Inference-time steering aims to alter the response characteristics of large language models (LLMs) without modifying their underlying parameters. A critical step in this process is the identification of internal modules within LLMs that are associated with the target behavior. However, current approaches to module selection often depend on superficial cues or ad-hoc heuristics, which can result in suboptimal or unintended outcomes. In this work, we propose a principled causal-attribution framework for identifying behavior-relevant attention heads in transformers. For each head, we train a vector-quantized autoencoder (VQ-AE) on its attention activations, partitioning the latent space into behavior-relevant and behavior-irrelevant subspaces, each quantized with a shared learnable codebook. We assess the behavioral relevance of each head by quantifying the separability of VQ-AE encodings for behavior-aligned versus behavior-violating responses using a binary classification metric. This yields a behavioral relevance score that reflects each head discriminative capacity with respect to the target behavior, guiding both selection and importance weighting. Experiments on seven LLMs from two model families and five behavioral steering datasets demonstrate that our method enables more accurate inference-time interventions, achieving superior performance on the truthfulness-steering task. Furthermore, the heads selected by our approach exhibit strong zero-shot generalization in cross-domain truthfulness-steering scenarios.
Counterfactual explanations for black-box models aim to pr ovide insight into an algorithmic decision to its recipient. For a binary classification problem an individual counterfactual details which features might be changed for the model to infer the opposite class. High-dimensional feature spaces that are typical of machine learning classification models admit many possible counterfactual examples to a decision, and so it is important to identify additional criteria to select the most useful counterfactuals. In this paper, we explore the idea that the counterfactuals should be maximally informative when considering the knowledge of a specific individual about the underlying classifier. To quantify this information gain we explicitly model the knowledge of the individual, and assess the uncertainty of predictions which the individual makes by the width of a conformal prediction interval. Regions of feature space where the prediction interval is wide correspond to areas where the confidence in decision making is low, and an additional counterfactual example might be more informative to an individual. To explore and evaluate our individualised conformal prediction interval counterfactuals (CPICFs), first we present a synthetic data set on a hypercube which allows us to fully visualise the decision boundary, conformal intervals via three different methods, and resultant CPICFs. Second, in this synthetic data set we explore the impact of a single CPICF on the knowledge of an individual locally around the original query. Finally, in both our synthetic data set and a complex real world dataset with a combination of continuous and discrete variables, we measure the utility of these counterfactuals via data augmentation, testing the performance on a held out set.
Law has long been a domain that has been popular in natural language processing (NLP) applications. Reasoning (ratiocination and the ability to make connections to precedent) is a core part of the practice of the law in the real world. Nevertheless, while multiple legal datasets exist, none have thus far focused specifically on reasoning tasks. We focus on a specific aspect of the legal landscape by introducing a corporate governance reasoning benchmark (CHANCERY) to test a model's ability to reason about whether executive/board/shareholder's proposed actions are consistent with corporate governance charters. This benchmark introduces a first-of-its-kind corporate governance reasoning test for language models - modeled after real world corporate governance law. The benchmark consists of a corporate charter (a set of governing covenants) and a proposal for executive action. The model's task is one of binary classification: reason about whether the action is consistent with the rules contained within the charter. We create the benchmark following established principles of corporate governance - 24 concrete corporate governance principles established in and 79 real life corporate charters selected to represent diverse industries from a total dataset of 10k real life corporate charters. Evaluations on state-of-the-art (SOTA) reasoning models confirm the difficulty of the benchmark, with models such as Claude 3.7 Sonnet and GPT-4o achieving 64.5% and 75.2% accuracy respectively. Reasoning agents exhibit superior performance, with agents based on the ReAct and CodeAct frameworks scoring 76.1% and 78.1% respectively, further confirming the advanced legal reasoning capabilities required to score highly on the benchmark. We also conduct an analysis of the types of questions which current reasoning models struggle on, revealing insights into the legal reasoning capabilities of SOTA models.
