P4Q: Learning to Prompt for Quantization in Visual-language Models

Large-scale pre-trained Vision-Language Models (VLMs) have gained prominence in various visual and multimodal tasks, yet the deployment of VLMs on downstream application platforms remains challenging due to their prohibitive requirements of training samples and computing resources. Fine-tuning and quantization of VLMs can substantially reduce the sample and computation costs, which are in urgent need. There are two prevailing paradigms in quantization, Quantization-Aware Training (QAT) can effectively quantize large-scale VLMs but incur a huge training cost, while low-bit Post-Training Quantization (PTQ) suffers from a notable performance drop. We propose a method that balances fine-tuning and quantization named ``Prompt for Quantization'' (P4Q), in which we design a lightweight architecture to leverage contrastive loss supervision to enhance the recognition performance of a PTQ model. Our method can effectively reduce the gap between image features and text features caused by low-bit quantization, based on learnable prompts to reorganize textual representations and a low-bit adapter to realign the distributions of image and text features. We also introduce a distillation loss based on cosine similarity predictions to distill the quantized model using a full-precision teacher. Extensive experimental results demonstrate that our P4Q method outperforms prior arts, even achieving comparable results to its full-precision counterparts. For instance, our 8-bit P4Q can theoretically compress the CLIP-ViT/B-32 by 4 $\times$ while achieving 66.94\% Top-1 accuracy, outperforming the learnable prompt fine-tuned full-precision model by 2.24\% with negligible additional parameters on the ImageNet dataset.

TAAT: Think and Act from Arbitrary Texts in Text2Motion

Text2Motion aims to generate human motions from texts. Existing datasets rely on the assumption that texts include action labels (such as "walk, bend, and pick up"), which is not flexible for practical scenarios. This paper redefines this problem with a more realistic assumption that the texts are arbitrary. Specifically, arbitrary texts include existing action texts composed of action labels (e.g., A person walks and bends to pick up something), and introduce scene texts without explicit action labels (e.g., A person notices his wallet on the ground ahead). To bridge the gaps between this realistic setting and existing datasets, we expand the action texts on the HumanML3D dataset to more scene texts, thereby creating a new HumanML3D++ dataset including arbitrary texts. In this challenging dataset, we benchmark existing state-of-the-art methods and propose a novel two-stage framework to extract action labels from arbitrary texts by the Large Language Model (LLM) and then generate motions from action labels. Extensive experiments are conducted under different application scenarios to validate the effectiveness of the proposed framework on existing and proposed datasets. The results indicate that Text2Motion in this realistic setting is very challenging, fostering new research in this practical direction. Our dataset and code will be released.

Cross-Level Distillation and Feature Denoising for Cross-Domain Few-Shot Classification

The conventional few-shot classification aims at learning a model on a large labeled base dataset and rapidly adapting to a target dataset that is from the same distribution as the base dataset. However, in practice, the base and the target datasets of few-shot classification are usually from different domains, which is the problem of cross-domain few-shot classification. We tackle this problem by making a small proportion of unlabeled images in the target domain accessible in the training stage. In this setup, even though the base data are sufficient and labeled, the large domain shift still makes transferring the knowledge from the base dataset difficult. We meticulously design a cross-level knowledge distillation method, which can strengthen the ability of the model to extract more discriminative features in the target dataset by guiding the network's shallow layers to learn higher-level information. Furthermore, in order to alleviate the overfitting in the evaluation stage, we propose a feature denoising operation which can reduce the feature redundancy and mitigate overfitting. Our approach can surpass the previous state-of-the-art method, Dynamic-Distillation, by 5.44% on 1-shot and 1.37% on 5-shot classification tasks on average in the BSCD-FSL benchmark. The implementation code will be available at https://github.com/jarucezh/cldfd.

Self-Enhancement Improves Text-Image Retrieval in Foundation Visual-Language Models

The emergence of cross-modal foundation models has introduced numerous approaches grounded in text-image retrieval. However, on some domain-specific retrieval tasks, these models fail to focus on the key attributes required. To address this issue, we propose a self-enhancement framework, A^{3}R, based on the CLIP-ViT/G-14, one of the largest cross-modal models. First, we perform an Attribute Augmentation strategy to enrich the textual description for fine-grained representation before model learning. Then, we propose an Adaption Re-ranking method to unify the representation space of textual query and candidate images and re-rank candidate images relying on the adapted query after model learning. The proposed framework is validated to achieve a salient improvement over the baseline and other teams' solutions in the cross-modal image retrieval track of the 1st foundation model challenge without introducing any additional samples. The code is available at \url{https://github.com/CapricornGuang/A3R}.

AttriCLIP: A Non-Incremental Learner for Incremental Knowledge Learning

Continual learning aims to enable a model to incrementally learn knowledge from sequentially arrived data. Previous works adopt the conventional classification architecture, which consists of a feature extractor and a classifier. The feature extractor is shared across sequentially arrived tasks or classes, but one specific group of weights of the classifier corresponding to one new class should be incrementally expanded. Consequently, the parameters of a continual learner gradually increase. Moreover, as the classifier contains all historical arrived classes, a certain size of the memory is usually required to store rehearsal data to mitigate classifier bias and catastrophic forgetting. In this paper, we propose a non-incremental learner, named AttriCLIP, to incrementally extract knowledge of new classes or tasks. Specifically, AttriCLIP is built upon the pre-trained visual-language model CLIP. Its image encoder and text encoder are fixed to extract features from both images and text. Text consists of a category name and a fixed number of learnable parameters which are selected from our designed attribute word bank and serve as attributes. As we compute the visual and textual similarity for classification, AttriCLIP is a non-incremental learner. The attribute prompts, which encode the common knowledge useful for classification, can effectively mitigate the catastrophic forgetting and avoid constructing a replay memory. We evaluate our AttriCLIP and compare it with CLIP-based and previous state-of-the-art continual learning methods in realistic settings with domain-shift and long-sequence learning. The results show that our method performs favorably against previous state-of-the-arts. The implementation code can be available at https://github.com/bhrqw/AttriCLIP.

