Vahid Reza Khazaie

Vahid Reza Khazaie, Anthony Wong, Mohammad Sabokrou

This paper presents a novel evaluation framework for Out-of-Distribution (OOD) detection that aims to assess the performance of machine learning models in more realistic settings. We observed that the real-world requirements for testing OOD detection methods are not satisfied by the current testing protocols. They usually encourage methods to have a strong bias towards a low level of diversity in normal data. To address this limitation, we propose new OOD test datasets (CIFAR-10-R, CIFAR-100-R, and ImageNet-30-R) that can allow researchers to benchmark OOD detection performance under realistic distribution shifts. Additionally, we introduce a Generalizability Score (GS) to measure the generalization ability of a model during OOD detection. Our experiments demonstrate that improving the performance on existing benchmark datasets does not necessarily improve the usability of OOD detection models in real-world scenarios. While leveraging deep pre-trained features has been identified as a promising avenue for OOD detection research, our experiments show that state-of-the-art pre-trained models tested on our proposed datasets suffer a significant drop in performance. To address this issue, we propose a post-processing stage for adapting pre-trained features under these distribution shifts before calculating the OOD scores, which significantly enhances the performance of state-of-the-art pre-trained models on our benchmarks.

Vahid Reza Khazaie, Anthony Wong, John Taylor Jewell et al.

Anomaly detection is to identify samples that do not conform to the distribution of the normal data. Due to the unavailability of anomalous data, training a supervised deep neural network is a cumbersome task. As such, unsupervised methods are preferred as a common approach to solve this task. Deep autoencoders have been broadly adopted as a base of many unsupervised anomaly detection methods. However, a notable shortcoming of deep autoencoders is that they provide insufficient representations for anomaly detection by generalizing to reconstruct outliers. In this work, we have designed an adversarial framework consisting of two competing components, an Adversarial Distorter, and an Autoencoder. The Adversarial Distorter is a convolutional encoder that learns to produce effective perturbations and the autoencoder is a deep convolutional neural network that aims to reconstruct the images from the perturbed latent feature space. The networks are trained with opposing goals in which the Adversarial Distorter produces perturbations that are applied to the encoder's latent feature space to maximize the reconstruction error and the autoencoder tries to neutralize the effect of these perturbations to minimize it. When applied to anomaly detection, the proposed method learns semantically richer representations due to applying perturbations to the feature space. The proposed method outperforms the existing state-of-the-art methods in anomaly detection on image and video datasets.

Vahid Reza Khazaie, Anthony Wong, Yalda Mohsenzadeh

Anomaly detection is to recognize samples that differ in some respect from the training observations. These samples which do not conform to the distribution of normal data are called outliers or anomalies. In real-world anomaly detection problems, the outliers are absent, not well defined, or have a very limited number of instances. Recent state-of-the-art deep learning-based anomaly detection methods suffer from high computational cost, complexity, unstable training procedures, and non-trivial implementation, making them difficult to deploy in real-world applications. To combat this problem, we leverage a simple learning procedure that trains a lightweight convolutional neural network, reaching state-of-the-art performance in anomaly detection. In this paper, we propose to solve anomaly detection as a supervised regression problem. We label normal and anomalous data using two separable distributions of continuous values. To compensate for the unavailability of anomalous samples during training time, we utilize straightforward image augmentation techniques to create a distinct set of samples as anomalies. The distribution of the augmented set is similar but slightly deviated from the normal data, whereas real anomalies are expected to have an even further distribution. Therefore, training a regressor on these augmented samples will result in more separable distributions of labels for normal and real anomalous data points. Anomaly detection experiments on image and video datasets show the superiority of the proposed method over the state-of-the-art approaches.

Shaina Raza, Ashmal Vayani, Aditya Jain et al.