Forest fire prediction involves estimating the likelihood of fire ignition or related risk levels in a specific area over a defined time period. With climate change intensifying fire behavior and frequency, accurate prediction has become one of the most pressing challenges in Artificial Intelligence (AI). Traditionally, fire ignition is approached as a binary classification task in the literature. However, this formulation oversimplifies the problem, especially from the perspective of end-users such as firefighters. In general, as is the case in France, firefighting units are organized by department, each with its terrain, climate conditions, and historical experience with fire events. Consequently, fire risk should be modeled in a way that is sensitive to local conditions and does not assume uniform risk across all regions. This paper proposes a new approach that tailors fire risk assessment to departmental contexts, offering more actionable and region-specific predictions for operational use. With this, we present the first national-scale AI benchmark for metropolitan France using state-of-the-art AI models on a relatively unexplored dataset. Finally, we offer a summary of important future works that should be taken into account. Supplementary materials are available on GitHub.
In recent years, deep learning (DL) approaches have demonstrated promising results in decoding hemodynamic responses captured by functional near-infrared spectroscopy (fNIRS), particularly in the context of brain-computer interface (BCI) applications. This work introduces AbsoluteNet, a novel deep learning architecture designed to classify auditory event-related responses recorded using fNIRS. The proposed network is built upon principles of spatio-temporal convolution and customized activation functions. Our model was compared against several models, namely fNIRSNET, MDNN, DeepConvNet, and ShallowConvNet. The results showed that AbsoluteNet outperforms existing models, reaching 87.0% accuracy, 84.8% sensitivity, and 89.2% specificity in binary classification, surpassing fNIRSNET, the second-best model, by 3.8% in accuracy. These findings underscore the effectiveness of our proposed deep learning model in decoding hemodynamic responses related to auditory processing and highlight the importance of spatio-temporal feature aggregation and customized activation functions to better fit fNIRS dynamics.
Recent research demonstrated that training large language models involves memorization of a significant fraction of training data. Such memorization can lead to privacy violations when training on sensitive user data and thus motivates the study of data memorization's role in learning. In this work, we develop a general approach for proving lower bounds on excess data memorization, that relies on a new connection between strong data processing inequalities and data memorization. We then demonstrate that several simple and natural binary classification problems exhibit a trade-off between the number of samples available to a learning algorithm, and the amount of information about the training data that a learning algorithm needs to memorize to be accurate. In particular, $\Omega(d)$ bits of information about the training data need to be memorized when $O(1)$ $d$-dimensional examples are available, which then decays as the number of examples grows at a problem-specific rate. Further, our lower bounds are generally matched (up to logarithmic factors) by simple learning algorithms. We also extend our lower bounds to more general mixture-of-clusters models. Our definitions and results build on the work of Brown et al. (2021) and address several limitations of the lower bounds in their work.
Generating realistic synthetic tabular data presents a critical challenge in machine learning. It adds another layer of complexity when this data contain class imbalance problems. This paper presents a novel token-aware data imputation method that leverages the in-context learning capabilities of large language models. This is achieved through the combination of a structured group-wise CSV-style prompting technique and the elimination of irrelevant contextual information in the input prompt. We test this approach with two class-imbalanced binary classification datasets and evaluate the effectiveness of imputation using classification-based evaluation metrics. The experimental results demonstrate that our approach significantly reduces the input prompt size while maintaining or improving imputation quality compared to our baseline prompt, especially for datasets that are of relatively smaller in size. The contributions of this presented work is two-fold -- 1) it sheds light on the importance of prompt design when leveraging LLMs for synthetic data generation and 2) it addresses a critical gap in LLM-based data imputation for class-imbalanced datasets with missing data by providing a practical solution within computational constraints. We hope that our work will foster further research and discussions about leveraging the incredible potential of LLMs and prompt engineering techniques for synthetic data generation.
The scientific literature is growing rapidly, making it hard to keep track of the state-of-the-art. Systematic literature reviews (SLRs) aim to identify and evaluate all relevant papers on a topic. After retrieving a set of candidate papers, the abstract screening phase determines initial relevance. To date, abstract screening methods using large language models (LLMs) focus on binary classification settings; existing question answering (QA) based ranking approaches suffer from error propagation. LLMs offer a unique opportunity to evaluate the SLR's inclusion and exclusion criteria, yet, existing benchmarks do not provide them exhaustively. We manually extract these criteria as well as research questions for 57 SLRs, mostly in the medical domain, enabling principled comparisons between approaches. Moreover, we propose LGAR, a zero-shot LLM Guided Abstract Ranker composed of an LLM based graded relevance scorer and a dense re-ranker. Our extensive experiments show that LGAR outperforms existing QA-based methods by 5-10 pp. in mean average precision. Our code and data is publicly available.
