Deep knowledge tracing models have achieved significant breakthroughs in modeling student learning trajectories. However, these architectures require substantial training time and are prone to overfitting on datasets with short sequences. In this paper, we explore a new paradigm for knowledge tracing by leveraging tabular foundation models (TFMs). Unlike traditional methods that require offline training on a fixed training set, our approach performs real-time ''live'' knowledge tracing in an online way via in-context learning. TFMs align testing sequences with relevant training sequences at inference time, therefore skipping the training step entirely. We demonstrate, using several datasets of increasing size, that our method achieves competitive predictive performance with up to 53x speedups on average, in a setting where student interactions are observed progressively over time.

Recent advancements in 3D foundation models have enabled the generation of high-fidelity assets, yet precise 3D manipulation remains a significant challenge. Existing 3D editing frameworks often face a difficult trade-off between visual controllability, geometric consistency, and scalability. Specifically, optimization-based methods are prohibitively slow, multi-view 2D propagation techniques suffer from visual drift, and training-free latent manipulation methods are inherently bound by frozen priors and cannot directly benefit from scaling. In this work, we present ShapeUP, a scalable, image-conditioned 3D editing framework that formulates editing as a supervised latent-to-latent translation within a native 3D representation. This formulation allows ShapeUP to build on a pretrained 3D foundation model, leveraging its strong generative prior while adapting it to editing through supervised training. In practice, ShapeUP is trained on triplets consisting of a source 3D shape, an edited 2D image, and the corresponding edited 3D shape, and learns a direct mapping using a 3D Diffusion Transformer (DiT). This image-as-prompt approach enables fine-grained visual control over both local and global edits and achieves implicit, mask-free localization, while maintaining strict structural consistency with the original asset. Our extensive evaluations demonstrate that ShapeUP consistently outperforms current trained and training-free baselines in both identity preservation and edit fidelity, offering a robust and scalable paradigm for native 3D content creation.

We present algorithms for diffusion model sampling which obtain $δ$-error in $\mathrm{polylog}(1/δ)$ steps, given access to $\widetilde O(δ)$-accurate score estimates in $L^2$. This is an exponential improvement over all previous results. Specifically, under minimal data assumptions, the complexity is $\widetilde O(d_\star \mathrm{polylog}(1/δ))$ where $d_\star$ is the intrinsic dimension of the data. Further, under a non-uniform $L$-Lipschitz condition, the complexity reduces to $\widetilde O(L \mathrm{polylog}(1/δ))$. Our approach also yields the first $\mathrm{polylog}(1/δ)$ complexity sampler for general log-concave distributions using only gradient evaluations.

In this letter, we propose an Attention-Based Neural-Augmented Kalman Filter (AttenNKF) for state estimation in legged robots. Foot slip is a major source of estimation error: when slip occurs, kinematic measurements violate the no-slip assumption and inject bias during the update step. Our objective is to estimate this slip-induced error and compensate for it. To this end, we augment an Invariant Extended Kalman Filter (InEKF) with a neural compensator that uses an attention mechanism to infer error conditioned on foot-slip severity and then applies this estimate as a post-update compensation to the InEKF state (i.e., after the filter update). The compensator is trained in a latent space, which aims to reduce sensitivity to raw input scales and encourages structured slip-conditioned compensations, while preserving the InEKF recursion. Experiments demonstrate improved performance compared to existing legged-robot state estimators, particularly under slip-prone conditions.

Separating liquid-liquid dispersions in gravity settlers is critical in chemical, pharmaceutical, and recycling processes. The dense-packed zone height is an important performance and safety indicator but it is often expensive and impractical to measure due to optical limitations. We propose a framework to estimate phase heights by combining a PINN model with readily available volume flow measurements, without requiring phase height measurements during deployment. To this end, a physics-informed neural network (PINN) is first pretrained on synthetic data and physics equations derived from a low-fidelity (approximate) mechanistic model to reduce the need for extensive experimental data. While the mechanistic model is used to generate synthetic training data, only volume balance equations are used in the PINN, as incorporating droplet coalescence and sedimentation submodels would be computationally prohibitive. The pretrained PINN is then fine-tuned with scarce experimental phase height and flow-rate data to capture the actual dynamics of the separator. We then deploy the differentiable PINN as a predictive model in an Extended Kalman Filter inspired state estimation framework, enabling the phase heights to be tracked and updated using flow-rate measurements only. We first test the two-stage trained PINN by forward simulation from a known initial state against the mechanistic model and a non-pretrained PINN. We then evaluate phase height estimation performance with the filter, comparing the two-stage trained PINN with a two-stage trained purely data-driven neural network. All model types are trained and evaluated using ensembles to account for model parameter uncertainty. In all evaluations, the two-stage trained PINN yields the most accurate phase-height estimates.

