Inner interpretability is a promising field aiming to uncover the internal mechanisms of AI systems through scalable, automated methods. While significant research has been conducted on large language models, limited attention has been paid to applying inner interpretability to large-scale image tasks, focusing primarily 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. Specifically, 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.

Modern vision models achieve remarkable accuracy, but explaining where evidence arises, what the model encodes, and how internal computations assemble that evidence remains fragmented. We introduce an iERF-centric framework that unifies local, global, and mechanistic interpretability around a single analysis unit: the pointwise feature vector (PFV) paired with its instance-specific Effective Receptive Field (iERF). On the local side, Sharing Ratio Decomposition (SRD) expresses each PFV as a mixture of upstream PFVs via sharing ratios and propagates iERFs to construct class-discriminative saliency maps. SRD yields high-resolution, activation-faithful explanations, is robust to targeted manipulation and noise, and remains activation-agnostic across common nonlinearities. For the global view, we introduce Concept-Anchored Feature Explanation (CAFE), which utilizes the iERF as a semantic label, grounding abstract latent vectors in verifiable pixel-level evidence. With CAFE, we address the challenge of non-localized sparse autoencoder latents-especially in Transformers, where early self-attention mixes distant context. To answer how representations are composed through depth, we propose the Interlayer Concept Graph with Interlayer Concept Attribution (ICAT), which quantifies concept-to-concept influence while isolating layer pairs; an interlayer insertion/deletion protocol identifies Integrated Gradients as the most faithful instantiation. Empirically, across ResNet50, VGG16, and ViTs, our framework outperforms baselines in both fidelity and robustness, successfully interprets dispersed SAE features, and exposes dominant concept routes in correct, misclassified, and adversarial cases. Grounded in iERFs, our approach provides a coherent, evidence-backed map from pixels to concepts to decisions.

A darknet market is an online marketplace where illicit goods or criminal services are provided. Because of the explosive development in the popularity of such marketplaces, there is a recognized requirement for automated analysis of cybercriminal activities in the darknet markets. However, the vendor's multiple accounts (Sybil accounts) make it difficult to identify the actual relationship between vendors. Recent studies proposed darknet market-based Sybil detection methods using images and item descriptions. However, relying solely on market information presents limitations, as it should depend on profiles and product data uploaded by vendors for inferring similarity and identifying Sybil accounts. This study introduces a novel forum-based model to complement market-based approaches, enhancing Sybil account detection by uncovering multiple accounts not easily detectable through market data alone. The proposed model filters Sybil-relevant posts, extracts vital information about Sybil vendors from forum posts, and validates vendors' presence in the marketplace. Evaluations on real-world datasets from major darknet markets show that 98 % accuracy in Sybil-related post classification and 93% accuracy in extracting information about vendors' multiple accounts are achieved. Our method successfully identifies a significant number of additional multiaccount pairs in comparison to the state-of-the-art market-based models, uncovering 541 Sybil pairs, a substantial 92 % of which are newly discovered multi-accounts not previously detected by existing models. To the best of our knowledge, this is the first study that utilizes forum data and demonstrates its efficacy for Sybil detection in the real world.

Although meta-learning seems promising performance in the realm of rapid adaptability, it is constrained by fixed cardinality. When faced with tasks of varying cardinalities that were unseen during training, the model lacks its ability. In this paper, we address and resolve this challenge by harnessing `label equivalence' emerged from stochastic numeric label assignments during episodic task sampling. Questioning what defines ``true" meta-learning, we introduce the ``any-way" learning paradigm, an innovative model training approach that liberates model from fixed cardinality constraints. Surprisingly, this model not only matches but often outperforms traditional fixed-way models in terms of performance, convergence speed, and stability. This disrupts established notions about domain generalization. Furthermore, we argue that the inherent label equivalence naturally lacks semantic information. To bridge this semantic information gap arising from label equivalence, we further propose a mechanism for infusing semantic class information into the model. This would enhance the model's comprehension and functionality. Experiments conducted on renowned architectures like MAML and ProtoNet affirm the effectiveness of our method.