This paper presents one of the top-performing solutions to the UNLP 2025 Shared Task on Detecting Manipulation in Social Media. The task focuses on detecting and classifying rhetorical and stylistic manipulation techniques used to influence Ukrainian Telegram users. For the classification subtask, we fine-tuned the Gemma 2 language model with LoRA adapters and applied a second-level classifier leveraging meta-features and threshold optimization. For span detection, we employed an XLM-RoBERTa model trained for multi-target, including token binary classification. Our approach achieved 2nd place in classification and 3rd place in span detection.
Adam is the de facto optimization algorithm for several deep learning applications, but an understanding of its implicit bias and how it differs from other algorithms, particularly standard first-order methods such as (stochastic) gradient descent (GD), remains limited. In practice, neural networks trained with SGD are known to exhibit simplicity bias -- a tendency to find simple solutions. In contrast, we show that Adam is more resistant to such simplicity bias. To demystify this phenomenon, in this paper, we investigate the differences in the implicit biases of Adam and GD when training two-layer ReLU neural networks on a binary classification task involving synthetic data with Gaussian clusters. We find that GD exhibits a simplicity bias, resulting in a linear decision boundary with a suboptimal margin, whereas Adam leads to much richer and more diverse features, producing a nonlinear boundary that is closer to the Bayes' optimal predictor. This richer decision boundary also allows Adam to achieve higher test accuracy both in-distribution and under certain distribution shifts. We theoretically prove these results by analyzing the population gradients. To corroborate our theoretical findings, we present empirical results showing that this property of Adam leads to superior generalization across datasets with spurious correlations where neural networks trained with SGD are known to show simplicity bias and don't generalize well under certain distributional shifts.
Deep learning techniques have enabled vast improvements in computer vision technologies. Nevertheless, these models are vulnerable to adversarial patch attacks which catastrophically impair performance. The physically realizable nature of these attacks calls for certifiable defenses, which feature provable guarantees on robustness. While certifiable defenses have been successfully applied to single-label classification, limited work has been done for multi-label classification. In this work, we present PatchDEMUX, a certifiably robust framework for multi-label classifiers against adversarial patches. Our approach is a generalizable method which can extend any existing certifiable defense for single-label classification; this is done by considering the multi-label classification task as a series of isolated binary classification problems to provably guarantee robustness. Furthermore, in the scenario where an attacker is limited to a single patch we propose an additional certification procedure that can provide tighter robustness bounds. Using the current state-of-the-art (SOTA) single-label certifiable defense PatchCleanser as a backbone, we find that PatchDEMUX can achieve non-trivial robustness on the MS-COCO and PASCAL VOC datasets while maintaining high clean performance
Placenta Accreta Spectrum Disorders (PAS) pose significant risks during pregnancy, frequently leading to postpartum hemorrhage during cesarean deliveries and other severe clinical complications, with bleeding severity correlating to the degree of placental invasion. Consequently, accurate prenatal diagnosis of PAS and its subtypes-placenta accreta (PA), placenta increta (PI), and placenta percreta (PP)-is crucial. However, existing guidelines and methodologies predominantly focus on the presence of PAS, with limited research addressing subtype recognition. Additionally, previous multi-class diagnostic efforts have primarily relied on inefficient two-stage cascaded binary classification tasks. In this study, we propose a novel convolutional neural network (CNN) architecture designed for efficient one-stage multiclass diagnosis of PAS and its subtypes, based on 4,140 magnetic resonance imaging (MRI) slices. Our model features two branches: the main classification branch utilizes a residual block architecture comprising multiple residual blocks, while the second branch integrates anatomical features of the uteroplacental area and the adjacent uterine serous layer to enhance the model's attention during classification. Furthermore, we implement a multitask learning strategy to leverage both branches effectively. Experiments conducted on a real clinical dataset demonstrate that our model achieves state-of-the-art performance.
