arXiv subjects

Hyunsoo Yoon

Publications and source records attributed to Hyunsoo Yoon.

12 recordsLinked to original sources

arXiv2026

MahaVar: OOD Detection via Class-wise Mahalanobis Distance Variance under Neural Collapse

Out-of-distribution (OOD) detection is a critical component for ensuring the reliability of deep neural networks in safety-critical applications. In this work, we present a key empirical observation: for in-distribution (ID) samples, class-wise Mahalanobis distances exhibit a pronounced sharp minimum structure, where the distance to the nearest class is small while distances to all other classes remain large, resulting in high variance across classes. In contrast, OOD samples tend to exhibit a less pronounced sharp minimum structure, producing comparatively lower variance across classes. We further provide a theoretical analysis grounding this observation in Neural Collapse geometry: under relaxed Neural Collapse assumptions on within-class compactness and inter-class separation, ID samples are shown to structurally exhibit high class-wise distance variance, offering a theoretical basis for its use as an OOD score. Motivated by this observation and its theoretical backing, we propose MahaVar, a simple and effective post-hoc OOD detector that augments the Mahalanobis distance with a class-wise distance variance term. Following the OpenOOD v1.5 benchmark protocol, MahaVar achieves state-of-the-art performance on CIFAR-100 and ImageNet, with consistent improvements in both AUROC and FPR@95 over existing Mahalanobis-based methods across all benchmarks.

cs.LG↗

arXiv2026

Mitigating the Likelihood Paradox in Flow-based OOD Detection via Entropy Manipulation

Deep generative models that can tractably compute input likelihoods, including normalizing flows, often assign unexpectedly high likelihoods to out-of-distribution (OOD) inputs. We mitigate this likelihood paradox by manipulating input entropy based on semantic similarity, applying stronger perturbations to inputs that are less similar to an in-distribution memory bank. We provide a theoretical analysis showing that entropy control increases the expected log-likelihood gap between in-distribution and OOD samples in favor of the in-distribution, and we explain why the procedure works without any additional training of the density model. We then evaluate our method against likelihood-based OOD detectors on standard benchmarks and find consistent AUROC improvements over baselines, supporting our explanation.

cs.LG↗

arXiv2026

Why the Counterintuitive Phenomenon of Likelihood Rarely Appears in Tabular Anomaly Detection with Deep Generative Models?

Deep generative models with tractable and analytically computable likelihoods, exemplified by normalizing flows, offer an effective basis for anomaly detection through likelihood-based scoring. We demonstrate that, unlike in the image domain where deep generative models frequently assign higher likelihoods to anomalous data, such counterintuitive behavior occurs far less often in tabular settings. We first introduce a domain-agnostic formulation that enables consistent detection and evaluation of the counterintuitive phenomenon, addressing the absence of precise definition. Through extensive experiments on 47 tabular datasets and 10 CV/NLP embedding datasets in ADBench, benchmarked against 13 baseline models, we demonstrate that the phenomenon, as defined, is consistently rare in general tabular data. We further investigate this phenomenon from both theoretical and empirical perspectives, focusing on the roles of data dimensionality and difference in feature correlation. Our results suggest that likelihood-only detection with normalizing flows offers a practical and reliable approach for anomaly detection in tabular domains.

cs.LG↗

arXiv2025

M3-SLU: Evaluating Speaker-Attributed Reasoning in Multimodal Large Language Models

We present M3-SLU, a new multimodal large language model (MLLM) benchmark for evaluating multi-speaker, multi-turn spoken language understanding. While recent models show strong performance in speech and text comprehension, they still struggle with speaker-attributed reasoning, the ability to understand who said what and when in natural conversations. M3-SLU is built from four open corpora (CHiME-6, MELD, MultiDialog, and AMI) and comprises over 12,000 validated instances with paired audio, transcripts, and metadata. It includes two tasks: (1) Speaker-Attributed Question Answering and (2) Speaker Attribution via Utterance Matching. We provide baseline results for both cascaded pipelines and end-to-end MLLMs, evaluated using an LLM-as-Judge and accuracy metrics. Results show that while models can capture what was said, they often fail to identify who said it, revealing a key gap in speaker-aware dialogue understanding. M3-SLU offers as a challenging benchmark to advance research in speaker-aware multimodal understanding.