The truthfulness of existing explanation methods in authentically elucidating the underlying model's decision-making process has been questioned. Existing methods have deviated from faithfully representing the model, thus susceptible to adversarial attacks. To address this, we propose a novel eXplainable AI (XAI) method called SRD (Sharing Ratio Decomposition), which sincerely reflects the model's inference process, resulting in significantly enhanced robustness in our explanations. Different from the conventional emphasis on the neuronal level, we adopt a vector perspective to consider the intricate nonlinear interactions between filters. We also introduce an interesting observation termed Activation-Pattern-Only Prediction (APOP), letting us emphasize the importance of inactive neurons and redefine relevance encapsulating all relevant information including both active and inactive neurons. Our method, SRD, allows for the recursive decomposition of a Pointwise Feature Vector (PFV), providing a high-resolution Effective Receptive Field (ERF) at any layer.

In this work, we introduce our method of outdoor scene relighting for Neural Radiance Fields (NeRF) named Sun-aligned Relighting TensoRF (SR-TensoRF). SR-TensoRF offers a lightweight and rapid pipeline aligned with the sun, thereby achieving a simplified workflow that eliminates the need for environment maps. Our sun-alignment strategy is motivated by the insight that shadows, unlike viewpoint-dependent albedo, are determined by light direction. We directly use the sun direction as an input during shadow generation, simplifying the requirements of the inference process significantly. Moreover, SR-TensoRF leverages the training efficiency of TensoRF by incorporating our proposed cubemap concept, resulting in notable acceleration in both training and rendering processes compared to existing methods.

Several side-channel attacks exploiting timing, cache, or power side channels have recently been proposed to obtain private information of a neural network. However, the hardware-based attacks require physical access to the system, using high-precision equipment to measure physical system behaviors such as power consumption or electromagnetic emanations, to exploit them as side channels. Whereas, the previous software-based side-channel attacks on neural networks can extract their model information only when the target architecture is known. In this paper, we propose the$\gamma$-Knife attack, a software-based power side-channel attack on a neural network, which can extract its architecture without any physical access or high-precision measuring equipment. Our work demonstrates that side-channels can be formed that leak architecture of neural networks by utilizing statistical metrics without high-resolution power data. The$\gamma$-Knife attack can reduce the search space of candidate architectures by obtaining private information such as filter size, depth of convolutional layer, and activation functions in the target architecture, as accurately as hardware-based power side-channel attacks even when the target neural network is totally unknown. We demonstrated the efficacy of the$\gamma$-Knife attack by implementing the attack on the well-known neural networks VGGNet, ResNet, GoogleNet, and MobileNet, using the Pytorch library on Intel CPUs and AMD CPUs. The$\gamma$-Knife attack could identify the target neural network architecture with an accuracy of approximately 90%, and efficiently extract its private information, by significantly reducing the search space of the target architecture.

A darknet market is an online marketplace typically implemented over Tor, where vendors sell illegal products or criminal services. Due to dramatic growth in the popularity of such markets, there is a recognized need for automatic investigation of the market’s ecosystem and identification of anonymous vendors. However, as they often create multiple accounts (or Sybil accounts) within or across different marketplaces, detecting Sybil accounts becomes the key to understanding the ecosystem of darknet markets and identifying the actual relationship between the vendors. This study presents a novel Sybil detection method that extracts multiple features of vendors from photographs in a fine-grained level (e.g., image similarity, main category, subcategory, and text data), and reveals the multiple Sybil accounts of them simultaneously. Each feature is extracted from multiple rich sources using an image hash algorithm, Deep Neural Network (DNN) classifier, image restoration, and text recognition tool; and merged using a weighted feature embedding model. The matching score of each vendor is then calculated to identify not only the exact Sybil accounts, but multiple potential accounts suspected of being associated to a single operator. We evaluate the efficacy of our method using real-world datasets from four large darknet markets (i.e., SilkRoad2, Agora, Evolution, Alphabay) from 2014 to 2015. Because of the anonymity of darknet market, we construct the ground-truth of Sybil accounts by randomly splitting the dataset of vendors into two even parts. We used the first set to train the model, and linked the second set to the original vendor in the first set to evaluate performance. Our experimental results demonstrated that the proposed method outperforms the existing photography-based system with an accuracy of 98%, identifying up to 700% more candidate Sybil accounts than prior work. Additionally, our method detects multiple Sybil accounts for 90% of evaluated test cases, presenting a very different picture of darknet marketplace dynamics than methods that can only detect a single Sybil account for each target vendor. Due to its fine-grained multiple feature extraction from photographs, our method can be more generically applied to various darknet markets for Sybil detection regardless of their languages or categories of items.