Locate-then-Merge: Neuron-Level Parameter Fusion for Mitigating Catastrophic Forgetting in Multimodal LLMs

• URL: http://arxiv.org/abs/2505.16703v1
• Date: Thu, 22 May 2025 14:04:43 GMT
• Title: Locate-then-Merge: Neuron-Level Parameter Fusion for Mitigating Catastrophic Forgetting in Multimodal LLMs
• Authors: Zeping Yu, Sophia Ananiadou,
• Abstract summary: We propose a training-free parameter fusion framework that locates important parameters and then selectively merges them.<n>We introduce Neuron-Fusion, a neuron-level strategy that preserves the influence of neurons with large parameter shifts.<n>Experiments on 13 benchmarks across both language and visual tasks show that Neuron-Fusion consistently outperforms existing model merging methods.
• Score: 19.472889262384818
• License: http://creativecommons.org/licenses/by/4.0/
• Abstract: Although multimodal large language models (MLLMs) have achieved impressive performance, the multimodal instruction tuning stage often causes catastrophic forgetting of the base LLM's language ability, even in strong models like Llama3. To address this, we propose Locate-then-Merge, a training-free parameter fusion framework that first locates important parameters and then selectively merges them. We further introduce Neuron-Fusion, a neuron-level strategy that preserves the influence of neurons with large parameter shifts--neurons likely responsible for newly acquired visual capabilities--while attenuating the influence of neurons with smaller changes that likely encode general-purpose language skills. This design enables better retention of visual adaptation while mitigating language degradation. Experiments on 13 benchmarks across both language and visual tasks show that Neuron-Fusion consistently outperforms existing model merging methods. Further analysis reveals that our method effectively reduces context hallucination in generation.

