|
Canada-BC-NELSON Azienda Directories
|
Azienda News:
- Finding and Editing Multi-Modal Neurons in Pre-Trained Transformers
In this paper, we propose a novel method to identify key neurons for interpretability -- how multi-modal LLMs bridge visual and textual concepts for captioning Our method improves conventional works upon efficiency and applied range by removing needs of costly gradient computation
- Finding and Editing Multi-Modal Neurons in Pre-Trained . . . - GitHub
This repository is the official implementation of the paper "Finding and Editing Multi-Modal Neurons in Pre-Trained Transformers" (Findings of ACL 2024)
- Model Edit 一定要先 locate 再 edit？ - 知乎
TL,DR: 之前的方法都是先 locate 到知识存储在哪个参数，再进行 edit。 作者意外发现，二者并不是正相关关系，虽然确实效果很不错，但是效果好可能只是因为编辑的是浅层的 FFN，locate 并不会带来增益。
- LLMs之ROME：ROME的简介 (定位和编辑GPT中的事实 . . .
该研究提出了一种新方法来定位和编辑大型语言模型内部事实关联的知识表示，可以有效调试和修复模型中的具体事实错误，既 有助于理解这些庞大模型内部机制，也为应用提供了新的可能。 1、语言模型内的事实存储在哪里？ ：模型权重、语境 2、为什么要定位事实？ ：可解释性+修正错误知识 3、我们发现了什么？ ：事实关联可以沿着三个维度局部化、通过对单个 MLP 模块进行小的秩一改变可以改变单个事实关联
- Finding and Editing Multi-Modal Neurons in Pre-Trained Transformers
PDF | On Jan 1, 2024, Haowen Pan and others published Finding and Editing Multi-Modal Neurons in Pre-Trained Transformers | Find, read and cite all the research you need on ResearchGate
- Finding and Editing Multi-Modal Neurons in Pre-Trained Transformers
In this paper, we propose a novel method to identify key neurons for interpretability — how multi-modal LLMs bridge visual and textual concepts for captioning Our method improves conventional works upon efficiency and applied range by removing needs of costly gradient computation
- Locating and Editing Factual Associations in GPT - OpenReview
Our results confirm an important role for mid-layer feed-forward modules in storing factual associations and suggest that direct manipulation of computational mechanisms may be a feasible approach for model editing The code, dataset, visualizations, and an interactive demo notebook are available in the supplemental materials
- [2202. 05262] Locating and Editing Factual Associations in GPT
We analyze the storage and recall of factual associations in autoregressive transformer language models, finding evidence that these associations correspond to localized, directly-editable computations We first develop a causal intervention for identifying neuron activations that are decisive in a model's factual predictions
- Locating and Editing Factual Associations in GPT
We analyze the storage and recall of factual associations in autoregressive trans-former language models, finding evidence that these associations correspond to localized, directly-editable computations We first develop a causal intervention for identifying neuron activations that are decisive in a model’s factual predictions
- Where you edit is what you get: Text-guided image editing with region . . .
We summarize our contributions as follows: First, we propose a novel framework that enables stable training of multi-text image editing within one model Second, we adopt a region-based attention mechanism to ensure spatially-localized editing, in which we utilize the semantic properties of StyleGAN’s latent space
|
|