Clustering is a long-standing research problem and a fundamental tool in AI and data analysis. The traditional k-center problem, a fundamental theoretical challenge in clustering, has a best possible approximation ratio of 2, and any improvement to a ratio of 2 - ε would imply P = NP. In this work, we study the constrained k-center clustering problem, where instance-level cannot-link (CL) and must-link (ML) constraints are incorporated as background knowledge. Although general CL constraints significantly increase the hardness of approximation, previous work has shown that disjoint CL sets permit constant-factor approximations. However, whether local search can achieve such a guarantee in this setting remains an open question. To this end, we propose a novel local search framework based on a transformation to a dominating matching set problem, achieving the best possible approximation ratio of 2. The experimental results on both real-world and synthetic datasets demonstrate that our algorithm outperforms baselines in solution quality.

Accelerated dynamic volumetric magnetic resonance imaging (4DMRI) is essential for applications relying on motion resolution. Existing 4DMRI produces acceptable artifacts of averaged breathing phases, which can blur and misrepresent instantaneous dynamic information. Recovery of such information requires a new paradigm to reconstruct extremely undersampled non-Cartesian k-space data. We propose B-FIRE, a binning-free diffusion implicit neural representation framework for hyper-accelerated MR reconstruction capable of reflecting instantaneous 3D abdominal anatomy. B-FIRE employs a CNN-INR encoder-decoder backbone optimized using diffusion with a comprehensive loss that enforces image-domain fidelity and frequency-aware constraints. Motion binned image pairs were used as training references, while inference was performed on binning-free undersampled data. Experiments were conducted on a T1-weighted StarVIBE liver MRI cohort, with accelerations ranging from 8 spokes per frame (RV8) to RV1. B-FIRE was compared against direct NuFFT, GRASP-CS, and an unrolled CNN method. Reconstruction fidelity, motion trajectory consistency, and inference latency were evaluated.

Deploying Automatic Speech Recognition (ASR) models on memory-constrained edge devices requires aggressive low-bit weight quantization. Layer-wise post-training quantization is practical and effective, but it suffers from cross-layer error accumulation. Existing compensation methods typically use a single global strength for all layers, which is ill-suited to encoder-decoder ASR models whose acoustic encoder and linguistic decoder exhibit markedly different sensitivities to quantization noise. We propose FADE, a diagnostic-driven framework that assigns each layer an adaptive compensation coefficient by combining two complementary signals: an intrinsic vulnerability score from weight geometry and a calibration reliability score from the data-driven solution. The resulting layer-wise coefficient balances local quantization fidelity against cross-layer error correction, enabling tailored compensation without retraining or hyperparameter search. Experiments on Whisper, Moonshine, and Qwen3-ASR across four benchmarks show that FADE consistently improves mean Word Error Rate over strong baselines at both 3- and 4-bit precision while substantially reducing run-to-run variance.

We present NeuroSPICE, a physics-informed neural network (PINN) framework for device and circuit simulation. Unlike conventional SPICE, which relies on time-discretized numerical solvers, NeuroSPICE leverages PINNs to solve circuit differential-algebraic equations (DAEs) by minimizing the residual of the equations through backpropagation. It models device and circuit waveforms using analytical equations in time domain with exact temporal derivatives. While PINNs do not outperform SPICE in speed or accuracy during training, they offer unique advantages such as surrogate models for design optimization and inverse problems. NeuroSPICE's flexibility enables the simulation of emerging devices, including highly nonlinear systems such as ferroelectric memories.

