I Have Covered All the Bases Here: Interpreting Reasoning Features in Large Language Models via Sparse Autoencoders

Large Language Models (LLMs) have achieved remarkable success in natural language processing. Recent advances have led to the developing of a new class of reasoning LLMs; for example, open-source DeepSeek-R1 has achieved state-of-the-art performance by integrating deep thinking and complex reasoning. Despite these impressive capabilities, the internal reasoning mechanisms of such models remain unexplored. In this work, we employ Sparse Autoencoders (SAEs), a method to learn a sparse decomposition of latent representations of a neural network into interpretable features, to identify features that drive reasoning in the DeepSeek-R1 series of models. First, we propose an approach to extract candidate ''reasoning features'' from SAE representations. We validate these features through empirical analysis and interpretability methods, demonstrating their direct correlation with the model's reasoning abilities. Crucially, we demonstrate that steering these features systematically enhances reasoning performance, offering the first mechanistic account of reasoning in LLMs. Code available at https://github.com/AIRI-Institute/SAE-Reasoning

DIPLI: Deep Image Prior Lucky Imaging for Blind Astronomical Image Restoration

Contemporary image restoration and super-resolution techniques effectively harness deep neural networks, markedly outperforming traditional methods. However, astrophotography presents unique challenges for deep learning due to limited training data. This work explores hybrid strategies, such as the Deep Image Prior (DIP) model, which facilitates blind training but is susceptible to overfitting, artifact generation, and instability when handling noisy images. We propose enhancements to the DIP model's baseline performance through several advanced techniques. First, we refine the model to process multiple frames concurrently, employing the Back Projection method and the TVNet model. Next, we adopt a Markov approach incorporating Monte Carlo estimation, Langevin dynamics, and a variational input technique to achieve unbiased estimates with minimal variance and counteract overfitting effectively. Collectively, these modifications reduce the likelihood of noise learning and mitigate loss function fluctuations during training, enhancing result stability. We validated our algorithm across multiple image sets of astronomical and celestial objects, achieving performance that not only mitigates limitations of Lucky Imaging, a classical computer vision technique that remains a standard in astronomical image reconstruction but surpasses the original DIP model, state of the art transformer- and diffusion-based models, underscoring the significance of our improvements.

CLEAR: Character Unlearning in Textual and Visual Modalities

Machine Unlearning (MU) is critical for enhancing privacy and security in deep learning models, particularly in large multimodal language models (MLLMs), by removing specific private or hazardous information. While MU has made significant progress in textual and visual modalities, multimodal unlearning (MMU) remains significantly underexplored, partially due to the absence of a suitable open-source benchmark. To address this, we introduce CLEAR, a new benchmark designed to evaluate MMU methods. CLEAR contains 200 fictitious individuals and 3,700 images linked with corresponding question-answer pairs, enabling a thorough evaluation across modalities. We assess 10 MU methods, adapting them for MMU, and highlight new challenges specific to multimodal forgetting. We also demonstrate that simple $\ell_1$ regularization on LoRA weights significantly mitigates catastrophic forgetting, preserving model performance on retained data. The dataset is available at https://huggingface.co/datasets/therem/CLEAR

AASIST3: KAN-Enhanced AASIST Speech Deepfake Detection using SSL Features and Additional Regularization for the ASVspoof 2024 Challenge

Automatic Speaker Verification (ASV) systems, which identify speakers based on their voice characteristics, have numerous applications, such as user authentication in financial transactions, exclusive access control in smart devices, and forensic fraud detection. However, the advancement of deep learning algorithms has enabled the generation of synthetic audio through Text-to-Speech (TTS) and Voice Conversion (VC) systems, exposing ASV systems to potential vulnerabilities. To counteract this, we propose a novel architecture named AASIST3. By enhancing the existing AASIST framework with Kolmogorov-Arnold networks, additional layers, encoders, and pre-emphasis techniques, AASIST3 achieves a more than twofold improvement in performance. It demonstrates minDCF results of 0.5357 in the closed condition and 0.1414 in the open condition, significantly enhancing the detection of synthetic voices and improving ASV security.

GLiRA: Black-Box Membership Inference Attack via Knowledge Distillation

While Deep Neural Networks (DNNs) have demonstrated remarkable performance in tasks related to perception and control, there are still several unresolved concerns regarding the privacy of their training data, particularly in the context of vulnerability to Membership Inference Attacks (MIAs). In this paper, we explore a connection between the susceptibility to membership inference attacks and the vulnerability to distillation-based functionality stealing attacks. In particular, we propose {GLiRA}, a distillation-guided approach to membership inference attack on the black-box neural network. We observe that the knowledge distillation significantly improves the efficiency of likelihood ratio of membership inference attack, especially in the black-box setting, i.e., when the architecture of the target model is unknown to the attacker. We evaluate the proposed method across multiple image classification datasets and models and demonstrate that likelihood ratio attacks when guided by the knowledge distillation, outperform the current state-of-the-art membership inference attacks in the black-box setting.

