Search

Inner interpretability is a promising field focused on uncovering the internal mechanisms of AI systems and developing scalable, automated methods to understand these systems at a mechanistic level. While significant research has explored top-down approaches starting from high-level problems or algorithmic hypotheses and bottom-up approaches building higher-level abstractions from low-level or circuit-level descriptions, most efforts have concentrated on analyzing large language models. Moreover, limited attention has been given to applying inner interpretability to large-scale image tasks, primarily focusing on architectural and functional levels to visualize learned concepts. In this paper, we first present a conceptual framework that supports inner interpretability and multilevel analysis for large-scale image classification tasks. We introduce the Bi-directional Interaction between Concept and Input Embeddings (Bi-ICE) module, which facilitates interpretability across the computational, algorithmic, and implementation levels. This module enhances transparency by generating predictions based on human-understandable concepts, quantifying their contributions, and localizing them within the inputs. Finally, we showcase enhanced transparency in image classification, measuring concept contributions and pinpointing their locations within the inputs. Our approach highlights algorithmic interpretability by demonstrating the process of concept learning and its convergence., Comment: The first two authors equally contributed to this work, 27 pages, 19 figures, 9 tables

All-in-one image restoration aims to handle multiple degradation types using one model. This paper proposes a simple pipeline for all-in-one blind image restoration to Restore Anything with Masks (RAM). We focus on the image content by utilizing Mask Image Modeling to extract intrinsic image information rather than distinguishing degradation types like other methods. Our pipeline consists of two stages: masked image pre-training and fine-tuning with mask attribute conductance. We design a straightforward masking pre-training approach specifically tailored for all-in-one image restoration. This approach enhances networks to prioritize the extraction of image content priors from various degradations, resulting in a more balanced performance across different restoration tasks and achieving stronger overall results. To bridge the gap of input integrity while preserving learned image priors as much as possible, we selectively fine-tuned a small portion of the layers. Specifically, the importance of each layer is ranked by the proposed Mask Attribute Conductance (MAC), and the layers with higher contributions are selected for finetuning. Extensive experiments demonstrate that our method achieves state-of-the-art performance. Our code and model will be released at \href{https://github.com/Dragonisss/RAM}{https://github.com/Dragonisss/RAM}., Comment: Accepted by ECCV 2024

Blood oxygenation level-dependent (BOLD) functional magnetic resonance imaging (fMRI) is widely used to visualize brain activation regions by detecting hemodynamic responses associated with increased metabolic demand. While alternative MRI methods have been employed to monitor functional activities, the investigation of in-vivo electrical property changes during brain function remains limited. In this study, we explored the relationship between fMRI signals and electrical conductivity (measured at the Larmor frequency) changes using phase-based electrical properties tomography (EPT). Our results revealed consistent patterns: conductivity changes showed negative correlations, with conductivity decreasing in the functionally active regions whereas B1 phase mapping exhibited positive correlations around activation regions. These observations were consistent across both motor and visual cortex activations. To further substantiate these findings, we conducted electromagnetic radio-frequency simulations that modeled activation states with varying conductivity, which demonstrated trends similar to our in-vivo results for both B1 phase and conductivity. These findings suggest that in-vivo electrical conductivity changes can indeed be measured during brain activity. However, further investigation is needed to fully understand the underlying mechanisms driving these measurements.

With the advancement of deep neural networks in computer vision, artificial intelligence (AI) is widely employed in real-world applications. However, AI still faces limitations in mimicking high-level human capabilities, such as novel category discovery, for practical use. While some methods utilizing offline continual learning have been proposed for novel category discovery, they neglect the continuity of data streams in real-world settings. In this work, we introduce Online Continuous Generalized Category Discovery (OCGCD), which considers the dynamic nature of data streams where data can be created and deleted in real time. Additionally, we propose a novel method, DEAN, Discovery via Energy guidance and feature AugmentatioN, which can discover novel categories in an online manner through energy-guided discovery and facilitate discriminative learning via energy-based contrastive loss. Furthermore, DEAN effectively pseudo-labels unlabeled data through variance-based feature augmentation. Experimental results demonstrate that our proposed DEAN achieves outstanding performance in proposed OCGCD scenario.

Random forests are considered a cornerstone in machine learning for their robustness and versatility. Despite these strengths, their conventional centralized training is ill-suited for the modern landscape of data that is often distributed, sensitive, and subject to privacy concerns. Federated learning (FL) provides a compelling solution to this problem, enabling models to be trained across a group of clients while maintaining the privacy of each client's data. However, adapting tree-based methods like random forests to federated settings introduces significant challenges, particularly when it comes to non-identically distributed (non-IID) data across clients, which is a common scenario in real-world applications. This paper presents a federated random forest approach that employs a novel ensemble construction method aimed at improving performance under non-IID data. Instead of growing trees independently in each client, our approach ensures each decision tree in the ensemble is iteratively and collectively grown across clients. To preserve the privacy of the client's data, we confine the information stored in the leaf nodes to the majority class label identified from the samples of the client's local data that reach each node. This limited disclosure preserves the confidentiality of the underlying data distribution of clients, thereby enhancing the privacy of the federated learning process. Furthermore, our collaborative ensemble construction strategy allows the ensemble to better reflect the data's heterogeneity across different clients, enhancing its performance on non-IID data, as our experimental results confirm., Comment: This is the authors' accepted manuscript of an article published in the journal Expert Systems With Applications. Published version available at: https://www.sciencedirect.com/science/article/pii/S0957417424016099. 22 pages, 3 figures

