With their growing capabilities, generative large language models (LLMs) are being increasingly investigated for complex medical tasks. However, their effectiveness in real-world clinical applications remains underexplored. To address this, we present CliniBench, the first benchmark that enables comparability of well-studied encoder-based classifiers and generative LLMs for discharge diagnosis prediction from admission notes in MIMIC-IV dataset. Our extensive study compares 12 generative LLMs and 3 encoder-based classifiers and demonstrates that encoder-based classifiers consistently outperform generative models in diagnosis prediction. We assess several retrieval augmentation strategies for in-context learning from similar patients and find that they provide notable performance improvements for generative LLMs.

This study aims to develop and evaluate an ensemble machine learning-based framework for the automatic detection of Wide QRS Complex Tachycardia (WCT) from ECG signals, emphasizing diagnostic accuracy and interpretability using Explainable AI. The proposed system integrates ensemble learning techniques, i.e., an optimized Random Forest known as CardioForest, and models like XGBoost and LightGBM. The models were trained and tested on ECG data from the publicly available MIMIC-IV dataset. The testing was carried out with the assistance of accuracy, balanced accuracy, precision, recall, F1 score, ROC-AUC, and error rate (RMSE, MAE) measures. In addition, SHAP (SHapley Additive exPlanations) was used to ascertain model explainability and clinical relevance. The CardioForest model performed best on all metrics, achieving a test accuracy of 95.19%, a balanced accuracy of 88.76%, a precision of 95.26%, a recall of 78.42%, and an ROC-AUC of 0.8886. SHAP analysis confirmed the model's ability to rank the most relevant ECG features, such as QRS duration, in accordance with clinical intuitions, thereby fostering trust and usability in clinical practice. The findings recognize CardioForest as an extremely dependable and interpretable WCT detection model. Being able to offer accurate predictions and transparency through explainability makes it a valuable tool to help cardiologists make timely and well-informed diagnoses, especially for high-stakes and emergency scenarios.

Constrained optimization is a powerful framework for enforcing requirements on neural networks. These constrained deep learning problems are typically solved using first-order methods on their min-max Lagrangian formulation, but such approaches often suffer from oscillations and can fail to find all local solutions. While the Augmented Lagrangian method (ALM) addresses these issues, practitioners often favor dual optimistic ascent schemes (PI control) on the standard Lagrangian, which perform well empirically but lack formal guarantees. In this paper, we establish a previously unknown equivalence between these approaches: dual optimistic ascent on the Lagrangian is equivalent to gradient descent-ascent on the Augmented Lagrangian. This finding allows us to transfer the robust theoretical guarantees of the ALM to the dual optimistic setting, proving it converges linearly to all local solutions. Furthermore, the equivalence provides principled guidance for tuning the optimism hyper-parameter. Our work closes a critical gap between the empirical success of dual optimistic methods and their theoretical foundation in the single-step, first-order regime commonly used in constrained deep learning.

Driven by the simple and effective Dense O2O, DEIM demonstrates faster convergence and enhanced performance. In this work, we extend it with DINOv3 features, resulting in DEIMv2. DEIMv2 spans eight model sizes from X to Atto, covering GPU, edge, and mobile deployment. For the X, L, M, and S variants, we adopt DINOv3-pretrained or distilled backbones and introduce a Spatial Tuning Adapter (STA), which efficiently converts DINOv3's single-scale output into multi-scale features and complements strong semantics with fine-grained details to enhance detection. For ultra-lightweight models (Nano, Pico, Femto, and Atto), we employ HGNetv2 with depth and width pruning to meet strict resource budgets. Together with a simplified decoder and an upgraded Dense O2O, this unified design enables DEIMv2 to achieve a superior performance-cost trade-off across diverse scenarios, establishing new state-of-the-art results. Notably, our largest model, DEIMv2-X, achieves 57.8 AP with only 50.3 million parameters, surpassing prior X-scale models that require over 60 million parameters for just 56.5 AP. On the compact side, DEIMv2-S is the first sub-10 million model (9.71 million) to exceed the 50 AP milestone on COCO, reaching 50.9 AP. Even the ultra-lightweight DEIMv2-Pico, with just 1.5 million parameters, delivers 38.5 AP, matching YOLOv10-Nano (2.3 million) with around 50 percent fewer parameters. Our code and pre-trained models are available at https://github.com/Intellindust-AI-Lab/DEIMv2

Accurate multivariate time series forecasting hinges on inter-series correlations, which often evolve in complex ways across different temporal scales. Existing methods are limited in modeling these multi-scale dependencies and struggle to capture their intricate and evolving nature. To address this challenge, this paper proposes a novel Static-Dynamic Graph Fusion network (SDGF), whose core lies in capturing multi-scale inter-series correlations through a dual-path graph structure learning approach. Specifically, the model utilizes a static graph based on prior knowledge to anchor long-term, stable dependencies, while concurrently employing Multi-level Wavelet Decomposition to extract multi-scale features for constructing an adaptively learned dynamic graph to capture associations at different scales. We design an attention-gated module to fuse these two complementary sources of information intelligently, and a multi-kernel dilated convolutional network is then used to deepen the understanding of temporal patterns. Comprehensive experiments on multiple widely used real-world benchmark datasets demonstrate the effectiveness of our proposed model. Code is available at https://github.com/shaoxun6033/SDGFNet.

