Why do LLaVA Vision-Language Models Reply to Images in English?

We uncover a surprising multilingual bias occurring in a popular class of multimodal vision-language models (VLMs). Including an image in the query to a LLaVA-style VLM significantly increases the likelihood of the model returning an English response, regardless of the language of the query. This paper investigates the causes of this loss with a two-pronged approach that combines extensive ablation of the design space with a mechanistic analysis of the models' internal representations of image and text inputs. Both approaches indicate that the issue stems in the language modelling component of the LLaVA model. Statistically, we find that switching the language backbone for a bilingual language model has the strongest effect on reducing this error. Mechanistically, we provide compelling evidence that visual inputs are not mapped to a similar space as text ones, and that intervening on intermediary attention layers can reduce this bias. Our findings provide important insights to researchers and engineers seeking to understand the crossover between multimodal and multilingual spaces, and contribute to the goal of developing capable and inclusive VLMs for non-English contexts.

LVLM-Intrepret: An Interpretability Tool for Large Vision-Language Models

In the rapidly evolving landscape of artificial intelligence, multi-modal large language models are emerging as a significant area of interest. These models, which combine various forms of data input, are becoming increasingly popular. However, understanding their internal mechanisms remains a complex task. Numerous advancements have been made in the field of explainability tools and mechanisms, yet there is still much to explore. In this work, we present a novel interactive application aimed towards understanding the internal mechanisms of large vision-language models. Our interface is designed to enhance the interpretability of the image patches, which are instrumental in generating an answer, and assess the efficacy of the language model in grounding its output in the image. With our application, a user can systematically investigate the model and uncover system limitations, paving the way for enhancements in system capabilities. Finally, we present a case study of how our application can aid in understanding failure mechanisms in a popular large multi-modal model: LLaVA.

Uncovering Bias in Large Vision-Language Models with Counterfactuals

With the advent of Large Language Models (LLMs) possessing increasingly impressive capabilities, a number of Large Vision-Language Models (LVLMs) have been proposed to augment LLMs with visual inputs. Such models condition generated text on both an input image and a text prompt, enabling a variety of use cases such as visual question answering and multimodal chat. While prior studies have examined the social biases contained in text generated by LLMs, this topic has been relatively unexplored in LVLMs. Examining social biases in LVLMs is particularly challenging due to the confounding contributions of bias induced by information contained across the text and visual modalities. To address this challenging problem, we conduct a large-scale study of text generated by different LVLMs under counterfactual changes to input images. Specifically, we present LVLMs with identical open-ended text prompts while conditioning on images from different counterfactual sets, where each set contains images which are largely identical in their depiction of a common subject (e.g., a doctor), but vary only in terms of intersectional social attributes (e.g., race and gender). We comprehensively evaluate the text produced by different LVLMs under this counterfactual generation setting and find that social attributes such as race, gender, and physical characteristics depicted in input images can significantly influence toxicity and the generation of competency-associated words.

Probing and Mitigating Intersectional Social Biases in Vision-Language Models with Counterfactual Examples

While vision-language models (VLMs) have achieved remarkable performance improvements recently, there is growing evidence that these models also posses harmful biases with respect to social attributes such as gender and race. Prior studies have primarily focused on probing such bias attributes individually while ignoring biases associated with intersections between social attributes. This could be due to the difficulty of collecting an exhaustive set of image-text pairs for various combinations of social attributes. To address this challenge, we employ text-to-image diffusion models to produce counterfactual examples for probing intserctional social biases at scale. Our approach utilizes Stable Diffusion with cross attention control to produce sets of counterfactual image-text pairs that are highly similar in their depiction of a subject (e.g., a given occupation) while differing only in their depiction of intersectional social attributes (e.g., race & gender). Through our over-generate-then-filter methodology, we produce SocialCounterfactuals, a high-quality dataset containing over 171k image-text pairs for probing intersectional biases related to gender, race, and physical characteristics. We conduct extensive experiments to demonstrate the usefulness of our generated dataset for probing and mitigating intersectional social biases in state-of-the-art VLMs.

Analyzing Zero-Shot Abilities of Vision-Language Models on Video Understanding Tasks

Foundational multimodal models pre-trained on large scale image-text pairs or video-text pairs or both have shown strong generalization abilities on downstream tasks. However unlike image-text models, pretraining video-text models is always not feasible due to the difficulty in collecting large-scale clean and aligned data, and exponential computational costs involved in the pretraining phase. Therefore, the pertinent question to ask is: Can image-text models be adapted to video tasks and is there any benefit to using these models over pretraining directly on videos? In this work, we focus on this question by proposing a detailed study on the generalization abilities of image-text models when evaluated on video understanding tasks in a zero-shot setting. We investigate 9 foundational image-text models on a diverse set of video tasks that include video action recognition (video AR), video retrieval (video RT), video question answering (video QA), video multiple choice (video MC) and video captioning (video CP). Our experiments show that image-text models exhibit impressive performance on video AR, video RT and video MC. Furthermore, they perform moderately on video captioning and poorly on video QA. These findings shed a light on the benefits of adapting foundational image-text models to an array of video tasks while avoiding the costly pretraining step.

