Recent vision-language foundation models still frequently produce outputs misaligned with their inputs, evidenced by object hallucination in captioning and prompt misalignment in the text-to-image generation model. Recent studies have explored methods for identifying misaligned elements, aiming not only to enhance interpretability but also to improve model performance. However, current approaches primarily rely on large foundation models in a zero-shot manner or fine-tuned models with human annotations, which limits scalability due to significant computational costs. This work proposes a novel approach, dubbed CLIP4DM, for detecting dense misalignments from pre-trained CLIP, specifically focusing on pinpointing misaligned words between image and text. We carefully revamp the gradient-based attribution computation method, enabling negative gradient of individual text tokens to indicate misalignment. We also propose F-CLIPScore, which aggregates misaligned attributions with a global alignment score. We evaluate our method on various dense misalignment detection benchmarks, covering various image and text domains and misalignment types. Our method demonstrates state-of-the-art performance among zero-shot models and competitive performance with fine-tuned models while maintaining superior efficiency. Our qualitative examples show that our method has a unique strength to detect entity-level objects, intangible objects, and attributes that can not be easily detected for existing works. We conduct ablation studies and analyses to highlight the strengths and limitations of our approach. Our code is publicly available at https://github.com/naver-ai/CLIP4DM.

Vision-language models (VLMs) embed aligned image-text pairs into a joint space but often rely on deterministic embeddings, assuming a one-to-one correspondence between images and texts. This oversimplifies real-world relationships, which are inherently many-to-many, with multiple captions describing a single image and vice versa. We introduce Probabilistic Language-Image Pre-training (ProLIP), the first probabilistic VLM pre-trained on a billion-scale image-text dataset using only probabilistic objectives, achieving a strong zero-shot capability (e.g., 74.6% ImageNet zero-shot accuracy with ViT-B/16). ProLIP efficiently estimates uncertainty by an "uncertainty token" without extra parameters. We also introduce a novel inclusion loss that enforces distributional inclusion relationships between image-text pairs and between original and masked inputs. Experiments demonstrate that, by leveraging uncertainty estimates, ProLIP benefits downstream tasks and aligns with intuitive notions of uncertainty, e.g., shorter texts being more uncertain and more general inputs including specific ones. Utilizing text uncertainties, we further improve ImageNet accuracy from 74.6% to 75.8% (under a few-shot setting), supporting the practical advantages of our probabilistic approach. The code is available at https://github.com/naver-ai/prolip

Continuously learning a variety of audio-video semantics over time is crucial for audio-related reasoning tasks in our ever-evolving world. However, this is a nontrivial problem and poses two critical challenges: sparse spatio-temporal correlation between audio-video pairs and multimodal correlation overwriting that forgets audio-video relations. To tackle this problem, we propose a new continual audio-video pre-training method with two novel ideas: (1) Localized Patch Importance Scoring: we introduce a multimodal encoder to determine the importance score for each patch, emphasizing semantically intertwined audio-video patches. (2) Replay-guided Correlation Assessment: to reduce the corruption of previously learned audiovisual knowledge due to drift, we propose to assess the correlation of the current patches on the past steps to identify the patches exhibiting high correlations with the past steps. Based on the results from the two ideas, we perform probabilistic patch selection for effective continual audio-video pre-training. Experimental validation on multiple benchmarks shows that our method achieves a 3.69%p of relative performance gain in zero-shot retrieval tasks compared to strong continual learning baselines, while reducing memory consumption by ~45%.

Extracting semantic representations from mobile user interfaces (UI) and using the representations for designers' decision-making processes have shown the potential to be effective computational design support tools. Current approaches rely on machine learning models trained on small-sized mobile UI datasets to extract semantic vectors and use screenshot-to-screenshot comparison to retrieve similar-looking UIs given query screenshots. However, the usability of these methods is limited because they are often not open-sourced and have complex training pipelines for practitioners to follow, and are unable to perform screenshot set-to-set (i.e., app-to-app) retrieval. To this end, we (1) employ visual models trained with large web-scale images and test whether they could extract a UI representation in a zero-shot way and outperform existing specialized models, and (2) use mathematically founded methods to enable app-to-app retrieval and design consistency analysis. Our experiments show that our methods not only improve upon previous retrieval models but also enable multiple new applications.

Recent studies have proposed unified user modeling frameworks that leverage user behavior data from various applications. Many of them benefit from utilizing users' behavior sequences as plain texts, representing rich information in any domain or system without losing generality. Hence, a question arises: Can language modeling for user history corpus help improve recommender systems? While its versatile usability has been widely investigated in many domains, its applications to recommender systems still remain underexplored. We show that language modeling applied directly to task-specific user histories achieves excellent results on diverse recommendation tasks. Also, leveraging additional task-agnostic user histories delivers significant performance benefits. We further demonstrate that our approach can provide promising transfer learning capabilities for a broad spectrum of real-world recommender systems, even on unseen domains and services.

