shown impressive capabilities in various downstream tasks but typically face Catastrophic Forgetting (CF) during fine-tuning.
Approach: They propose a pruning-based approach to balance CF and downstream task performance by integrating the ratio of the task vector to pre-trained model parameters into the pruning criteria.
Outcome: The proposed pruning-based approach limits CF to just 0.25% while maintaining 99.67% accuracy on downstream tasks.
Attention Speaks Volumes: Localizing and Mitigating Bias in Language Models (2025.acl-long)

Copied to clipboard

Challenge: ambiguous comparative prompting frameworks generate biased content when confronted with ambiguously prompts . ambiguity in LLMs can manifest in various forms, including reinforcing stereotypes, displaying gender bias, or demonstrating preferential treatment towards specific demographic groups.
Approach: They propose a technique to localize bias to specific layers of an LLM by analyzing attention scores and then reduce bias by scaling attention in these biased layers.
Outcome: The proposed technique reduces bias by scaling attention in biased layers.
Enhancing Translation Ability of Large Language Models by Leveraging Task-Related Layers (2024.lrec-main)

Copied to clipboard

Challenge: Experimental validation shows that adjusting task-related layers significantly improves performance on translation tasks while maintaining stability and accuracy on other tasks.
Approach: They propose to adjust task-related layers in large models to better harness their machine translation capabilities by revealing the structure and characteristics of attention weights through singular value decomposition.
Outcome: The proposed method reduces computational resource consumption and catastrophic forgetting while maintaining stability and accuracy on other tasks.
Controllable Memorization in LLMs via Weight Pruning (2025.emnlp-main)

Copied to clipboard

Challenge: Existing studies have focused on mitigating memorization, but the deliberate control of memorisation has been underexplored.
Approach: They propose a gradient-based weight pruning framework to control memorization rates in large language models by fine-grained control over pruning parameters.
Outcome: The proposed framework enables models to suppress or enhance memorization based on application-specific requirements.
From Local to Global: Revisiting Structured Pruning Paradigms for Large Language Models (2026.acl-long)

Copied to clipboard

Challenge: Structured pruning is a practical approach to deploying large language models (LLMs) but it fails to capitalize on modest task-specific calibration signals, causing limited downstream gains.
Approach: They propose a method that removes attention heads and MLP channels using loss-based important scores . they use perplexity for language modeling and a margin-based objective for decision-style tasks .
Outcome: The proposed method lowers perplexity and improves accuracy at higher sparsity . it also stabilizes accuracy and mitigates perxity collapse without fine-tuning .
Attention Basin: Why Contextual Position Matters in Large Language Models (2026.acl-long)

Copied to clipboard

Challenge: Large Language Models (LLMs) are sensitive to the contextual position of information in input.
Approach: They introduce Attention-Driven Reranking (AttnRank) which estimates a model’s intrinsic positional attention preferences using a small calibration set and reorders retrieved documents or few-shot examples to align the most salient content with these high-attention positions.
Outcome: Experiments on multi-hop QA and few-shot in-context learning tasks show that AttnRank achieves substantial improvements across 10 large language models of varying architectures and scales, without modifying model parameters or training procedures.
Fortify the Shortest Stave in Attention: Enhancing Context Awareness of Large Language Models for Effective Tool Use (2024.acl-long)

Copied to clipboard

Challenge: In this paper, we demonstrate that an inherent waveform pattern in the attention allocation of large language models significantly affects their performance in tasks demanding a high degree of context awareness.
Approach: They propose a method that compensates an attention trough with an attention peak by a process to enhance the model's awareness to various contextual positions.
Outcome: The proposed method improves the performance of a 7B model on the largest tool-use benchmark, comparable to that of GPT-4.
Toward Inclusive Language Models: Sparsity-Driven Calibration for Systematic and Interpretable Mitigation of Social Biases in LLMs (2025.findings-emnlp)

Copied to clipboard

Challenge: a new method to mitigate stereotypical bias in large language models is needed . inherent biases from training on vast Internet datasets can amplify harmful stereotypes .
Approach: They propose a method to identify stereotypical bias in decoder-only transformer models . they apply a localization mechanism that correlates internal activations with a new Context Influence score .
Outcome: The proposed method reduces stereotypical biases on BBQ, StereoSet, and CrowS-Pairs while improving reasoning performance on MMLU by 10%.
Gradient-based Intra-attention Pruning on Pre-trained Language Models (2023.acl-long)

Copied to clipboard

Challenge: Pre-trained language models are computationally expensive and slow in inference due to their large sizes.
Approach: They propose a structured pruning method which combines pruning with knowledge distillation to yield highly effective models.
Outcome: The proposed method outperforms other pruning methods in sparsity regimes while maintaining 93% 99% performance.
On the Limitations of Language-targeted Pruning: Investigating the Calibration Language Impact in Multilingual LLM Pruning (2026.tacl-1)

Copied to clipboard

Challenge: Recent advances in large language model pruning have shown high predictive performance in post-training settings.
Approach: They conduct an empirical study on the performance and internal representation changes associated with pruning multilingual models for monolingual applications.
Outcome: The proposed pruning methods retain perplexity and yield high signal-to-noise ratios, but not consistently improve downstream tasks.

What is GenGO?

GenGO is an NLP powered publication search system. It currenctly indexes 30k+ papers from ACL Anthology, and implements multi-aspect summarization, semantic search, and more!

Information

About
Limitations