Unsupervised domain adaptation (UDA) for semantic segmentation aims to transfer the pixel-wise knowledge from the labeled source domain to the unlabeled target domain. However, current UDA methods typically assume a shared label space between source and target, limiting their applicability in real-world scenarios where novel categories may emerge in the target domain. In this paper, we introduce Open-Set Domain Adaptation for Semantic Segmentation (OSDA-SS) for the first time, where the target domain includes unknown classes. We identify two major problems in the OSDA-SS scenario as follows: 1) the existing UDA methods struggle to predict the exact boundary of the unknown classes, and 2) they fail to accurately predict the shape of the unknown classes. To address these issues, we propose Boundary and Unknown Shape-Aware open-set domain adaptation, coined BUS. Our BUS can accurately discern the boundaries between known and unknown classes in a contrastive manner using a novel dilation-erosion-based contrastive loss. In addition, we propose OpenReMix, a new domain mixing augmentation method that guides our model to effectively learn domain and size-invariant features for improving the shape detection of the known and unknown classes. Through extensive experiments, we demonstrate that our proposed BUS effectively detects unknown classes in the challenging OSDA-SS scenario compared to the previous methods by a large margin. The code is available at https://github.com/KHU-AGI/BUS., Comment: 14 pages, 5 figures, 13 tables, CVPR 2024 Poster

Few-Shot Class Incremental Learning (FSCIL) is a task that requires a model to learn new classes incrementally without forgetting when only a few samples for each class are given. FSCIL encounters two significant challenges: catastrophic forgetting and overfitting, and these challenges have driven prior studies to primarily rely on shallow models, such as ResNet-18. Even though their limited capacity can mitigate both forgetting and overfitting issues, it leads to inadequate knowledge transfer during few-shot incremental sessions. In this paper, we argue that large models such as vision and language transformers pre-trained on large datasets can be excellent few-shot incremental learners. To this end, we propose a novel FSCIL framework called PriViLege, Pre-trained Vision and Language transformers with prompting functions and knowledge distillation. Our framework effectively addresses the challenges of catastrophic forgetting and overfitting in large models through new pre-trained knowledge tuning (PKT) and two losses: entropy-based divergence loss and semantic knowledge distillation loss. Experimental results show that the proposed PriViLege significantly outperforms the existing state-of-the-art methods with a large margin, e.g., +9.38% in CUB200, +20.58% in CIFAR-100, and +13.36% in miniImageNet. Our implementation code is available at https://github.com/KHU-AGI/PriViLege., Comment: Accepted by CVPR 2024

We introduce HyperCLOVA X, a family of large language models (LLMs) tailored to the Korean language and culture, along with competitive capabilities in English, math, and coding. HyperCLOVA X was trained on a balanced mix of Korean, English, and code data, followed by instruction-tuning with high-quality human-annotated datasets while abiding by strict safety guidelines reflecting our commitment to responsible AI. The model is evaluated across various benchmarks, including comprehensive reasoning, knowledge, commonsense, factuality, coding, math, chatting, instruction-following, and harmlessness, in both Korean and English. HyperCLOVA X exhibits strong reasoning capabilities in Korean backed by a deep understanding of the language and cultural nuances. Further analysis of the inherent bilingual nature and its extension to multilingualism highlights the model's cross-lingual proficiency and strong generalization ability to untargeted languages, including machine translation between several language pairs and cross-lingual inference tasks. We believe that HyperCLOVA X can provide helpful guidance for regions or countries in developing their sovereign LLMs., Comment: 44 pages; updated authors list and fixed author names

Recent studies have demonstrated that diffusion models are capable of generating high-quality samples, but their quality heavily depends on sampling guidance techniques, such as classifier guidance (CG) and classifier-free guidance (CFG). These techniques are often not applicable in unconditional generation or in various downstream tasks such as image restoration. In this paper, we propose a novel sampling guidance, called Perturbed-Attention Guidance (PAG), which improves diffusion sample quality across both unconditional and conditional settings, achieving this without requiring additional training or the integration of external modules. PAG is designed to progressively enhance the structure of samples throughout the denoising process. It involves generating intermediate samples with degraded structure by substituting selected self-attention maps in diffusion U-Net with an identity matrix, by considering the self-attention mechanisms' ability to capture structural information, and guiding the denoising process away from these degraded samples. In both ADM and Stable Diffusion, PAG surprisingly improves sample quality in conditional and even unconditional scenarios. Moreover, PAG significantly improves the baseline performance in various downstream tasks where existing guidances such as CG or CFG cannot be fully utilized, including ControlNet with empty prompts and image restoration such as inpainting and deblurring., Comment: Project page is available at https://ku-cvlab.github.io/Perturbed-Attention-Guidance