This paper presents our submission to the PROCESS Challenge 2025, focusing on spontaneous speech analysis for early dementia detection. For the three-class classification task (Healthy Control, Mild Cognitive Impairment, and Dementia), we propose a cascaded binary classification framework that fine-tunes pre-trained language models and incorporates pause encoding to better capture disfluencies. This design streamlines multi-class classification and addresses class imbalance by restructuring the decision process. For the Mini-Mental State Examination score regression task, we develop an enhanced multimodal fusion system that combines diverse acoustic and linguistic features. Separate regression models are trained on individual feature sets, with ensemble learning applied through score averaging. Experimental results on the test set outperform the baselines provided by the organizers in both tasks, demonstrating the robustness and effectiveness of our approach.
Structural health monitoring (SHM) involves sensor deployment, data acquisition, and data interpretation, commonly implemented via a tedious wired system. The information processing in current practice majorly depends on electronic computers, albeit with universal applications, delivering challenges such as high energy consumption and low throughput due to the nature of digital units. In recent years, there has been a renaissance interest in shifting computations from electronic computing units to the use of real physical systems, a concept known as physical computation. This approach provides the possibility of thinking out of the box for SHM, seamlessly integrating sensing and computing into a pure-physical entity, without relying on external electronic power supplies, thereby properly coping with resource-restricted scenarios. The latest advances of metamaterials (MM) hold great promise for this proactive idea. In this paper, we introduce a programmable metamaterial-based sensor (termed as MM-sensor) for physically processing structural vibration information to perform specific SHM tasks, such as structural damage warning (binary classification) in this initiation, without the need for further information processing or resource-consuming, that is, the data collection and analysis are completed in-situ at the sensor level. We adopt the configuration of a locally resonant metamaterial plate (LRMP) to achieve the first fabrication of the MM-sensor. We take advantage of the bandgap properties of LRMP to physically differentiate the dynamic behavior of structures before and after damage. By inversely designing the geometric parameters, our current approach allows for adjustments to the bandgap features. This is effective for engineering systems with a first natural frequency ranging from 9.54 Hz to 81.86 Hz.
With the rapid development of Large Language Models (LLMs), aligning these models with human preferences and values is critical to ensuring ethical and safe applications. However, existing alignment techniques such as RLHF or DPO often require direct fine-tuning on LLMs with billions of parameters, resulting in substantial computational costs and inefficiencies. To address this, we propose Micro token-level Accept-Reject Aligning (MARA) approach designed to operate independently of the language models. MARA simplifies the alignment process by decomposing sentence-level preference learning into token-level binary classification, where a compact three-layer fully-connected network determines whether candidate tokens are "Accepted" or "Rejected" as part of the response. Extensive experiments across seven different LLMs and three open-source datasets show that MARA achieves significant improvements in alignment performance while reducing computational costs. The source code and implementation details are publicly available at https://github.com/IAAR-Shanghai/MARA, and the trained models are released at https://huggingface.co/IAAR-Shanghai/MARA_AGENTS.
Podcasts have become daily companions for half a billion users. Given the enormous amount of podcast content available, highlights provide a valuable signal that helps viewers get the gist of an episode and decide if they want to invest in listening to it in its entirety. However, identifying highlights automatically is challenging due to the unstructured and long-form nature of the content. We introduce Rhapsody, a dataset of 13K podcast episodes paired with segment-level highlight scores derived from YouTube's 'most replayed' feature. We frame the podcast highlight detection as a segment-level binary classification task. We explore various baseline approaches, including zero-shot prompting of language models and lightweight finetuned language models using segment-level classification heads. Our experimental results indicate that even state-of-the-art language models like GPT-4o and Gemini struggle with this task, while models finetuned with in-domain data significantly outperform their zero-shot performance. The finetuned model benefits from leveraging both speech signal features and transcripts. These findings highlight the challenges for fine-grained information access in long-form spoken media.
The proliferation of misinformation in digital platforms reveals the limitations of traditional detection methods, which mostly rely on static classification and fail to capture the intricate process of real-world fact-checking. Despite advancements in Large Language Models (LLMs) that enhance automated reasoning, their application to misinformation detection remains hindered by issues of logical inconsistency and superficial verification. In response, we introduce Debate-to-Detect (D2D), a novel Multi-Agent Debate (MAD) framework that reformulates misinformation detection as a structured adversarial debate. Inspired by fact-checking workflows, D2D assigns domain-specific profiles to each agent and orchestrates a five-stage debate process, including Opening Statement, Rebuttal, Free Debate, Closing Statement, and Judgment. To transcend traditional binary classification, D2D introduces a multi-dimensional evaluation mechanism that assesses each claim across five distinct dimensions: Factuality, Source Reliability, Reasoning Quality, Clarity, and Ethics. Experiments with GPT-4o on two fakenews datasets demonstrate significant improvements over baseline methods, and the case study highlight D2D's capability to iteratively refine evidence while improving decision transparency, representing a substantial advancement towards robust and interpretable misinformation detection. The code will be open-sourced in a future release.
