Large language models (LLMs) are becoming bigger to boost performance. However, little is known about how explainability is affected by this trend. This work explores LIME explanations for DeBERTaV3 models of four different sizes on natural language inference (NLI) and zero-shot classification (ZSC) tasks. We evaluate the explanations based on their faithfulness to the models' internal decision processes and their plausibility, i.e. their agreement with human explanations. The key finding is that increased model size does not correlate with plausibility despite improved model performance, suggesting a misalignment between the LIME explanations and the models' internal processes as model size increases. Our results further suggest limitations regarding faithfulness metrics in NLI contexts.
In the recent years, the dual-encoder vision-language models (\eg CLIP) have achieved remarkable text-to-image retrieval performance. However, we discover that these models usually results in very different retrievals for a pair of paraphrased queries. Such behavior might render the retrieval system less predictable and lead to user frustration. In this work, we consider the task of paraphrased text-to-image retrieval where a model aims to return similar results given a pair of paraphrased queries. To start with, we collect a dataset of paraphrased image descriptions to facilitate quantitative evaluation for this task. We then hypothesize that the undesired behavior of existing dual-encoder model is due to their text towers which are trained on image-sentence pairs and lack the ability to capture the semantic similarity between paraphrased queries. To improve on this, we investigate multiple strategies for training a dual-encoder model starting from a language model pretrained on a large text corpus. Compared to public dual-encoder models such as CLIP and OpenCLIP, the model trained with our best adaptation strategy achieves a significantly higher ranking similarity for paraphrased queries while maintaining similar zero-shot classification and retrieval accuracy.
Source-free domain adaptation (SFDA) aims to adapt a source model trained on a fully-labeled source domain to a related but unlabeled target domain. While the source model is a key avenue for acquiring target pseudolabels, the generated pseudolabels may exhibit source bias. In the conventional SFDA pipeline, a large data (e.g. ImageNet) pre-trained feature extractor is used to initialize the source model at the start of source training, and subsequently discarded. Despite having diverse features important for generalization, the pre-trained feature extractor can overfit to the source data distribution during source training and forget relevant target domain knowledge. Rather than discarding this valuable knowledge, we introduce an integrated framework to incorporate pre-trained networks into the target adaptation process. The proposed framework is flexible and allows us to plug modern pre-trained networks into the adaptation process to leverage their stronger representation learning capabilities. For adaptation, we propose the Co-learn algorithm to improve target pseudolabel quality collaboratively through the source model and a pre-trained feature extractor. Building on the recent success of the vision-language model CLIP in zero-shot image recognition, we present an extension Co-learn++ to further incorporate CLIP's zero-shot classification decisions. We evaluate on 4 benchmark datasets and include more challenging scenarios such as open-set, partial-set and open-partial SFDA. Experimental results demonstrate that our proposed strategy improves adaptation performance and can be successfully integrated with existing SFDA methods. Project code is available at https://github.com/zwenyu/colearn-plus.