Learning dexterous bimanual manipulation policies critically depends on large-scale, high-quality demonstrations, yet current paradigms face inherent trade-offs: teleoperation provides physically grounded data but is prohibitively labor-intensive, while simulation-based synthesis scales efficiently but suffers from sim-to-real gaps. We present BiDemoSyn, a framework that synthesizes contact-rich, physically feasible bimanual demonstrations from a single real-world example. The key idea is to decompose tasks into invariant coordination blocks and variable, object-dependent adjustments, then adapt them through vision-guided alignment and lightweight trajectory optimization. This enables the generation of thousands of diverse and feasible demonstrations within several hours, without repeated teleoperation or reliance on imperfect simulation. Across six dual-arm tasks, we show that policies trained on BiDemoSyn data generalize robustly to novel object poses and shapes, significantly outperforming recent strong baselines. Beyond the one-shot setting, BiDemoSyn naturally extends to few-shot-based synthesis, improving object-level diversity and out-of-distribution generalization while maintaining strong data efficiency. Moreover, policies trained on BiDemoSyn data exhibit zero-shot cross-embodiment transfer to new robotic platforms, enabled by object-centric observations and a simplified 6-DoF end-effector action representation that decouples policies from embodiment-specific dynamics. By bridging the gap between efficiency and real-world fidelity, BiDemoSyn provides a scalable path toward practical imitation learning for complex bimanual manipulation without compromising physical grounding.

Visual and textual soft prompt tuning can effectively improve the adaptability of Vision-Language Models (VLMs) in downstream tasks. However, fine-tuning on video tasks impairs the model's generalization ability to unseen classes. Existing methods attempt to mitigate this forgetting effect by regularizing the gap between hand-crafted prompts and soft prompts, but this also weakens the learning ability of soft prompts. To address this challenge, we propose a plug-and-play coupling prompt learning framework to optimize the generalization performance of V-L models in video tasks, with the core motivation of mitigating semantic space narrowing during fine-tuning by introducing an externally supervised prompt. Specifically, for textual prompts, we introduce pre-trained prompts from other datasets as hard prompt tokens. These are concatenated with soft prompt tokens and coupled via a learnable mapping layer. This competitive prompting approach prevents the semantic space from overfitting to supervised categories. In addition, we introduce a set of well-designed irrelevant video sets and negative prompts as generic attribute anchors to maintain the generic relevance of the attributes in the pre-trained semantic space, thus preserving the generalization ability. Experiments on video tasks demonstrate that our method significantly outperforms state-of-the-art prompt tuning approaches across generalization benchmarks, particularly on base-to-new class prediction.

In computer vision, machine unlearning aims to remove the influence of specific visual concepts or training images without retraining from scratch. Studies show that existing approaches often modify the classifier while leaving internal representations intact, resulting in incomplete forgetting. In this work, we extend the notion of unlearning to the representation level, deriving a three-term interplay between forgetting efficacy, retention fidelity, and class separation. Building on Neural Collapse theory, we show that the orthogonal projection of a simplex Equiangular Tight Frame (ETF) remains an ETF in a lower dimensional space, yielding a provably optimal forgetting operator. We further introduce the Representation Unlearning Score (RUS) to quantify representation-level forgetting and retention fidelity. Building on this, we introduce POUR (Provably Optimal Unlearning of Representations), a geometric projection method with closed-form (POUR-P) and a feature-level unlearning variant under a distillation scheme (POUR-D). Experiments on CIFAR-10/100 and PathMNIST demonstrate that POUR achieves effective unlearning while preserving retained knowledge, outperforming state-of-the-art unlearning methods on both classification-level and representation-level metrics.

Distributed Fiber Optic Sensing (DFOS) is promising for long-range perimeter security, yet practical deployment faces three key obstacles: severe cross-deployment domain shift, scarce or unavailable labels at new sites, and limited within-class coverage even in source deployments. We propose DUPLE, a prototype-based meta-learning framework tailored for cross-deployment DFOS recognition. The core idea is to jointly exploit complementary time- and frequency-domain cues and adapt class representations to sample-specific statistics: (i) a dual-domain learner constructs multi-prototype class representations to cover intra-class heterogeneity; (ii) a lightweight statistical guidance mechanism estimates the reliability of each domain from raw signal statistics; and (iii) a query-adaptive aggregation strategy selects and combines the most relevant prototypes for each query. Extensive experiments on two real-world cross-deployment benchmarks demonstrate consistent improvements over strong deep learning and meta-learning baselines, achieving more accurate and stable recognition under label-scarce target deployments.