Few-Shot Learning with Visual Distribution Calibration and Cross-Modal Distribution Alignment

Pre-trained vision-language models have inspired much research on few-shot learning. However, with only a few training images, there exist two crucial problems: (1) the visual feature distributions are easily distracted by class-irrelevant information in images, and (2) the alignment between the visual and language feature distributions is difficult. To deal with the distraction problem, we propose a Selective Attack module, which consists of trainable adapters that generate spatial attention maps of images to guide the attacks on class-irrelevant image areas. By messing up these areas, the critical features are captured and the visual distributions of image features are calibrated. To better align the visual and language feature distributions that describe the same object class, we propose a cross-modal distribution alignment module, in which we introduce a vision-language prototype for each class to align the distributions, and adopt the Earth Mover's Distance (EMD) to optimize the prototypes. For efficient computation, the upper bound of EMD is derived. In addition, we propose an augmentation strategy to increase the diversity of the images and the text prompts, which can reduce overfitting to the few-shot training images. Extensive experiments on 11 datasets demonstrate that our method consistently outperforms prior arts in few-shot learning. The implementation code will be available at https://github.com/bhrqw/SADA.

Rethinking the Number of Shots in Robust Model-Agnostic Meta-Learning

Robust Model-Agnostic Meta-Learning (MAML) is usually adopted to train a meta-model which may fast adapt to novel classes with only a few exemplars and meanwhile remain robust to adversarial attacks. The conventional solution for robust MAML is to introduce robustness-promoting regularization during meta-training stage. With such a regularization, previous robust MAML methods simply follow the typical MAML practice that the number of training shots should match with the number of test shots to achieve an optimal adaptation performance. However, although the robustness can be largely improved, previous methods sacrifice clean accuracy a lot. In this paper, we observe that introducing robustness-promoting regularization into MAML reduces the intrinsic dimension of clean sample features, which results in a lower capacity of clean representations. This may explain why the clean accuracy of previous robust MAML methods drops severely. Based on this observation, we propose a simple strategy, i.e., increasing the number of training shots, to mitigate the loss of intrinsic dimension caused by robustness-promoting regularization. Though simple, our method remarkably improves the clean accuracy of MAML without much loss of robustness, producing a robust yet accurate model. Extensive experiments demonstrate that our method outperforms prior arts in achieving a better trade-off between accuracy and robustness. Besides, we observe that our method is less sensitive to the number of fine-tuning steps during meta-training, which allows for a reduced number of fine-tuning steps to improve training efficiency.

Anti-Retroactive Interference for Lifelong Learning

Humans can continuously learn new knowledge. However, machine learning models suffer from drastic dropping in performance on previous tasks after learning new tasks. Cognitive science points out that the competition of similar knowledge is an important cause of forgetting. In this paper, we design a paradigm for lifelong learning based on meta-learning and associative mechanism of the brain. It tackles the problem from two aspects: extracting knowledge and memorizing knowledge. First, we disrupt the sample's background distribution through a background attack, which strengthens the model to extract the key features of each task. Second, according to the similarity between incremental knowledge and base knowledge, we design an adaptive fusion of incremental knowledge, which helps the model allocate capacity to the knowledge of different difficulties. It is theoretically analyzed that the proposed learning paradigm can make the models of different tasks converge to the same optimum. The proposed method is validated on the MNIST, CIFAR100, CUB200 and ImageNet100 datasets.

Confidence Dimension for Deep Learning based on Hoeffding Inequality and Relative Evaluation

Research on the generalization ability of deep neural networks (DNNs) has recently attracted a great deal of attention. However, due to their complex architectures and large numbers of parameters, measuring the generalization ability of specific DNN models remains an open challenge. In this paper, we propose to use multiple factors to measure and rank the relative generalization of DNNs based on a new concept of confidence dimension (CD). Furthermore, we provide a feasible framework in our CD to theoretically calculate the upper bound of generalization based on the conventional Vapnik-Chervonenk dimension (VC-dimension) and Hoeffding's inequality. Experimental results on image classification and object detection demonstrate that our CD can reflect the relative generalization ability for different DNNs. In addition to full-precision DNNs, we also analyze the generalization ability of binary neural networks (BNNs), whose generalization ability remains an unsolved problem. Our CD yields a consistent and reliable measure and ranking for both full-precision DNNs and BNNs on all the tasks.

Associative Adversarial Learning Based on Selective Attack

A human's attention can intuitively adapt to corrupted areas of an image by recalling a similar uncorrupted image they have previously seen. This observation motivates us to improve the attention of adversarial images by considering their clean counterparts. To accomplish this, we introduce Associative Adversarial Learning (AAL) into adversarial learning to guide a selective attack. We formulate the intrinsic relationship between attention and attack (perturbation) as a coupling optimization problem to improve their interaction. This leads to an attention backtracking algorithm that can effectively enhance the attention's adversarial robustness. Our method is generic and can be used to address a variety of tasks by simply choosing different kernels for the associative attention that select other regions for a specific attack. Experimental results show that the selective attack improves the model's performance. We show that our method improves the recognition accuracy of adversarial training on ImageNet by 8.32% compared with the baseline. It also increases object detection mAP on PascalVOC by 2.02% and recognition accuracy of few-shot learning on miniImageNet by 1.63%.