Pre-trained vision-language models (VLMs), exemplified by CLIP, demonstrate remarkable adaptability across zero-shot classification tasks without additional training. However, their performance diminishes in the presence of domain shifts. In this study, we introduce CLIP Adaptation duRing Test-Time (CLIPArTT), a fully test-time adaptation (TTA) approach for CLIP, which involves automatic text prompts construction during inference for their use as text supervision. Our method employs a unique, minimally invasive text prompt tuning process, wherein multiple predicted classes are aggregated into a single new text prompt, used as \emph{pseudo label} to re-classify inputs in a transductive manner. Additionally, we pioneer the standardization of TTA benchmarks (e.g., TENT) in the realm of VLMs. Our findings demonstrate that, without requiring additional transformations nor new trainable modules, CLIPArTT enhances performance dynamically across non-corrupted datasets such as CIFAR-100, corrupted datasets like CIFAR-100-C and ImageNet-C, alongside synthetic datasets such as VisDA-C. This research underscores the potential for improving VLMs' adaptability through novel test-time strategies, offering insights for robust performance across varied datasets and environments. The code can be found at: https://github.com/dosowiechi/CLIPArTT.git
There are a thousand ways to caption an image. Contrastive Language Pretraining (CLIP) on the other hand, works by mapping an image and its caption to a single vector -- limiting how well CLIP-like models can represent the diverse ways to describe an image. In this work, we introduce Llip, Latent Language Image Pretraining, which models the diversity of captions that could match an image. Llip's vision encoder outputs a set of visual features that are mixed into a final representation by conditioning on information derived from the text. We show that Llip outperforms non-contextualized baselines like CLIP and SigLIP on a variety of tasks even with large-scale encoders. Llip improves zero-shot classification by an average of 2.9% zero-shot classification benchmarks with a ViT-G/14 encoder. Specifically, Llip attains a zero-shot top-1 accuracy of 83.5% on ImageNet outperforming a similarly sized CLIP by 1.4%. We also demonstrate improvement on zero-shot retrieval on MS-COCO by 6.0%. We provide a comprehensive analysis of the components introduced by the method and demonstrate that Llip leads to richer visual representations.
State space models and Mamba-based models have been increasingly applied across various domains, achieving state-of-the-art performance. This technical report introduces the first attempt to train a transferable Mamba model utilizing contrastive language-image pretraining (CLIP). We have trained Mamba models of varying sizes and undertaken comprehensive evaluations of these models on 26 zero-shot classification datasets and 16 out-of-distribution (OOD) datasets. Our findings reveal that a Mamba model with 67 million parameters is on par with a 307 million-parameter Vision Transformer (ViT) model in zero-shot classification tasks, highlighting the parameter efficiency of Mamba models. In tests of OOD generalization, Mamba-based models exhibit exceptional performance in conditions of OOD image contrast or when subjected to high-pass filtering. However, a Hessian analysis indicates that Mamba models feature a sharper and more non-convex landscape compared to ViT-based models, making them more challenging to train. The source code is available at https://github.com/raytrun/mamba-clip.
In recent years, zero-shot learning has attracted the focus of many researchers, due to its flexibility and generality. Many approaches have been proposed to achieve the zero-shot classification of the point clouds for 3D object understanding, following the schema of CLIP. However, in the real world, the point clouds could be extremely sparse, dramatically limiting the effectiveness of the 3D point cloud encoders, and resulting in the misalignment of point cloud features and text embeddings. To the point cloud encoders to fit the extremely sparse point clouds without re-running the pre-training procedure which could be time-consuming and expensive, in this work, we propose an unsupervised model adaptation approach to enhance the point cloud encoder for the extremely sparse point clouds. We propose a novel fused-cross attention layer that expands the pre-trained self-attention layer with additional learnable tokens and attention blocks, which effectively modifies the point cloud features while maintaining the alignment between point cloud features and text embeddings. We also propose a complementary learning-based self-distillation schema that encourages the modified features to be pulled apart from the irrelevant text embeddings without overfitting the feature space to the observed text embeddings. Extensive experiments demonstrate that the proposed approach effectively increases the zero-shot capability on extremely sparse point clouds, and overwhelms other state-of-the-art model adaptation approaches.
Social anxiety represents a prevalent challenge in modern society, affecting individuals across personal and professional spheres. Left unaddressed, this condition can yield substantial negative consequences, impacting social interactions and performance. Further understanding its diverse physical and emotional symptoms becomes pivotal for comprehensive diagnosis and tailored therapeutic interventions. This study analyze prevalence and frequency of social anxiety symptoms taken from Mayo Clinic, exploring diverse human experiences from utilizing a large Reddit dataset dedicated to this issue. Leveraging these platforms, the research aims to extract insights and examine a spectrum of physical and emotional symptoms linked to social anxiety disorder. Upholding ethical considerations, the study maintains strict user anonymity within the dataset. By employing a novel approach, the research utilizes BART-based multi-label zero-shot classification to identify and measure symptom prevalence and significance in the form of probability score for each symptom under consideration. Results uncover distinctive patterns: "Trembling" emerges as a prevalent physical symptom, while emotional symptoms like "Fear of being judged negatively" exhibit high frequencies. These findings offer insights into the multifaceted nature of social anxiety, aiding clinical practices and interventions tailored to its diverse expressions.