ManagerTower: Aggregating the Insights of Uni-Modal Experts for Vision-Language Representation Learning

Two-Tower Vision-Language (VL) models have shown promising improvements on various downstream VL tasks. Although the most advanced work improves performance by building bridges between encoders, it suffers from ineffective layer-by-layer utilization of uni-modal representations and cannot flexibly exploit different levels of uni-modal semantic knowledge. In this work, we propose ManagerTower, a novel VL model architecture that gathers and combines the insights of pre-trained uni-modal experts at different levels. The managers introduced in each cross-modal layer can adaptively aggregate uni-modal semantic knowledge to facilitate more comprehensive cross-modal alignment and fusion. ManagerTower outperforms previous strong baselines both with and without Vision-Language Pre-training (VLP). With only 4M VLP data, ManagerTower achieves superior performances on various downstream VL tasks, especially 79.15% accuracy on VQAv2 Test-Std, 86.56% IR@1 and 95.64% TR@1 on Flickr30K. Code and checkpoints are available at https://github.com/LooperXX/ManagerTower.

Shifted and Squeezed 8-bit Floating Point format for Low-Precision Training of Deep Neural Networks

Training with larger number of parameters while keeping fast iterations is an increasingly adopted strategy and trend for developing better performing Deep Neural Network (DNN) models. This necessitates increased memory footprint and computational requirements for training. Here we introduce a novel methodology for training deep neural networks using 8-bit floating point (FP8) numbers. Reduced bit precision allows for a larger effective memory and increased computational speed. We name this method Shifted and Squeezed FP8 (S2FP8). We show that, unlike previous 8-bit precision training methods, the proposed method works out-of-the-box for representative models: ResNet-50, Transformer and NCF. The method can maintain model accuracy without requiring fine-tuning loss scaling parameters or keeping certain layers in single precision. We introduce two learnable statistics of the DNN tensors - shifted and squeezed factors that are used to optimally adjust the range of the tensors in 8-bits, thus minimizing the loss in information due to quantization.

Correlation of Auroral Dynamics and GNSS Scintillation with an Autoencoder

High energy particles originating from solar activity travel along the the Earth's magnetic field and interact with the atmosphere around the higher latitudes. These interactions often manifest as aurora in the form of visible light in the Earth's ionosphere. These interactions also result in irregularities in the electron density, which cause disruptions in the amplitude and phase of the radio signals from the Global Navigation Satellite Systems (GNSS), known as 'scintillation'. In this paper we use a multi-scale residual autoencoder (Res-AE) to show the correlation between specific dynamic structures of the aurora and the magnitude of the GNSS phase scintillations ($\sigma_{\phi}$). Auroral images are encoded in a lower dimensional feature space using the Res-AE, which in turn are clustered with t-SNE and UMAP. Both methods produce similar clusters, and specific clusters demonstrate greater correlations with observed phase scintillations. Our results suggest that specific dynamic structures of auroras are highly correlated with GNSS phase scintillations.

Prediction of GNSS Phase Scintillations: A Machine Learning Approach

A Global Navigation Satellite System (GNSS) uses a constellation of satellites around the earth for accurate navigation, timing, and positioning. Natural phenomena like space weather introduce irregularities in the Earth's ionosphere, disrupting the propagation of the radio signals that GNSS relies upon. Such disruptions affect both the amplitude and the phase of the propagated waves. No physics-based model currently exists to predict the time and location of these disruptions with sufficient accuracy and at relevant scales. In this paper, we focus on predicting the phase fluctuations of GNSS radio waves, known as phase scintillations. We propose a novel architecture and loss function to predict 1 hour in advance the magnitude of phase scintillations within a time window of plus-minus 5 minutes with state-of-the-art performance.

Using Scene Graph Context to Improve Image Generation

Generating realistic images from scene graphs asks neural networks to be able to reason about object relationships and compositionality. As a relatively new task, how to properly ensure the generated images comply with scene graphs or how to measure task performance remains an open question. In this paper, we propose to harness scene graph context to improve image generation from scene graphs. We introduce a scene graph context network that pools features generated by a graph convolutional neural network that are then provided to both the image generation network and the adversarial loss. With the context network, our model is trained to not only generate realistic looking images, but also to better preserve non-spatial object relationships. We also define two novel evaluation metrics, the relation score and the mean opinion relation score, for this task that directly evaluate scene graph compliance. We use both quantitative and qualitative studies to demonstrate that our pro-posed model outperforms the state-of-the-art on this challenging task.