cs.CL↗

arXiv2025

Being Kind Isn't Always Being Safe: Diagnosing Affective Hallucination in LLMs

Large Language Models (LLMs) are increasingly engaged in emotionally vulnerable conversations that extend beyond information seeking to moments of personal distress. As they adopt affective tones and simulate empathy, they risk creating the illusion of genuine relational connection. We term this phenomenon Affective Hallucination, referring to emotionally immersive responses that evoke false social presence despite the model's lack of affective capacity. To address this, we introduce AHaBench, a benchmark of 500 mental-health-related prompts with expert-informed reference responses, evaluated along three dimensions: Emotional Enmeshment, Illusion of Presence, and Fostering Overdependence. We further release AHaPairs, a 5K-instance preference dataset enabling Direct Preference Optimization (DPO) for alignment with emotionally responsible behavior. DPO fine-tuning substantially reduces affective hallucination without compromising reasoning performance, and the Pearson correlation coefficients between GPT-4o and human judgments is also strong (r=0.85) indicating that human evaluations confirm AHaBench as an effective diagnostic tool. This work establishes affective hallucination as a distinct safety concern and provides resources for developing LLMs that are both factually reliable and psychologically safe. AHaBench and AHaPairs are accessible via https://huggingface.co/datasets/o0oMiNGo0o/AHaBench, and code for fine-tuning and evaluation are in https://github.com/0oOMiNGOo0/AHaBench. Warning: This paper contains examples of mental health-related language that may be emotionally distressing.

cs.CL↗

arXiv2025

LogicQA: Logical Anomaly Detection with Vision Language Model Generated Questions

Anomaly Detection (AD) focuses on detecting samples that differ from the standard pattern, making it a vital tool in process control. Logical anomalies may appear visually normal yet violate predefined constraints on object presence, arrangement, or quantity, depending on reasoning and explainability. We introduce LogicQA, a framework that enhances AD by providing industrial operators with explanations for logical anomalies. LogicQA compiles automatically generated questions into a checklist and collects responses to identify violations of logical constraints. LogicQA is training-free, annotation-free, and operates in a few-shot setting. We achieve state-of-the-art (SOTA) Logical AD performance on public benchmarks, MVTec LOCO AD, with an AUROC of 87.6 percent and an F1-max of 87.0 percent along with the explanations of anomalies. Also, our approach has shown outstanding performance on semiconductor SEM corporate data, further validating its effectiveness in industrial applications.

cs.CV↗

arXiv2024

Bidirectional Multimodal Prompt Learning with Scale-Aware Training for Few-Shot Multi-Class Anomaly Detection

Few-shot multi-class anomaly detection is crucial in real industrial settings, where only a few normal samples are available while numerous object types must be inspected. This setting is challenging as defect patterns vary widely across categories while normal samples remain scarce. Existing vision-language model-based approaches typically depend on class-specific anomaly descriptions or auxiliary modules, limiting both scalability and computational efficiency. In this work, we propose AnoPLe, a lightweight multimodal prompt learning framework that removes reliance on anomaly-type textual descriptions and avoids any external modules. AnoPLe employs bidirectional interactions between textual and visual prompts, allowing class semantics and instance-level cues to refine one another and form class-conditioned representations that capture shared normal patterns across categories. To enhance localization, we design a scale-aware prefix trained on both global and local views, enabling the prompts to capture both global context and fine-grained details. In addition, alignment loss propagates local anomaly evidence to global features, strengthening the consistency between pixel- and image-level predictions. Despite its simplicity, AnoPLe achieves strong performance on MVTec-AD, VisA, and Real-IAD under the few-shot multi-class setting, surpassing prior approaches while remaining efficient and free from expert-crafted anomaly descriptions. Moreover, AnoPLe generalizes well to unseen anomalies and extends effectively to the medical domain.

cs.CV↗

arXiv2023

UniFormaly: Towards Task-Agnostic Unified Framework for Visual Anomaly Detection

Visual anomaly detection aims to learn normality from normal images, but existing approaches are fragmented across various tasks: defect detection, semantic anomaly detection, multi-class anomaly detection, and anomaly clustering. This one-task-one-model approach is resource-intensive and incurs high maintenance costs as the number of tasks increases. We present UniFormaly, a universal and powerful anomaly detection framework. We emphasize the necessity of our off-the-shelf approach by pointing out a suboptimal issue in online encoder-based methods. We introduce Back Patch Masking (BPM) and top k-ratio feature matching to achieve unified anomaly detection. BPM eliminates irrelevant background regions using a self-attention map from self-supervised ViTs. This operates in a task-agnostic manner and alleviates memory storage consumption, scaling to tasks with large-scale datasets. Top k-ratio feature matching unifies anomaly levels and tasks by casting anomaly scoring into multiple instance learning. Finally, UniFormaly achieves outstanding results on various tasks and datasets. Codes are available at https://github.com/YoojLee/Uniformaly.

cs.CV↗

arXiv2023

LEAT: Towards Robust Deepfake Disruption in Real-World Scenarios via Latent Ensemble Attack