Multi-modal foundation models such as CLIP have showcased impressive zero-shot capabilities. However, their applicability in resource-constrained environments is limited due to their large number of parameters and high inference time. While existing approaches have scaled down the entire CLIP architecture, we focus on training smaller variants of the image encoder, which suffices for efficient zero-shot classification. The use of synthetic data has shown promise in distilling representations from larger teachers, resulting in strong few-shot and linear probe performance. However, we find that this approach surprisingly fails in true zero-shot settings when using contrastive losses. We identify the exploitation of spurious features as being responsible for poor generalization between synthetic and real data. However, by using the image feature-based L2 distillation loss, we mitigate these problems and train students that achieve zero-shot performance which on four domain-specific datasets is on-par with a ViT-B/32 teacher model trained on DataCompXL, while featuring up to 92% fewer parameters.
Recent open-world 3D representation learning methods using Vision-Language Models (VLMs) to align 3D point cloud with image-text information have shown superior 3D zero-shot performance. However, CAD-rendered images for this alignment often lack realism and texture variation, compromising alignment robustness. Moreover, the volume discrepancy between 3D and 2D pretraining datasets highlights the need for effective strategies to transfer the representational abilities of VLMs to 3D learning. In this paper, we present OpenDlign, a novel open-world 3D model using depth-aligned images generated from a diffusion model for robust multimodal alignment. These images exhibit greater texture diversity than CAD renderings due to the stochastic nature of the diffusion model. By refining the depth map projection pipeline and designing depth-specific prompts, OpenDlign leverages rich knowledge in pre-trained VLM for 3D representation learning with streamlined fine-tuning. Our experiments show that OpenDlign achieves high zero-shot and few-shot performance on diverse 3D tasks, despite only fine-tuning 6 million parameters on a limited ShapeNet dataset. In zero-shot classification, OpenDlign surpasses previous models by 8.0% on ModelNet40 and 16.4% on OmniObject3D. Additionally, using depth-aligned images for multimodal alignment consistently enhances the performance of other state-of-the-art models.
Vision-language models enable open-world classification of objects without the need for any retraining. While this zero-shot paradigm marks a significant advance, even today's best models exhibit skewed performance when objects are dissimilar from their typical depiction. Real world objects such as pears appear in a variety of forms -- from diced to whole, on a table or in a bowl -- yet standard VLM classifiers map all instances of a class to a \it{single vector based on the class label}. We argue that to represent this rich diversity within a class, zero-shot classification should move beyond a single vector. We propose a method to encode and account for diversity within a class using inferred attributes, still in the zero-shot setting without retraining. We find our method consistently outperforms standard zero-shot classification over a large suite of datasets encompassing hierarchies, diverse object states, and real-world geographic diversity, as well finer-grained datasets where intra-class diversity may be less prevalent. Importantly, our method is inherently interpretable, offering faithful explanations for each inference to facilitate model debugging and enhance transparency. We also find our method scales efficiently to a large number of attributes to account for diversity -- leading to more accurate predictions for atypical instances. Finally, we characterize a principled trade-off between overall and worst class accuracy, which can be tuned via a hyperparameter of our method. We hope this work spurs further research into the promise of zero-shot classification beyond a single class vector for capturing diversity in the world, and building transparent AI systems without compromising performance.