Over 8,024 wildfire incidents have been documented in 2024 alone, affecting thousands of fatalities and significant damage to infrastructure and ecosystems. Wildfires in the United States have inflicted devastating losses. Wildfires are becoming more frequent and intense, which highlights how urgently efficient warning systems are needed to avoid disastrous outcomes. The goal of this study is to enhance the accuracy of wildfire detection by using Convolutional Neural Network (CNN) built on the VGG16 architecture. The D-FIRE dataset, which includes several kinds of wildfire and non-wildfire images, was employed in the study. Low-resolution images, dataset imbalance, and the necessity for real-time applicability are some of the main challenges. These problems were resolved by enriching the dataset using data augmentation techniques and optimizing the VGG16 model for binary classification. The model produced a low false negative rate, which is essential for reducing unexplored fires, despite dataset boundaries. In order to help authorities execute fast responses, this work shows that deep learning models such as VGG16 can offer a reliable, automated approach for early wildfire recognition. For the purpose of reducing the impact of wildfires, our future work will concentrate on connecting to systems with real-time surveillance networks and enlarging the dataset to cover more varied fire situations.
Recent studies on the safety alignment of large language models (LLMs) have revealed that existing approaches often operate superficially, leaving models vulnerable to various adversarial attacks. Despite their significance, these studies generally fail to offer actionable solutions beyond data augmentation for achieving more robust safety mechanisms. This paper identifies a fundamental cause of this superficiality: existing alignment approaches often presume that models can implicitly learn a safety-related reasoning task during the alignment process, enabling them to refuse harmful requests. However, the learned safety signals are often diluted by other competing objectives, leading models to struggle with drawing a firm safety-conscious decision boundary when confronted with adversarial attacks. Based on this observation, by explicitly introducing a safety-related binary classification task and integrating its signals with our attention and decoding strategies, we eliminate this ambiguity and allow models to respond more responsibly to malicious queries. We emphasize that, with less than 0.2x overhead cost, our approach enables LLMs to assess the safety of both the query and the previously generated tokens at each necessary generating step. Extensive experiments demonstrate that our method significantly improves the resilience of LLMs against various adversarial attacks, offering a promising pathway toward more robust generative AI systems.
Monkeypox is a viral disease characterized by distinctive skin lesions and has been reported in many countries. The recent global outbreak has emphasized the urgent need for scalable, accessible, and accurate diagnostic solutions to support public health responses. In this study, we developed ITMAINN, an intelligent, AI-driven healthcare system specifically designed to detect Monkeypox from skin lesion images using advanced deep learning techniques. Our system consists of three main components. First, we trained and evaluated several pretrained models using transfer learning on publicly available skin lesion datasets to identify the most effective models. For binary classification (Monkeypox vs. non-Monkeypox), the Vision Transformer, MobileViT, Transformer-in-Transformer, and VGG16 achieved the highest performance, each with an accuracy and F1-score of 97.8%. For multiclass classification, which contains images of patients with Monkeypox and five other classes (chickenpox, measles, hand-foot-mouth disease, cowpox, and healthy), ResNetViT and ViT Hybrid models achieved 92% accuracy, with F1 scores of 92.24% and 92.19%, respectively. The best-performing and most lightweight model, MobileViT, was deployed within the mobile application. The second component is a cross-platform smartphone application that enables users to detect Monkeypox through image analysis, track symptoms, and receive recommendations for nearby healthcare centers based on their location. The third component is a real-time monitoring dashboard designed for health authorities to support them in tracking cases, analyzing symptom trends, guiding public health interventions, and taking proactive measures. This system is fundamental in developing responsive healthcare infrastructure within smart cities. Our solution, ITMAINN, is part of revolutionizing public health management.