Robotic Foundation Models (RFMs) hold great promise as generalist, end-to-end systems for robot control. Yet their ability to generalize across new environments, tasks, and embodiments remains limited. We argue that a major bottleneck lies in their foundations: most RFMs are built by fine-tuning internet-pretrained Vision-Language Models (VLMs). However, these VLMs are trained on 2D image-language tasks and lack the 3D spatial reasoning inherently required for embodied control in the 3D world. Bridging this gap directly with large-scale robotic data is costly and difficult to scale. Instead, we propose to enrich easy-to-collect non-robotic image data with 3D annotations and enhance a pretrained VLM with 3D understanding capabilities. Following this strategy, we train SPEAR-VLM, a 3D-aware VLM that infers object coordinates in 3D space from a single 2D image. Building on SPEAR-VLM, we introduce our main contribution, $~\textbf{SPEAR-1}$: a robotic foundation model that integrates grounded 3D perception with language-instructed embodied control. Trained on $\sim$45M frames from 24 Open X-Embodiment datasets, SPEAR-1 outperforms or matches state-of-the-art models such as $π_0$-FAST and $π_{0.5}$, while it uses 20$\times$ fewer robot demonstrations. This carefully-engineered training strategy unlocks new VLM capabilities and as a consequence boosts the reliability of embodied control beyond what is achievable with only robotic data. We make our model weights and 3D-annotated datasets publicly available at https://spear.insait.ai.

Heart failure (HF) discharge planning depends on identifying patients at risk of deterioration or death, yet accurate prediction from routinely collected electronic health records (EHRs) remains challenging. We developed and validated sequence models for three one-year prediction tasks in a Swedish HF cohort (N = 42,820): clinical instability (a rehospitalization phenotype) and mortality after the initial in-hospital HF diagnosis, and mortality after the latest hospitalization. A modular three-component framework transforms structured EHRs into patient sequences by specifying tokenization strategies, temporal representations, and model configurations. Patient data included diagnoses, vital signs, laboratories, medications, and procedures. Autoregressive next-token prediction models consistently outperformed alternative objectives in short-context settings (<= 512 tokens). The best model (Llama) achieved AUPRCs (95% CI) of 0.555 (0.535-0.575), 0.582 (0.558-0.608), and 0.854 (0.842-0.865), with robust calibration. Ablations show Llama and Mamba variants learn efficient patient representations, with tiny configurations surpassing larger conventional baselines, indicating that model size alone does not improve performance. With limited clinical concepts or training data, Llama maintains strong performance, frequently surpassing full-data baselines. Combining clinical instability and mortality predictions defines four distinct care pathways, from standard primary care to intensive home care, supporting patient-centered decisions at discharge. These findings demonstrate accurate risk prediction from routine hospital data, provide actionable development guidance, and support post-discharge risk stratification.

Developing reliable iris recognition and presentation attack detection methods requires diverse datasets that capture realistic variations in iris features and a wide spectrum of anomalies. Because of the rich texture of iris images, which spans a wide range of spatial frequencies, synthesizing same-identity iris images while controlling specific attributes remains challenging. In this work, we introduce a new iris image augmentation strategy by traversing a generative model's latent space toward latent codes that represent same-identity samples but with some desired iris image properties manipulated. The latent space traversal is guided by a gradient of specific geometrical, textural, or quality-related iris image features (e.g., sharpness, pupil size, iris size, or pupil-to-iris ratio) and preserves the identity represented by the image being manipulated. The proposed approach can be easily extended to manipulate any attribute for which a differentiable loss term can be formulated. Additionally, our approach can use either randomly generated images using either a pre-train GAN model or real-world iris images. We can utilize GAN inversion to project any given iris image into the latent space and obtain its corresponding latent code.

We propose patching for large language models (LLMs) like software versions, a lightweight and modular approach for addressing safety vulnerabilities. While vendors release improved LLM versions, major releases are costly, infrequent, and difficult to tailor to customer needs, leaving released models with known safety gaps. Unlike full-model fine-tuning or major version updates, our method enables rapid remediation by prepending a compact, learnable prefix to an existing model. This "patch" introduces only 0.003% additional parameters, yet reliably steers model behavior toward that of a safer reference model. Across three critical domains (toxicity mitigation, bias reduction, and harmfulness refusal) policy patches achieve safety improvements comparable to next-generation safety-aligned models while preserving fluency. Our results demonstrate that LLMs can be "patched" much like software, offering vendors and practitioners a practical mechanism for distributing scalable, efficient, and composable safety updates between major model releases.