This study is part of the debate on the efficiency of large versus small language models for text classification by prompting.We assess the performance of small language models in zero-shot text classification, challenging the prevailing dominance of large models.Across 15 datasets, our investigation benchmarks language models from 77M to 40B parameters using different architectures and scoring functions. Our findings reveal that small models can effectively classify texts, getting on par with or surpassing their larger counterparts.We developed and shared a comprehensive open-source repository that encapsulates our methodologies. This research underscores the notion that bigger isn't always better, suggesting that resource-efficient small models may offer viable solutions for specific data classification challenges.
The high volume and rapid evolution of content on social media present major challenges for studying the stance of social media users. In this work, we develop a two stage stance labeling method that utilizes the user-hashtag bipartite graph and the user-user interaction graph. In the first stage, a simple and efficient heuristic for stance labeling uses the user-hashtag bipartite graph to iteratively update the stance association of user and hashtag nodes via a label propagation mechanism. This set of soft labels is then integrated with the user-user interaction graph to train a graph neural network (GNN) model using semi-supervised learning. We evaluate this method on two large-scale datasets containing tweets related to climate change from June 2021 to June 2022 and gun control from January 2022 to January 2023. Our experiments demonstrate that enriching text-based embeddings of users with network information from the user interaction graph using our semi-supervised GNN method outperforms both classifiers trained on user textual embeddings and zero-shot classification using LLMs such as GPT4. We discuss the need for integrating nuanced understanding from social science with the scalability of computational methods to better understand how polarization on social media occurs for divisive issues such as climate change and gun control.
In this paper, we consider the problem of visual representation learning for computational pathology, by exploiting large-scale image-text pairs gathered from public resources, along with the domain-specific knowledge in pathology. Specifically, we make the following contributions: (i) We curate a pathology knowledge tree that consists of 50,470 informative attributes for 4,718 diseases requiring pathology diagnosis from 32 human tissues. To our knowledge, this is the first comprehensive structured pathology knowledge base; (ii) We develop a knowledge-enhanced visual-language pretraining approach, where we first project pathology-specific knowledge into latent embedding space via a language model, and use it to guide the visual representation learning; (iii) We conduct thorough experiments to validate the effectiveness of our proposed components, demonstrating significant performance improvement on various downstream tasks, including cross-modal retrieval, zero-shot classification on pathology patches, and zero-shot tumor subtyping on whole slide images (WSIs).
Multimodal pre-trained models, such as CLIP, are popular for zero-shot classification due to their open-vocabulary flexibility and high performance. However, vision-language models, which compute similarity scores between images and class labels, are largely black-box, with limited interpretability, risk for bias, and inability to discover new visual concepts not written down. Moreover, in practical settings, the vocabulary for class names and attributes of specialized concepts will not be known, preventing these methods from performing well on images uncommon in large-scale vision-language datasets. To address these limitations, we present a novel method that discovers interpretable yet discriminative sets of attributes for visual recognition. We introduce an evolutionary search algorithm that uses a large language model and its in-context learning abilities to iteratively mutate a concept bottleneck of attributes for classification. Our method produces state-of-the-art, interpretable fine-grained classifiers. We outperform the latest baselines by 18.4% on five fine-grained iNaturalist datasets and by 22.2% on two KikiBouba datasets, despite the baselines having access to privileged information about class names.
Neural Radiance Fields (NeRFs) have emerged as a standard framework for representing 3D scenes and objects, introducing a novel data type for information exchange and storage. Concurrently, significant progress has been made in multimodal representation learning for text and image data. This paper explores a novel research direction that aims to connect the NeRF modality with other modalities, similar to established methodologies for images and text. To this end, we propose a simple framework that exploits pre-trained models for NeRF representations alongside multimodal models for text and image processing. Our framework learns a bidirectional mapping between NeRF embeddings and those obtained from corresponding images and text. This mapping unlocks several novel and useful applications, including NeRF zero-shot classification and NeRF retrieval from images or text.