Detecting disinformation that blends manipulated text and images has become increasingly challenging, as AI tools make synthetic content easy to generate and disseminate. While most existing AI safety benchmarks focus on single modality misinformation (i.e., false content shared without intent to deceive), intentional multimodal disinformation, such as propaganda or conspiracy theories that imitate credible news, remains largely unaddressed. We introduce the Vision-Language Disinformation Detection Benchmark (VLDBench), the first large-scale resource supporting both unimodal (text-only) and multimodal (text + image) disinformation detection. VLDBench comprises approximately 62,000 labeled text-image pairs across 13 categories, curated from 58 news outlets. Using a semi-automated pipeline followed by expert review, 22 domain experts invested over 500 hours to produce high-quality annotations with substantial inter-annotator agreement. Evaluations of state-of-the-art Large Language Models (LLMs) and Vision-Language Models (VLMs) on VLDBench show that incorporating visual cues improves detection accuracy by 5 to 35 percentage points over text-only models. VLDBench provides data and code for evaluation, fine-tuning, and robustness testing to support disinformation analysis. Developed in alignment with AI governance frameworks (e.g., the MIT AI Risk Repository), VLDBench offers a principled foundation for advancing trustworthy disinformation detection in multimodal media. Project: https://vectorinstitute.github.io/VLDBench/ Dataset: https://huggingface.co/datasets/vector-institute/VLDBench Code: https://github.com/VectorInstitute/VLDBench

Ananya Raval, Aravind Narayanan, Vahid Reza Khazaie et al.

Large Multimodal Models (LMMs) are typically trained on vast corpora of image-text data but are often limited in linguistic coverage, leading to biased and unfair outputs across languages. While prior work has explored multimodal evaluation, less emphasis has been placed on assessing multilingual capabilities. In this work, we introduce LinguaMark, a benchmark designed to evaluate state-of-the-art LMMs on a multilingual Visual Question Answering (VQA) task. Our dataset comprises 6,875 image-text pairs spanning 11 languages and five social attributes. We evaluate models using three key metrics: Bias, Answer Relevancy, and Faithfulness. Our findings reveal that closed-source models generally achieve the highest overall performance. Both closed-source (GPT-4o and Gemini2.5) and open-source models (Gemma3, Qwen2.5) perform competitively across social attributes, and Qwen2.5 demonstrates strong generalization across multiple languages. We release our benchmark and evaluation code to encourage reproducibility and further research.

Negin Baghbanzadeh, Adibvafa Fallahpour, Yasaman Parhizkar et al.

Despite the growing scale of medical Vision-Language datasets, the impact of dataset quality on model performance remains under-explored. We introduce Open-PMC, a high-quality medical dataset from PubMed Central, containing 2.2 million image-text pairs, enriched with image modality annotations, subfigures, and summarized in-text references. Notably, the in-text references provide richer medical context, extending beyond the abstract information typically found in captions. Through extensive experiments, we benchmark Open-PMC against larger datasets across retrieval and zero-shot classification tasks. Our results show that dataset quality-not just size-drives significant performance gains. We complement our benchmark with an in-depth analysis of feature representation. Our findings highlight the crucial role of data curation quality in advancing multimodal medical AI. We release Open-PMC, along with the trained models and our codebase.

Sana Ayromlou, Vahid Reza Khazaie, Fereshteh Forghani et al.

The rapid advancement in self-supervised representation learning has highlighted its potential to leverage unlabeled data for learning rich visual representations. However, the existing techniques, particularly those employing different augmentations of the same image, often rely on a limited set of simple transformations that cannot fully capture variations in the real world. This constrains the diversity and quality of samples, which leads to sub-optimal representations. In this paper, we introduce a framework that enriches the self-supervised learning (SSL) paradigm by utilizing generative models to produce semantically consistent image augmentations. By directly conditioning generative models on a source image, our method enables the generation of diverse augmentations while maintaining the semantics of the source image, thus offering a richer set of data for SSL. Our extensive experimental results on various joint-embedding SSL techniques demonstrate that our framework significantly enhances the quality of learned visual representations by up to 10\% Top-1 accuracy in downstream tasks. This research demonstrates that incorporating generative models into the joint-embedding SSL workflow opens new avenues for exploring the potential of synthetic data. This development paves the way for more robust and versatile representation learning techniques.

Aravind Narayanan, Vahid Reza Khazaie, Shaina Raza

Large vision-language models (VLMs) can jointly interpret images and text, but they are also prone to absorbing and reproducing harmful social stereotypes when visual cues such as age, gender, race, clothing, or occupation are present. To investigate these risks, we introduce a news-image benchmark consisting of 1,343 image-question pairs drawn from diverse outlets, which we annotated with ground-truth answers and demographic attributes (age, gender, race, occupation, and sports). We evaluate a range of state-of-the-art VLMs and employ a large language model (LLM) as judge, with human verification. Our findings show that: (i) visual context systematically shifts model outputs in open-ended settings; (ii) bias prevalence varies across attributes and models, with particularly high risk for gender and occupation; and (iii) higher faithfulness does not necessarily correspond to lower bias. We release the benchmark prompts, evaluation rubric, and code to support reproducible and fairness-aware multimodal assessment.

