Identifying Multi-modal Knowledge Neurons in Pretrained Transformers via Two-stage Filtering

Recent advances in large language models (LLMs) have led to the development of multimodal LLMs (MLLMs) in the fields of natural language processing (NLP) and computer vision. Although these models allow for integrated visual and language understanding, they present challenges such as opaque internal processing and the generation of hallucinations and misinformation. Therefore, there is a need for a method to clarify the location of knowledge in MLLMs. In this study, we propose a method to identify neurons associated with specific knowledge using MiniGPT-4, a Transformer-based MLLM. Specifically, we extract knowledge neurons through two stages: activation differences filtering using inpainting and gradient-based filtering using GradCAM. Experiments on the image caption generation task using the MS COCO 2017 dataset, BLEU, ROUGE, and BERTScore quantitative evaluation, and qualitative evaluation using an activation heatmap showed that our method is able to locate knowledge with higher accuracy than existing methods. This study contributes to the visualization and explainability of knowledge in MLLMs and shows the potential for future knowledge editing and control.

RecTable: Fast Modeling Tabular Data with Rectified Flow

Score-based or diffusion models generate high-quality tabular data, surpassing GAN-based and VAE-based models. However, these methods require substantial training time. In this paper, we introduce RecTable, which uses the rectified flow modeling, applied in such as text-to-image generation and text-to-video generation. RecTable features a simple architecture consisting of a few stacked gated linear unit blocks. Additionally, our training strategies are also simple, incorporating a mixed-type noise distribution and a logit-normal timestep distribution. Our experiments demonstrate that RecTable achieves competitive performance compared to the several state-of-the-art diffusion and score-based models while reducing the required training time. Our code is available at https://github.com/fmp453/rectable.

Erasing with Precision: Evaluating Specific Concept Erasure from Text-to-Image Generative Models

Studies have been conducted to prevent specific concepts from being generated from pretrained text-to-image generative models, achieving concept erasure in various ways. However, the performance evaluation of these studies is still largely reliant on visualization, with the superiority of studies often determined by human subjectivity. The metrics of quantitative evaluation also vary, making comprehensive comparisons difficult. We propose EraseEval, an evaluation method that differs from previous evaluation methods in that it involves three fundamental evaluation criteria: (1) How well does the prompt containing the target concept be reflected, (2) To what extent the concepts related to the erased concept can reduce the impact of the erased concept, and (3) Whether other concepts are preserved. These criteria are evaluated and integrated into a single metric, such that a lower score is given if any of the evaluations are low, leading to a more robust assessment. We experimentally evaluated baseline concept erasure methods, organized their characteristics, and identified challenges with them. Despite being fundamental evaluation criteria, some concept erasure methods failed to achieve high scores, which point toward future research directions for concept erasure methods. Our code is available at https://github.com/fmp453/erase-eval.

Erasing Concepts from Text-to-Image Diffusion Models with Few-shot Unlearning

Generating images from text has become easier because of the scaling of diffusion models and advancements in the field of vision and language. These models are trained using vast amounts of data from the Internet. Hence, they often contain undesirable content such as copyrighted material. As it is challenging to remove such data and retrain the models, methods for erasing specific concepts from pre-trained models have been investigated. We propose a novel concept-erasure method that updates the text encoder using few-shot unlearning in which a few real images are used. The discussion regarding the generated images after erasing a concept has been lacking. While there are methods for specifying the transition destination for concepts, the validity of the specified concepts is unclear. Our method implicitly achieves this by transitioning to the latent concepts inherent in the model or the images. Our method can erase a concept within 10 s, making concept erasure more accessible than ever before. Implicitly transitioning to related concepts leads to more natural concept erasure. We applied the proposed method to various concepts and confirmed that concept erasure can be achieved tens to hundreds of times faster than with current methods. By varying the parameters to be updated, we obtained results suggesting that, like previous research, knowledge is primarily accumulated in the feed-forward networks of the text encoder.

Application of frozen large-scale models to multimodal task-oriented dialogue

In this study, we use the existing Large Language Models ENnhanced to See Framework (LENS Framework) to test the feasibility of multimodal task-oriented dialogues. The LENS Framework has been proposed as a method to solve computer vision tasks without additional training and with fixed parameters of pre-trained models. We used the Multimodal Dialogs (MMD) dataset, a multimodal task-oriented dialogue benchmark dataset from the fashion field, and for the evaluation, we used the ChatGPT-based G-EVAL, which only accepts textual modalities, with arrangements to handle multimodal data. Compared to Transformer-based models in previous studies, our method demonstrated an absolute lift of 10.8% in fluency, 8.8% in usefulness, and 5.2% in relevance and coherence. The results show that using large-scale models with fixed parameters rather than using models trained on a dataset from scratch improves performance in multimodal task-oriented dialogues. At the same time, we show that Large Language Models (LLMs) are effective for multimodal task-oriented dialogues. This is expected to lead to efficient applications to existing systems.

ResBit: Residual Bit Vector for Categorical Values

The one-hot vector has long been widely used in machine learning as a simple and generic method for representing discrete data. However, this method increases the number of dimensions linearly with the categorical data to be represented, which is problematic from the viewpoint of spatial computational complexity in deep learning, which requires a large amount of data. Recently, Analog Bits, a method for representing discrete data as a sequence of bits, was proposed on the basis of the high expressiveness of diffusion models. However, since the number of category types to be represented in a generation task is not necessarily at a power of two, there is a discrepancy between the range that Analog Bits can represent and the range represented as category data. If such a value is generated, the problem is that the original category value cannot be restored. To address this issue, we propose Residual Bit Vector (ResBit), which is a hierarchical bit representation. Although it is a general-purpose representation method, in this paper, we treat it as numerical data and show that it can be used as an extension of Analog Bits using Table Residual Bit Diffusion (TRBD), which is incorporated into TabDDPM, a tabular data generation method. We experimentally confirmed that TRBD can generate diverse and high-quality data from small-scale table data to table data containing diverse category values faster than TabDDPM. Furthermore, we show that ResBit can also serve as an alternative to the one-hot vector by utilizing ResBit for conditioning in GANs and as a label expression in image classification.