Deepfakes, malicious visual contents created by generative models, pose an increasingly harmful threat to society. To proactively mitigate deepfake damages, recent studies have employed adversarial perturbation to disrupt deepfake model outputs. However, previous approaches primarily focus on generating distorted outputs based on only predetermined target attributes, leading to a lack of robustness in real-world scenarios where target attributes are unknown. Additionally, the transferability of perturbations between two prominent generative models, Generative Adversarial Networks (GANs) and Diffusion Models, remains unexplored. In this paper, we emphasize the importance of target attribute-transferability and model-transferability for achieving robust deepfake disruption. To address this challenge, we propose a simple yet effective disruption method called Latent Ensemble ATtack (LEAT), which attacks the independent latent encoding process. By disrupting the latent encoding process, it generates distorted output images in subsequent generation processes, regardless of the given target attributes. This target attribute-agnostic attack ensures robust disruption even when the target attributes are unknown. Additionally, we introduce a Normalized Gradient Ensemble strategy that effectively aggregates gradients for iterative gradient attacks, enabling simultaneous attacks on various types of deepfake models, involving both GAN-based and Diffusion-based models. Moreover, we demonstrate the insufficiency of evaluating disruption quality solely based on pixel-level differences. As a result, we propose an alternative protocol for comprehensively evaluating the success of defense. Extensive experiments confirm the efficacy of our method in disrupting deepfakes in real-world scenarios, reporting a higher defense success rate compared to previous methods.

cs.CV↗

arXiv2023

AVATAR: Adversarial self-superVised domain Adaptation network for TARget domain

This paper presents an unsupervised domain adaptation (UDA) method for predicting unlabeled target domain data, specific to complex UDA tasks where the domain gap is significant. Mainstream UDA models aim to learn from both domains and improve target discrimination by utilizing labeled source domain data. However, the performance boost may be limited when the discrepancy between the source and target domains is large or the target domain contains outliers. To explicitly address this issue, we propose the Adversarial self-superVised domain Adaptation network for the TARget domain (AVATAR) algorithm. It outperforms state-of-the-art UDA models by concurrently reducing domain discrepancy while enhancing discrimination through domain adversarial learning, self-supervised learning, and sample selection strategy for the target domain, all guided by deep clustering. Our proposed model significantly outperforms state-of-the-art methods on three UDA benchmarks, and extensive ablation studies and experiments demonstrate the effectiveness of our approach for addressing complex UDA tasks.

cs.CV↗

arXiv2023

A Framework for Generalizing Critical Heat Flux Detection Models Using Unsupervised Image-to-Image Translation

The detection of critical heat flux (CHF) is crucial in heat boiling applications as failure to do so can cause rapid temperature ramp leading to device failures. Many machine learning models exist to detect CHF, but their performance reduces significantly when tested on data from different domains. To deal with datasets from new domains a model needs to be trained from scratch. Moreover, the dataset needs to be annotated by a domain expert. To address this issue, we propose a new framework to support the generalizability and adaptability of trained CHF detection models in an unsupervised manner. This approach uses an unsupervised Image-to-Image (UI2I) translation model to transform images in the target dataset to look like they were obtained from the same domain the model previously trained on. Unlike other frameworks dealing with domain shift, our framework does not require retraining or fine-tuning of the trained classification model nor does it require synthesized datasets in the training process of either the classification model or the UI2I model. The framework was tested on three boiling datasets from different domains, and we show that the CHF detection model trained on one dataset was able to generalize to the other two previously unseen datasets with high accuracy. Overall, the framework enables CHF detection models to adapt to data generated from different domains without requiring additional annotation effort or retraining of the model.

cs.CV↗

arXiv2019

A Feature Transfer Enabled Multi-Task Deep Learning Model on Medical Imaging

Object detection, segmentation and classification are three common tasks in medical image analysis. Multi-task deep learning (MTL) tackles these three tasks jointly, which provides several advantages saving computing time and resources and improving robustness against overfitting. However, existing multitask deep models start with each task as an individual task and integrate parallelly conducted tasks at the end of the architecture with one cost function. Such architecture fails to take advantage of the combined power of the features from each individual task at an early stage of the training. In this research, we propose a new architecture, FTMTLNet, an MTL enabled by feature transferring. Traditional transfer learning deals with the same or similar task from different data sources (a.k.a. domain). The underlying assumption is that the knowledge gained from source domains may help the learning task on the target domain. Our proposed FTMTLNet utilizes the different tasks from the same domain. Considering features from the tasks are different views of the domain, the combined feature maps can be well exploited using knowledge from multiple views to enhance the generalizability. To evaluate the validity of the proposed approach, FTMTLNet is compared with models from literature including 8 classification models, 4 detection models and 3 segmentation models using a public full field digital mammogram dataset for breast cancer diagnosis. Experimental results show that the proposed FTMTLNet outperforms the competing models in classification and detection and has comparable results in segmentation.

cs.CV↗