Certification of Speaker Recognition Models to Additive Perturbations

Speaker recognition technology is applied in various tasks ranging from personal virtual assistants to secure access systems. However, the robustness of these systems against adversarial attacks, particularly to additive perturbations, remains a significant challenge. In this paper, we pioneer applying robustness certification techniques to speaker recognition, originally developed for the image domain. In our work, we cover this gap by transferring and improving randomized smoothing certification techniques against norm-bounded additive perturbations for classification and few-shot learning tasks to speaker recognition. We demonstrate the effectiveness of these methods on VoxCeleb 1 and 2 datasets for several models. We expect this work to improve voice-biometry robustness, establish a new certification benchmark, and accelerate research of certification methods in the audio domain.

FS-Net: Full Scale Network and Adaptive Threshold for Improving Extraction of Micro-Retinal Vessel Structures

Retinal vascular segmentation, is a widely researched subject in biomedical image processing, aims to relieve ophthalmologists' workload when treating and detecting retinal disorders. However, segmenting retinal vessels has its own set of challenges, with prior techniques failing to generate adequate results when segmenting branches and microvascular structures. The neural network approaches used recently are characterized by the inability to keep local and global properties together and the failure to capture tiny end vessels make it challenging to attain the desired result. To reduce this retinal vessel segmentation problem, we propose a full-scale micro-vessel extraction mechanism based on an encoder-decoder neural network architecture, sigmoid smoothing, and an adaptive threshold method. The network consists of of residual, encoder booster, bottleneck enhancement, squeeze, and excitation building blocks. All of these blocks together help to improve the feature extraction and prediction of the segmentation map. The proposed solution has been evaluated using the DRIVE, CHASE-DB1, and STARE datasets, and competitive results are obtained when compared with previous studies. The AUC and accuracy on the DRIVE dataset are 0.9884 and 0.9702, respectively. On the CHASE-DB1 dataset, the scores are 0.9903 and 0.9755, respectively. On the STARE dataset, the scores are 0.9916 and 0.9750, respectively. The performance achieved is one step ahead of what has been done in previous studies, and this results in a higher chance of having this solution in real-life diagnostic centers that seek ophthalmologists attention.

DeepLOC: Deep Learning-based Bone Pathology Localization and Classification in Wrist X-ray Images

In recent years, computer-aided diagnosis systems have shown great potential in assisting radiologists with accurate and efficient medical image analysis. This paper presents a novel approach for bone pathology localization and classification in wrist X-ray images using a combination of YOLO (You Only Look Once) and the Shifted Window Transformer (Swin) with a newly proposed block. The proposed methodology addresses two critical challenges in wrist X-ray analysis: accurate localization of bone pathologies and precise classification of abnormalities. The YOLO framework is employed to detect and localize bone pathologies, leveraging its real-time object detection capabilities. Additionally, the Swin, a transformer-based module, is utilized to extract contextual information from the localized regions of interest (ROIs) for accurate classification.

Medical Image Captioning via Generative Pretrained Transformers

The automatic clinical caption generation problem is referred to as proposed model combining the analysis of frontal chest X-Ray scans with structured patient information from the radiology records. We combine two language models, the Show-Attend-Tell and the GPT-3, to generate comprehensive and descriptive radiology records. The proposed combination of these models generates a textual summary with the essential information about pathologies found, their location, and the 2D heatmaps localizing each pathology on the original X-Ray scans. The proposed model is tested on two medical datasets, the Open-I, MIMIC-CXR, and the general-purpose MS-COCO. The results measured with the natural language assessment metrics prove their efficient applicability to the chest X-Ray image captioning.

Autofocusing+: Noise-Resilient Motion Correction in Magnetic Resonance Imaging

Image corruption by motion artifacts is an ingrained problem in Magnetic Resonance Imaging (MRI). In this work, we propose a neural network-based regularization term to enhance Autofocusing, a classic optimization-based method to remove motion artifacts. The method takes the best of both worlds: the optimization-based routine iteratively executes the blind demotion and deep learning-based prior penalizes for unrealistic restorations and speeds up the convergence. We validate the method on three models of motion trajectories, using synthetic and real noisy data. The method proves resilient to noise and anatomic structure variation, outperforming the state-of-the-art demotion methods.