Vision-Language Models (VLMs) have demonstrated impressive performance on zero-shot classification, i.e. classification when provided merely with a list of class names. In this paper, we tackle the case of zero-shot classification in the presence of unlabeled data. We leverage the graph structure of the unlabeled data and introduce ZLaP, a method based on label propagation (LP) that utilizes geodesic distances for classification. We tailor LP to graphs containing both text and image features and further propose an efficient method for performing inductive inference based on a dual solution and a sparsification step. We perform extensive experiments to evaluate the effectiveness of our method on 14 common datasets and show that ZLaP outperforms the latest related works. Code: https://github.com/vladan-stojnic/ZLaP
In NLP, zero-shot classification (ZSC) is the task of assigning labels to textual data without any labeled examples for the target classes. A common method for ZSC is to fine-tune a language model on a Natural Language Inference (NLI) dataset and then use it to infer the entailment between the input document and the target labels. However, this approach faces certain challenges, particularly for languages with limited resources. In this paper, we propose an alternative solution that leverages dictionaries as a source of data for ZSC. We focus on Luxembourgish, a low-resource language spoken in Luxembourg, and construct two new topic relevance classification datasets based on a dictionary that provides various synonyms, word translations and example sentences. We evaluate the usability of our dataset and compare it with the NLI-based approach on two topic classification tasks in a zero-shot manner. Our results show that by using the dictionary-based dataset, the trained models outperform the ones following the NLI-based approach for ZSC. While we focus on a single low-resource language in this study, we believe that the efficacy of our approach can also transfer to other languages where such a dictionary is available.
Recent advancements in open vocabulary models, like CLIP, have notably advanced zero-shot classification and segmentation by utilizing natural language for class-specific embeddings. However, most research has focused on improving model accuracy through prompt engineering, prompt learning, or fine-tuning with limited labeled data, thereby overlooking the importance of refining the class descriptors. This paper introduces a new approach to text-supervised semantic segmentation using supervision by a large language model (LLM) that does not require extra training. Our method starts from an LLM, like GPT-3, to generate a detailed set of subclasses for more accurate class representation. We then employ an advanced text-supervised semantic segmentation model to apply the generated subclasses as target labels, resulting in diverse segmentation results tailored to each subclass's unique characteristics. Additionally, we propose an assembly that merges the segmentation maps from the various subclass descriptors to ensure a more comprehensive representation of the different aspects in the test images. Through comprehensive experiments on three standard benchmarks, our method outperforms traditional text-supervised semantic segmentation methods by a marked margin.
Despite that deep learning methods have achieved remarkable performance in pathology image classification, they heavily rely on labeled data, demanding extensive human annotation efforts. In this study, we present a novel human annotation-free method for pathology image classification by leveraging pre-trained Vision-Language Models (VLMs). Without human annotation, pseudo labels of the training set are obtained by utilizing the zero-shot inference capabilities of VLM, which may contain a lot of noise due to the domain shift between the pre-training data and the target dataset. To address this issue, we introduce VLM-CPL, a novel approach based on consensus pseudo labels that integrates two noisy label filtering techniques with a semi-supervised learning strategy. Specifically, we first obtain prompt-based pseudo labels with uncertainty estimation by zero-shot inference with the VLM using multiple augmented views of an input. Then, by leveraging the feature representation ability of VLM, we obtain feature-based pseudo labels via sample clustering in the feature space. Prompt-feature consensus is introduced to select reliable samples based on the consensus between the two types of pseudo labels. By rejecting low-quality pseudo labels, we further propose High-confidence Cross Supervision (HCS) to learn from samples with reliable pseudo labels and the remaining unlabeled samples. Experimental results showed that our method obtained an accuracy of 87.1% and 95.1% on the HPH and LC25K datasets, respectively, and it largely outperformed existing zero-shot classification and noisy label learning methods. The code is available at https://github.com/lanfz2000/VLM-CPL.