Shaina Raza, Caesar Saleh, Azib Farooq et al.

Detecting bias in multimodal news requires models that reason over text--image pairs, not just classify text. In response, we present ViLBias, a VQA-style benchmark and framework for detecting and reasoning about bias in multimodal news. The dataset comprises 40,945 text--image pairs from diverse outlets, each annotated with a bias label and concise rationale using a two-stage LLM-as-annotator pipeline with hierarchical majority voting and human-in-the-loop validation. We evaluate Small Language Models (SLMs), Large Language Models (LLMs), and Vision--Language Models (VLMs) across closed-ended classification and open-ended reasoning (oVQA), and compare parameter-efficient tuning strategies. Results show that incorporating images alongside text improves detection accuracy by 3--5\%, and that LLMs/VLMs better capture subtle framing and text--image inconsistencies than SLMs. Parameter-efficient methods (LoRA/QLoRA/Adapters) recover 97--99\% of full fine-tuning performance with $<5\%$ trainable parameters. For oVQA, reasoning accuracy spans 52--79\% and faithfulness 68--89\%, both improved by instruction tuning; closed accuracy correlates strongly with reasoning ($r = 0.91$). ViLBias offers a scalable benchmark and strong baselines for multimodal bias detection and rationale quality.

Mohammad Akhavan Anvari, Rojina Kashefi, Vahid Reza Khazaie et al.

Anomaly detection involves identifying instances within a dataset that deviate from the norm and occur infrequently. Current benchmarks tend to favor methods biased towards low diversity in normal data, which does not align with real-world scenarios. Despite advancements in these benchmarks, contemporary anomaly detection methods often struggle with out-of-distribution generalization, particularly in classifying samples with subtle transformations during testing. These methods typically assume that normal samples during test time have distributions very similar to those in the training set, while anomalies are distributed much further away. However, real-world test samples often exhibit various levels of distribution shift while maintaining semantic consistency. Therefore, effectively generalizing to samples that have undergone semantic-preserving transformations, while accurately detecting normal samples whose semantic meaning has changed after transformation as anomalies, is crucial for the trustworthiness and reliability of a model. For example, although it is clear that rotation shifts the meaning for a car in the context of anomaly detection but preserves the meaning for a bird, current methods are likely to detect both as abnormal. This complexity underscores the necessity for dynamic learning procedures rooted in the intrinsic concept of outliers. To address this issue, we propose new testing protocols and a novel method called Knowledge Exposure (KE), which integrates external knowledge to comprehend concept dynamics and differentiate transformations that induce semantic shifts. This approach enhances generalization by utilizing insights from a pre-trained CLIP model to evaluate the significance of anomalies for each concept. Evaluation on CIFAR-10, CIFAR-100, and SVHN with the new protocols demonstrates superior performance compared to previous methods.

Shuvendu Roy, Yasaman Parhizkar, Franklin Ogidi et al.

We perform a comprehensive benchmarking of contrastive frameworks for learning multimodal representations in the medical domain. Through this study, we aim to answer the following research questions: (i) How transferable are general-domain representations to the medical domain? (ii) Is multimodal contrastive training sufficient, or does it benefit from unimodal training as well? (iii) What is the impact of feature granularity on the effectiveness of multimodal medical representation learning? To answer these questions, we investigate eight contrastive learning approaches under identical training setups, and train them on 2.8 million image-text pairs from four datasets, and evaluate them on 25 downstream tasks, including classification (zero-shot and linear probing), image-to-text and text-to-image retrieval, and visual question-answering. Our findings suggest a positive answer to the first question, a negative answer to the second question, and the benefit of learning fine-grained features. Finally, we make our code publicly available.