A central feature of many deliberative processes, such as citizens' assemblies and deliberative polls, is the opportunity for participants to engage directly with experts. While participants are typically invited to propose questions for expert panels, only a limited number can be selected due to time constraints. This raises the challenge of how to choose a small set of questions that best represent the interests of all participants. We introduce an auditing framework for measuring the level of representation provided by a slate of questions, based on the social choice concept known as justified representation (JR). We present the first algorithms for auditing JR in the general utility setting, with our most efficient algorithm achieving a runtime of $O(mn\log n)$, where $n$ is the number of participants and $m$ is the number of proposed questions. We apply our auditing methods to historical deliberations, comparing the representativeness of (a) the actual questions posed to the expert panel (chosen by a moderator), (b) participants' questions chosen via integer linear programming, (c) summary questions generated by large language models (LLMs). Our results highlight both the promise and current limitations of LLMs in supporting deliberative processes. By integrating our methods into an online deliberation platform that has been used for over hundreds of deliberations across more than 50 countries, we make it easy for practitioners to audit and improve representation in future deliberations.

Ongoing breakthroughs in large language models (LLMs) are reshaping scholarly search and discovery interfaces. While these systems offer new possibilities for navigating scientific knowledge, they also raise concerns about fairness and representational bias rooted in the models' memorized training data. As LLMs are increasingly used to answer queries about researchers and research communities, their ability to accurately reconstruct scholarly coauthor lists becomes an important but underexamined issue. In this study, we investigate how memorization in LLMs affects the reconstruction of coauthor lists and whether this process reflects existing inequalities across academic disciplines and world regions. We evaluate three prominent models, DeepSeek R1, Llama 4 Scout, and Mixtral 8x7B, by comparing their generated coauthor lists against bibliographic reference data. Our analysis reveals a systematic advantage for highly cited researchers, indicating that LLM memorization disproportionately favors already visible scholars. However, this pattern is not uniform: certain disciplines, such as Clinical Medicine, and some regions, including parts of Africa, exhibit more balanced reconstruction outcomes. These findings highlight both the risks and limitations of relying on LLM-generated relational knowledge in scholarly discovery contexts and emphasize the need for careful auditing of memorization-driven biases in LLM-based systems.

Using Reinforcement Learning with Verifiable Rewards (RLVR) to optimize Large Language Models (LLMs) can be conceptualized as progressively editing a query's `Reasoning Tree'. This process involves exploring nodes (tokens) and dynamically modifying the model's policy at each node. When combined with data scheduling, this process yields further gains in data efficiency and accuracy. However, existing RLVR data scheduling methods typically rely on path-based metrics to rank queries, overlooking the reasoning tree structures of these queries. In this paper, we introduce a novel metric, namely Reasoning Score (r-score), which measures the query's learning difficulty based on the structure of its reasoning tree. Based on the r-score, we propose the Reasoning Tree Schedule (Re-Schedule), a scheduling algorithm that constructs a curriculum progressing from structurally simple (high r-score) to complex (low r-score) queries. Experiments on six math-reasoning benchmarks show that Re-Schedule significantly improves average accuracy, achieving gains of up to 3.2%. These strong results validate our approach and demonstrate that a structural understanding of the reasoning tree provides a more powerful and principled foundation for RLVR data scheduling.

Eigenvalue problems are among the most important topics in many scientific disciplines. With the recent surge and development of machine learning, neural eigenvalue methods have attracted significant attention as a forward pass of inference requires only a tiny fraction of the computation time compared to traditional solvers. However, a key limitation is the requirement for large amounts of labeled data in training, including operators and their eigenvalues. To tackle this limitation, we propose a novel method, named Sorting Chebyshev Subspace Filter (SCSF), which significantly accelerates eigenvalue data generation by leveraging similarities between operators -- a factor overlooked by existing methods. Specifically, SCSF employs truncated fast Fourier transform sorting to group operators with similar eigenvalue distributions and constructs a Chebyshev subspace filter that leverages eigenpairs from previously solved problems to assist in solving subsequent ones, reducing redundant computations. To the best of our knowledge, SCSF is the first method to accelerate eigenvalue data generation. Experimental results show that SCSF achieves up to a 3.5 times speedup compared to various numerical solvers.