We present a comparative study on how and why contrastive learning (CL) and masked image modeling (MIM) differ in their representations and in their performance of downstream tasks. In particular, we demonstrate that self-supervised Vision Transformers (ViTs) have the following properties: (1) CL trains selfattentions to capture longer-range global patterns than MIM, such as the shape of an object, especially in the later layers of the ViT architecture. This CL property helps ViTs linearly separate images in their representation spaces. However, it also makes the self-attentions collapse into homogeneity for all query tokens and heads. Such homogeneity of self-attention reduces the diversity of representations, worsening scalability and dense prediction performance. Since low-and high-frequency information respectively represent shapes and textures, CL is more shape-oriented and MIM more texture-oriented. Upon these analyses, we find that CL and MIM can complement each other and observe that even the simplest harmonization can help leverage the advantages of both methods. Contrastive Learning (CL) (He et al., 2020; Chen et al., 2020a;b; 2021) has been the most popular self-supervised learning methods until recently. It aims to learn the invariant semantics of two random views (Tian et al., 2020a;b) by making global projections of representations similar for positive samples and dissimilar for negative samples. Since CL exploits the globally projected representations to contrast each other, it can be deemed as an "image-level" self-supervised learning approach. Deviating from CL, masked image modeling (MIM) (Bao et al., 2022; Xie et al., 2022b; He et al., 2022) has risen as a strong competitor of CL in the era of Vision Transformers (ViTs) (Dosovitskiy et al., 2021) with its impressive performances of downstream tasks. MIM trains ViTs by reconstructing the correct semantics of masked input patches. Unlike CL, it learns the semantics of patch tokens and this can be deemed as a "token-level" self-supervised learning approach. Since MIM outperforms CL in fine-tuning accuracy, it may appear prima facie as a more effective pre-training method than CL. However, a different trend is observed for linear probing accuracy with CL outperforming MIM (See Figure 1). For further exposition on CL and MIM, we refer the reader to Appendix B. Then, which method--CL or MIM--should we use for the self-supervised learning of ViTs? Although both methods are widely used, little is known about what they learn.

Generated synthetic data in medical research can substitute privacy and security-sensitive data with a large-scale curated dataset, reducing data collection and annotation costs. As part of this effort, we propose UniXGen, a unified chest X-ray and report generation model, with the following contributions. First, we design a unified model for bidirectional chest X-ray and report generation by adopting a vector quantization method to discretize chest X-rays into discrete visual tokens and formulating both tasks as sequence generation tasks. Second, we introduce several special tokens to generate chest X-rays with specific views that can be useful when the desired views are unavailable. Furthermore, UniXGen can flexibly take various inputs from single to multiple views to take advantage of the additional findings available in other X-ray views. We adopt an efficient transformer for computational and memory efficiency to handle the long-range input sequence of multi-view chest X-rays with high resolution and long paragraph reports. In extensive experiments, we show that our unified model has a synergistic effect on both generation tasks, as opposed to training only the task-specific models. We also find that view-specific special tokens can distinguish between different views and properly generate specific views even if they do not exist in the dataset, and utilizing multi-view chest X-rays can faithfully capture the abnormal findings in the additional X-rays. The source code is publicly available at: https://github.com/ttumyche/UniXGen.

Vision-and-Language Pretraining (VLP) has improved performance on various joint vision-and-language downstream tasks. Current approaches for VLP heavily rely on image feature extraction processes, most of which involve region supervisions (e.g., object detection) and the convolutional architecture (e.g., ResNet). Although disregarded in the literature, we find it problematic in terms of both (1) efficiency/speed, that simply extracting input features requires much more computation than the actual multimodal interaction steps; and (2) expressive power, as it is upper bounded to the expressive power of the visual encoder and its predefined visual vocabulary. In this paper, we present a minimal VLP model, Vision-and-Language Transformer (ViLT), monolithic in the sense that processing of visual inputs is drastically simplified to just the same convolution-free manner that we process textual inputs. We show that ViLT is up to 60 times faster than previous VLP models, yet with competitive or better downstream task performance.

Learning compact discrete representations of data is itself a key task in addition to facilitating subsequent processing. It is also relevant to meta-learning since a latent representation shared across relevant tasks enables a model to adapt to new tasks quickly. In this paper, we present a method for learning a stochastic encoder that yields discrete p-way codes of length d by maximizing the mutual information between representations and labels. We show that previous loss functions for deep metric learning are approximations to this information-theoretic objective function. Our model, Discrete InfoMax Codes (DIMCO), learns to produce a short representation of data that can be used to classify classes with few labeled examples. Our analysis shows that using shorter codes reduces overfitting in the context of few-shot classification. Experiments show that DIMCO requires less memory (i.e., code length) for performance similar to previous methods and that our method is particularly effective when the training dataset is small.

Without relevant human priors, neural networks may learn uninterpretable features. We propose Dynamics of Attention for Focus Transition (DAFT) as a human prior for machine reasoning. DAFT is a novel method that regularizes attention-based reasoning by modelling it as a continuous dynamical system using neural ordinary differential equations. As a proof of concept, we augment a state-of-the-art visual reasoning model with DAFT. Our experiments reveal that applying DAFT yields similar performance to the original model while using fewer reasoning steps, showing that it implicitly learns to skip unnecessary steps. We also propose a new metric, Total Length of Transition (TLT), which represents the effective reasoning step size by quantifying how much a given model's focus drifts while reasoning about a question. We show that adding DAFT results in lower TLT, demonstrating that our method indeed obeys the human prior towards shorter reasoning paths in addition to producing more interpretable attention maps.