John Taylor Jewell, Vahid Reza Khazaie, Yalda Mohsenzadeh

Novelty detection is the task of recognizing samples that do not belong to the distribution of the target class. During training, the novelty class is absent, preventing the use of traditional classification approaches. Deep autoencoders have been widely used as a base of many unsupervised novelty detection methods. In particular, context autoencoders have been successful in the novelty detection task because of the more effective representations they learn by reconstructing original images from randomly masked images. However, a significant drawback of context autoencoders is that random masking fails to consistently cover important structures of the input image, leading to suboptimal representations - especially for the novelty detection task. In this paper, to optimize input masking, we have designed a framework consisting of two competing networks, a Mask Module and a Reconstructor. The Mask Module is a convolutional autoencoder that learns to generate optimal masks that cover the most important parts of images. Alternatively, the Reconstructor is a convolutional encoder-decoder that aims to reconstruct unperturbed images from masked images. The networks are trained in an adversarial manner in which the Mask Module generates masks that are applied to images given to the Reconstructor. In this way, the Mask Module seeks to maximize the reconstruction error that the Reconstructor is minimizing. When applied to novelty detection, the proposed approach learns semantically richer representations compared to context autoencoders and enhances novelty detection at test time through more optimal masking. Novelty detection experiments on the MNIST and CIFAR-10 image datasets demonstrate the proposed approach's superiority over cutting-edge methods. In a further experiment on the UCSD video dataset for novelty detection, the proposed approach achieves state-of-the-art results.

Vahid Reza Khazaie, Nicky Bayat, Yalda Mohsenzadeh

State-of-the-art deep neural network models have reached near perfect face recognition accuracy rates on controlled high-resolution face images. However, their performance is drastically degraded when they are tested with very low-resolution face images. This is particularly critical in surveillance systems, where a low-resolution probe image is to be matched with high-resolution gallery images. super-resolution techniques aim at producing high-resolution face images from low-resolution counterparts. While they are capable of reconstructing images that are visually appealing, the identity-related information is not preserved. Here, we propose an identity-preserving end-to-end image-to-image translation deep neural network which is capable of super-resolving very low-resolution faces to their high-resolution counterparts while preserving identity-related information. We achieved this by training a very deep convolutional encoder-decoder network with a symmetric contracting path between corresponding layers. This network was trained with a combination of a reconstruction and an identity-preserving loss, on multi-scale low-resolution conditions. Extensive quantitative evaluations of our proposed model demonstrated that it outperforms competing super-resolution and low-resolution face recognition methods on natural and artificial low-resolution face data sets and even unseen identities.

Andrew Keyes, Nicky Bayat, Vahid Reza Khazaie et al.

Advanced Generative Adversarial Networks (GANs) are remarkable in generating intelligible audio from a random latent vector. In this paper, we examine the task of recovering the latent vector of both synthesized and real audio. Previous works recovered latent vectors of given audio through an auto-encoder inspired technique that trains an encoder network either in parallel with the GAN or after the generator is trained. With our approach, we train a deep residual neural network architecture to project audio synthesized by WaveGAN into the corresponding latent space with near identical reconstruction performance. To accommodate for the lack of an original latent vector for real audio, we optimize the residual network on the perceptual loss between the real audio samples and the reconstructed audio of the predicted latent vectors. In the case of synthesized audio, the Mean Squared Error (MSE) between the ground truth and recovered latent vector is minimized as well. We further investigated the audio reconstruction performance when several gradient optimization steps are applied to the predicted latent vector. Through our deep neural network based method of training on real and synthesized audio, we are able to predict a latent vector that corresponds to a reasonable reconstruction of real audio. Even though we evaluated our method on WaveGAN, our proposed method is universal and can be applied to any other GANs.

Nicky Bayat, Vahid Reza Khazaie, Yalda Mohsenzadeh

Generative adversarial networks (GANs) synthesize realistic images from a random latent vector. While many studies have explored various training configurations and architectures for GANs, the problem of inverting a generative model to extract latent vectors of given input images has been inadequately investigated. Although there is exactly one generated image per given random vector, the mapping from an image to its recovered latent vector can have more than one solution. We train a ResNet architecture to recover a latent vector for a given face that can be used to generate a face nearly identical to the target. We use a perceptual loss to embed face details in the recovered latent vector while maintaining visual quality using a pixel loss. The vast majority of studies on latent vector recovery perform well only on generated images, we argue that our method can be used to determine a mapping between real human faces and latent-space vectors that contain most of the important face style details. In addition, our proposed method projects generated faces to their latent-space with high fidelity and speed. At last, we demonstrate the performance of our approach on both real and generated faces.