Detecting out-of-distribution (OOD) data is critical for machine learning, be it for safety reasons or to enable open-ended learning. However, beyond mere detection, choosing an appropriate course of action typically hinges on the type of OOD data encountered. Unfortunately, the latter is generally not distinguished in practice, as modern OOD detection methods collapse distributional shifts into single scalar outlier scores. This work argues that scalar-based methods are thus insufficient for OOD data to be properly contextualized and prospectively exploited, a limitation we overcome with the introduction of DISC: Diffusion-based Statistical Characterization. DISC leverages the iterative denoising process of diffusion models to extract a rich, multi-dimensional feature vector that captures statistical discrepancies across multiple noise levels. Extensive experiments on image and tabular benchmarks show that DISC matches or surpasses state-of-the-art detectors for OOD detection and, crucially, also classifies OOD type, a capability largely absent from prior work. As such, our work enables a shift from simple binary OOD detection to a more granular detection.

Retrieval-augmented generation (RAG) has demonstrated its ability to enhance Large Language Models (LLMs) by integrating external knowledge sources. However, multi-hop questions, which require the identification of multiple knowledge targets to form a synthesized answer, raise new challenges for RAG systems. Under the multi-hop settings, existing methods often struggle to fully understand the questions with complex semantic structures and are susceptible to irrelevant noise during the retrieval of multiple information targets. To address these limitations, we propose a novel graph representation learning framework for multi-hop question retrieval. We first introduce a Multi-information Level Knowledge Graph (Multi-L KG) to model various information levels for a more comprehensive understanding of multi-hop questions. Based on this, we design a Question-Adaptive Graph Neural Network (Quest-GNN) for representation learning on the Multi-L KG. Quest-GNN employs intra/inter-level message passing mechanisms, and in each message passing the information aggregation is guided by the question, which not only facilitates multi-granular information aggregation but also significantly reduces the impact of noise. To enhance its ability to learn robust representations, we further propose two synthesized data generation strategies for pre-training the Quest-GNN. Extensive experimental results demonstrate the effectiveness of our framework in multi-hop scenarios, especially in high-hop questions the improvement can reach 33.8\%. The code is available at: https://github.com/Jerry2398/QSGNN.

Multi-label classification (MLC) offers a more comprehensive semantic understanding of Remote Sensing (RS) imagery compared to traditional single-label classification (SLC). However, obtaining complete annotations for MLC is particularly challenging due to the complexity and high cost of the labeling process. As a practical alternative, single-positive multi-label learning (SPML) has emerged, where each image is annotated with only one relevant label, and the model is expected to recover the full set of labels. While scalable, SPML introduces significant supervision ambiguity, demanding specialized solutions for model training. Although various SPML methods have been proposed in the computer vision domain, research in the RS context remains limited. To bridge this gap, we propose Adaptive Gradient Calibration (AdaGC), a novel and generalizable SPML framework tailored to RS imagery. AdaGC adopts a gradient calibration (GC) mechanism with a dual exponential moving average (EMA) module for robust pseudo-label generation. We introduce a theoretically grounded, training-dynamics-based indicator to adaptively trigger GC, which ensures GC's effectiveness by preventing it from being affected by model underfitting or overfitting to label noise. Extensive experiments on two benchmark RS datasets under two distinct label noise types demonstrate that AdaGC achieves state-of-the-art (SOTA) performance while maintaining strong robustness across diverse settings. The codes and data will be released at https://github.com/rslab-unitrento/AdaGC.

Deep learning has become increasingly important in remote sensing image classification due to its ability to extract semantic information from complex data. Classification tasks often include predefined label hierarchies that represent the semantic relationships among classes. However, these hierarchies are frequently overlooked, and most approaches focus only on fine-grained classification schemes. In this paper, we present a novel Semantics-Aware Hierarchical Consensus (SAHC) approach to learn hierarchical features and relationships by integrating hierarchy-specific classification heads within a deep network architecture, each specialized in different degrees of class granularity. The proposed approach employs trainable hierarchy matrices, which guide the network through the learning of the hierarchical structure in a self-consistent manner. Furthermore, we introduce a hierarchical consensus mechanism to ensure aligned probability distributions across different hierarchical levels. This mechanism acts as a weighted ensemble being able to effectively leverage the inherent structure of the hierarchical classification task. The proposed SAHC method is evaluated on two benchmark datasets with different degrees of hierarchical complexity on different tasks, considering varying spectral and spatial resolutions. Experimental results show both the effectiveness of the proposed approach in guiding network learning and the robustness of the hierarchical consensus for remote sensing image classification tasks. The codes will be released at https://github.com/rslab-unitrento/sahc.