Biased attributes, spuriously correlated with target labels in a dataset, can problematically lead to neural networks that learn improper shortcuts for classifications and limit their capabilities for out-of-distribution (OOD) generalization. Although many debiasing approaches have been proposed to ensure correct predictions from biased datasets, few studies have considered learning latent embedding consisting of intrinsic and biased attributes that contribute to improved performance and explain how the model pays attention to attributes. In this paper, we propose a novel debiasing framework, Debiasing Global Workspace, introducing attention-based information bottlenecks for learning compositional representations of attributes without defining specific bias types. Based on our observation that learning shape-centric representation helps robust performance on OOD datasets, we adopt those abilities to learn robust and generalizable representations of decomposable latent embeddings corresponding to intrinsic and biasing attributes. We conduct comprehensive evaluations on biased datasets, along with both quantitative and qualitative analyses, to showcase our approach's efficacy in attribute-centric representation learning and its ability to differentiate between intrinsic and bias-related features., Comment: 24 pages, 16 figures, 3 tables

Spaceborne radar remote sensing of the earth system is essential to study natural and man-made changes in the ecosystem, water and energy cycles, weather and air quality, sea level, and surface dynamics. A major challenge with current approaches is the lack of broad spectrum tunability due to narrow band microwave electronics, that limit systems to specific science variable retrievals. This results in a significant limitation in studying dynamic coupled earth system processes such as surface and subsurface hydrology, where broad spectrum radar remote sensing is needed to sense multiple variables simultaneously. Rydberg atomic sensors are highly sensitive broad-spectrum quantum detectors that can be dynamically tuned to cover micro-to-millimeter waves with no requirement for band-specific electronics. Rydberg atomic sensors can use existing transmitted signals such as navigation and communication satellites to enable remote sensing. We demonstrate remote sensing of soil moisture, an important earth system variable, via ground-based radar reflectometry with Rydberg atomic systems. To do this, we sensitize the atoms to XM satellite radio signals and use signal correlations to demonstrate use of these satellite signals for remote sensing of soil moisture. Our approach provides a step towards satellite-based broad-spectrum Rydberg atomic remote sensing.

Nanobubbles and related technologies are expected to be highly utilized in water resource-based industries such as water purification, crops, horticulture, medicine, bio, and sterilization. Ozone, a chemical with high sterilizing power, is known as a natural substance that is reduced to oxygen and water after reacting with pollutants. Ozone water, which is generated by dissolving ozone in water, has been used in various industrial sectors such as medical care, food, and environment. Due to the unstable molecular state of ozone, however, it is difficult to produce, use, and supply ozone at industrial sites in a stable manner. This study proposed a method for constructing a system that can generate high-concentration ozone water in large quantities using low power in real time and maintaining the concentration of the generated ozone water over the long term. Friction tubes (called 'nanotube') played a key role to generate nanobubble ozone water with an increased half-life for virus killing activity. In addition, the safety of ozone water during its spray into the air was explained, and virucidal activity test cases for the influenza A (H1N1/A/PR8) and COVID-19 (SARS-CoV-2) virus using high-concentration ozone water as well as its technical efficacy were described.

Nanobubble-related technologies have been confirmed to be useful in various fields such as climate change and the environment as well as water-based industries such as water purification, crops, horticulture, medical care, bio, and sterilization. However, a method of mass production in real time enough to apply nano-bubbles to the industry has not yet been developed. We explored the mechanism of nano-bubble water generation by friction between water and walls and developed a tube device applying the shape of the flow path to maximize the friction in the fluid passing through the flow path. It also describes the case of real-time and low-power mass production of nanobubbles and its technical utility. We found that the friction of nanotubes alone can easily and quickly improve the production of nanobubbles with small particle size in real time; by increasing the shearing pressure while increasing the effective friction constant value, the particle size of nanobubbles can be smaller while increasing the particle concentration., Comment: 24 pages, 24 figures, 6 tables

Interpreting answers to yes-no questions in social media is difficult. Yes and no keywords are uncommon, and the few answers that include them are rarely to be interpreted what the keywords suggest. In this paper, we present a new corpus of 4,442 yes-no question-answer pairs from Twitter. We discuss linguistic characteristics of answers whose interpretation is yes or no, as well as answers whose interpretation is unknown. We show that large language models are far from solving this problem, even after fine-tuning and blending other corpora for the same problem but outside social media., Comment: Accepted at the Findings of EMNLP 2023

Yes-no questions expect a yes or no for an answer, but people often skip polar keywords. Instead, they answer with long explanations that must be interpreted. In this paper, we focus on this challenging problem and release new benchmarks in eight languages. We present a distant supervision approach to collect training data. We also demonstrate that direct answers (i.e., with polar keywords) are useful to train models to interpret indirect answers (i.e., without polar keywords). Experimental results demonstrate that monolingual fine-tuning is beneficial if training data can be obtained via distant supervision for the language of interest (5 languages). Additionally, we show that cross-lingual fine-tuning is always beneficial (8 languages)., Comment: Accepted to EMNLP 2023 Findings