Related papers

• LLaVA-NeuMT: Selective Layer-Neuron Modulation for Efficient Multilingual Multimodal Translation [12.51212639515934]
  LLaVA-NeuMT is a novel framework that explicitly models language-specific and language-agnostic representations to mitigate multilingual interference.<n>Our approach consists of a layer selection mechanism that identifies the most informative layers for different language pairs.<n>We conduct extensive experiments on the M3-Multi30K and M3-AmbigCaps datasets, demonstrating that LLaVA-NeuMT, while fine-tuning only 40% of the model parameters, surpasses full fine-tuning approaches.
  arXiv  Detail & Related papers  (2025-07-25T04:23:24Z)
• The Emergence of Abstract Thought in Large Language Models Beyond Any Language [95.50197866832772]
  Large language models (LLMs) function effectively across a diverse range of languages.<n>Preliminary studies observe that the hidden activations of LLMs often resemble English, even when responding to non-English prompts.<n>Recent results show strong multilingual performance, even surpassing English performance on specific tasks in other languages.
  arXiv  Detail & Related papers  (2025-06-11T16:00:54Z)
• Probing Neural Topology of Large Language Models [15.34202977968525]
  We introduce graph probing, a method for uncovering the functional connectivity topology of LLM neurons.<n>We find a universal predictability of next-token prediction performance using only neural topology.<n>This predictability is robust even when retaining just 1% of neuron connections or probing models after only 8 pretraining steps.
  arXiv  Detail & Related papers  (2025-06-01T14:57:03Z)
• How does Alignment Enhance LLMs' Multilingual Capabilities? A Language Neurons Perspective [64.79894853375478]
  We propose a new finer-grained neuron identification algorithm, which detects language neurons(including language-specific neurons and language-related neurons) and language-agnostic neurons.<n>Based on the distributional characteristics of different types of neurons, we divide the LLMs' internal process for multilingual inference into four parts.<n>We systematically analyze the models before and after alignment with a focus on different types of neurons.
  arXiv  Detail & Related papers  (2025-05-27T17:59:52Z)
• The Unreasonable Effectiveness of Model Merging for Cross-Lingual Transfer in LLMs [54.59207567677249]
  Large language models (LLMs) still struggle across tasks outside of high-resource languages.<n>In this work, we investigate cross-lingual transfer to lower-resource languages where task-specific post-training data is scarce.
  arXiv  Detail & Related papers  (2025-05-23T20:28:31Z)
• Mechanistic Understanding and Mitigation of Language Confusion in English-Centric Large Language Models [49.09746599881631]
  We present the first mechanistic interpretability study of language confusion.<n>We show that confusion points (CPs) are central to this phenomenon.<n>We show that editing a small set of critical neurons, identified via comparative analysis with multilingual-tuned models, substantially mitigates confusion.
  arXiv  Detail & Related papers  (2025-05-22T11:29:17Z)
• EMMA: Efficient Visual Alignment in Multi-Modal LLMs [56.03417732498859]
  EMMA is a lightweight cross-modality module designed to efficiently fuse visual and textual encodings. EMMA boosts performance across multiple tasks by up to 9.3% while significantly improving robustness against hallucinations.
  arXiv  Detail & Related papers  (2024-10-02T23:00:31Z)
• LANDeRMT: Detecting and Routing Language-Aware Neurons for Selectively Finetuning LLMs to Machine Translation [43.26446958873554]
  Large language models (LLMs) have shown promising results in multilingual translation even with limited bilingual supervision. Recent advancements in large language models (LLMs) have shown promising results in multilingual translation even with limited bilingual supervision. LandeRMT is a framework that selectively finetunes LLMs to textbfMachine textbfTranslation with diverse translation training data.
  arXiv  Detail & Related papers  (2024-09-29T02:39:42Z)
• Let's Focus on Neuron: Neuron-Level Supervised Fine-tuning for Large Language Model [43.107778640669544]
  Large Language Models (LLMs) are composed of neurons that exhibit various behaviors and roles. Recent studies have revealed that not all neurons are active across different datasets. We introduce Neuron-Level Fine-Tuning (NeFT), a novel approach that refines the granularity of parameter training down to the individual neuron.
  arXiv  Detail & Related papers  (2024-03-18T09:55:01Z)
• How do Large Language Models Handle Multilingualism? [81.15060972112563]
  This study explores how large language models (LLMs) handle multilingualism. LLMs initially understand the query, converting multilingual inputs into English for task-solving. In the intermediate layers, they employ English for thinking and incorporate multilingual knowledge with self-attention and feed-forward structures.
  arXiv  Detail & Related papers  (2024-02-29T02:55:26Z)
• NeuroPrune: A Neuro-inspired Topological Sparse Training Algorithm for Large Language Models [35.10729451729596]
  Transformer-based Language Models have become ubiquitous in Natural Language Processing (NLP) However, expensive training as well as inference remains a significant impediment to their widespread applicability. Inspired by brain neuronal networks, we explore sparsity approaches through the lens of network topology.
  arXiv  Detail & Related papers  (2024-02-28T22:21:47Z)
• Language-Specific Neurons: The Key to Multilingual Capabilities in Large Language Models [117.20416338476856]
  Large language models (LLMs) demonstrate remarkable multilingual capabilities without being pre-trained on specially curated multilingual parallel corpora. We propose a novel detection method, language activation probability entropy (LAPE), to identify language-specific neurons within LLMs. Our findings indicate that LLMs' proficiency in processing a particular language is predominantly due to a small subset of neurons.
  arXiv  Detail & Related papers  (2024-02-26T09:36:05Z)
• Contextual Feature Extraction Hierarchies Converge in Large Language Models and the Brain [12.92793034617015]
  We show that as large language models (LLMs) achieve higher performance on benchmark tasks, they become more brain-like. We also show the importance of contextual information in improving model performance and brain similarity.
  arXiv  Detail & Related papers  (2024-01-31T08:48:35Z)
• Dependency-based Mixture Language Models [53.152011258252315]
  We introduce the Dependency-based Mixture Language Models. In detail, we first train neural language models with a novel dependency modeling objective. We then formulate the next-token probability by mixing the previous dependency modeling probability distributions with self-attention.
  arXiv  Detail & Related papers  (2022-03-19T06:28:30Z)

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

This site does not guarantee the quality of this site (including all information) and is not responsible for any consequences.