Compositional generalization, the ability to generate novel combinations of known concepts, is a key ingredient for visual generative models. Yet, not all mechanisms that enable or inhibit it are fully understood. In this work, we conduct a systematic study of how various design choices influence compositional generalization in image and video generation in a positive or negative way. Through controlled experiments, we identify two key factors: (i) whether the training objective operates on a discrete or continuous distribution, and (ii) to what extent conditioning provides information about the constituent concepts during training. Building on these insights, we show that relaxing the MaskGIT discrete loss with an auxiliary continuous JEPA-based objective can improve compositional performance in discrete models like MaskGIT.

Recent advances in causal interpretability have extended from language models to vision-language models (VLMs), seeking to reveal their internal mechanisms through input interventions. While textual interventions often target semantics, visual interventions typically rely on coarse pixel-level perturbations, limiting semantic insights on multimodal integration. In this study, we introduce V-SEAM, a novel framework that combines Visual Semantic Editing and Attention Modulating for causal interpretation of VLMs. V-SEAM enables concept-level visual manipulations and identifies attention heads with positive or negative contributions to predictions across three semantic levels: objects, attributes, and relationships. We observe that positive heads are often shared within the same semantic level but vary across levels, while negative heads tend to generalize broadly. Finally, we introduce an automatic method to modulate key head embeddings, demonstrating enhanced performance for both LLaVA and InstructBLIP across three diverse VQA benchmarks. Our data and code are released at: https://github.com/petergit1/V-SEAM.

Questionnaire-based surveys are foundational to social science research and public policymaking, yet traditional survey methods remain costly, time-consuming, and often limited in scale. Although prior work has explored large language models (LLMs) as virtual survey respondents, existing studies often address narrow task settings, focus on single sociological domains, or lack a unified evaluation framework that enables systematic comparison across diverse datasets and models. To address these gaps, we introduce two complementary task abstractions: Partial Attribute Simulation (PAS), where LLMs predict missing attributes from incomplete respondent profiles, and Full Attribute Simulation (FAS), where LLMs generate complete synthetic datasets under zero-context and context-enhanced conditions. Both are framed as diagnostic and exploratory tools rather than replacements for human data collection. We curate LLM-S^3 (Large Language Model-based Sociodemographic Survey Simulation), a benchmark spanning 11 real-world public datasets across four sociological domains, and evaluate GPT-3.5/4 Turbo and LLaMA 3.0/3.1-8B under zero-shot and few-shot settings. Our evaluation reveals consistent performance trends across model families, highlights failure modes in structured output generation, and demonstrates how context and prompt design affect simulation fidelity. Our code and dataset are available at: https://github.com/dart-lab-research/LLM-S-Cube-Benchmark

With the development of large-scale diffusion-based and language-modeling-based generative models, impressive progress has been achieved in text-to-audio generation. Despite producing high-quality outputs, existing text-to-audio models mainly aim to generate semantically aligned sound and fall short of controlling fine-grained acoustic characteristics of specific sounds. As a result, users who need specific sound content may find it difficult to generate the desired audio clips. In this paper, we present DreamAudio for customized text-to-audio generation (CTTA). Specifically, we introduce a new framework that is designed to enable the model to identify auditory information from user-provided reference concepts for audio generation. Given a few reference audio samples containing personalized audio events, our system can generate new audio samples that include these specific events. In addition, two types of datasets are developed for training and testing the proposed systems. The experiments show that DreamAudio generates audio samples that are highly consistent with the customized audio features and aligned well with the input text prompts. Furthermore, DreamAudio offers comparable performance in general text-to-audio tasks. We also provide a human-involved dataset containing audio events from real-world CTTA cases as the benchmark for customized generation tasks.

Over half of US adults with Alzheimer disease and related dementias remain undiagnosed, and speech-based screening offers a scalable detection approach. We compared large language model adaptation strategies for dementia detection using the DementiaBank speech corpus, evaluating nine text-only models and three multimodal audio-text models on recordings from DementiaBank speech corpus. Adaptations included in-context learning with different demonstration selection policies, reasoning-augmented prompting, parameter-efficient fine-tuning, and multimodal integration. Results showed that class-centroid demonstrations achieved the highest in-context learning performance, reasoning improved smaller models, and token-level fine-tuning generally produced the best scores. Adding a classification head substantially improved underperforming models. Among multimodal models, fine-tuned audio-text systems performed well but did not surpass the top text-only models. These findings highlight that model adaptation strategies, including demonstration selection, reasoning design, and tuning method, critically influence speech-based dementia detection, and that properly adapted open-weight models can match or exceed commercial systems.