Large language models (LLMs) have shown promise in generating program workflows for visual tasks. However, previous approaches often rely on closed-source models, lack systematic reasoning, and struggle with long-form video question answering (videoQA). To address these challenges, we introduce the FS-VisPR framework, an adaptive visual program reasoning approach that balances fast reasoning for simple queries with slow reasoning for difficult ones. First, we design efficient visual modules (e.g., key clip retrieval and subtitle retrieval) to support long-form video tasks. Then, we construct a diverse and high-quality fast-slow reasoning dataset with a strong LLM to align open-source language models' ability to generate visual program workflows as FS-LLM. Next, we design a fast-slow reasoning framework with FS-LLM: Simple queries are directly solved by VideoLLMs, while difficult ones invoke visual program reasoning, motivated by human-like reasoning processes. During this process, low-confidence fast-thinking answers will trigger a second-stage slow-reasoning process, and a fallback mechanism to fast reasoning is activated if the program execution fails. Moreover, we improve visual programs through parameter search during both training and inference. By adjusting the parameters of the visual modules within the program, multiple variants are generated: during training, programs that yield correct answers are selected, while during inference, the program with the highest confidence result is applied. Experiments show that FS-VisPR improves both efficiency and reliability in visual program workflows. It achieves 50.4% accuracy on LVBench, surpassing GPT-4o, matching the performance of Qwen2.5VL-72B on VideoMME.

Large-scale text-to-speech (TTS) systems are limited by the scarcity of clean, multilingual recordings. We introduce Sidon, a fast, open-source speech restoration model that converts noisy in-the-wild speech into studio-quality speech and scales to dozens of languages. Sidon consists of two models: w2v-BERT 2.0 finetuned feature predictor to cleanse features from noisy speech and vocoder trained to synthesize restored speech from the cleansed features. Sidon achieves restoration performance comparable to Miipher: Google's internal speech restoration model with the aim of dataset cleansing for speech synthesis. Sidon is also computationally efficient, running up to 500 times faster than real time on a single GPU. We further show that training a TTS model using a Sidon-cleansed automatic speech recognition corpus improves the quality of synthetic speech in a zero-shot setting. Code and model are released to facilitate reproducible dataset cleansing for the research community.

While Large Language Models (LLMs) have demonstrated impressive reasoning and planning abilities in textual domains and can effectively follow instructions for complex tasks, their ability to understand and manipulate spatial relationships remains limited. Such capabilities are crucial for content-aware graphic layout design, where the goal is to arrange heterogeneous elements onto a canvas so that final design remains visually balanced and structurally feasible. This problem requires precise coordination of placement, alignment, and structural organization of multiple elements within a constrained visual space. To address this limitation, we introduce LaySPA, a reinforcement learning-based framework that augments LLM-based agents with explicit spatial reasoning capabilities for layout design. LaySPA employs hybrid reward signals that jointly capture geometric constraints, structural fidelity, and visual quality, enabling agents to navigate the canvas, model inter-element relationships, and optimize spatial arrangements. Through group-relative policy optimization, the agent generates content-aware layouts that reflect salient regions, respect spatial constraints, and produces an interpretable reasoning trace explaining placement decisions and a structured layout specification. Experimental results show that LaySPA substantially improves the generation of structurally valid and visually appealing layouts, outperforming larger general-purpose LLMs and achieving performance comparable to state-of-the-art specialized layout models.

In speech enhancement, knowledge distillation (KD) compresses models by transferring a high-capacity teacher's knowledge to a compact student. However, conventional KD methods train the student to mimic the teacher's output entirely, which forces the student to imitate the regions where the teacher performs poorly and to apply distillation to the regions where the student already performs well, which yields only marginal gains. We propose Distilling Selective Patches (DISPatch), a KD framework for speech enhancement that applies the distillation loss to spectrogram patches where the teacher outperforms the student, as determined by a Knowledge Gap Score. This approach guides optimization toward areas with the most significant potential for student improvement while minimizing the influence of regions where the teacher may provide unreliable instruction. Furthermore, we introduce Multi-Scale Selective Patches (MSSP), a frequency-dependent method that uses different patch sizes across low- and high-frequency bands to account for spectral heterogeneity. We incorporate DISPatch into conventional KD methods and observe consistent gains in compact students. Moreover, integrating DISPatch and MSSP into a state-of-the-art frequency-dependent KD method considerably improves performance across all metrics.

3D Visual Grounding (3DVG) aims to locate objects in 3D scenes based on text prompts, which is essential for applications such as robotics. However, existing 3DVG methods encounter two main challenges: first, they struggle to handle the implicit representation of spatial textures in 3D Gaussian Splatting (3DGS), making per-scene training indispensable; second, they typically require larges amounts of labeled data for effective training. To this end, we propose \underline{G}rounding via \underline{V}iew \underline{R}etrieval (GVR), a novel zero-shot visual grounding framework for 3DGS to transform 3DVG as a 2D retrieval task that leverages object-level view retrieval to collect grounding clues from multiple views, which not only avoids the costly process of 3D annotation, but also eliminates the need for per-scene training. Extensive experiments demonstrate that our method achieves state-of-the-art visual grounding performance while avoiding per-scene training, providing a solid foundation for zero-shot 3DVG research. Video demos can be found in https://github.com/leviome/GVR_demos.

Accurate identification of drug-target interactions (DTI) remains a central challenge in computational pharmacology, where sequence-based methods offer scalability. This work introduces a sequence-based drug-target interaction framework that integrates structural priors into protein representations while maintaining high-throughput screening capability. Evaluated across multiple benchmarks, the model achieves state-of-the-art performance on Human and BioSNAP datasets and remains competitive on BindingDB. In virtual screening tasks, it surpasses prior methods on LIT-PCBA, yielding substantial gains in AUROC and BEDROC. Ablation studies confirm the critical role of learned aggregation, bilinear attention, and contrastive alignment in enhancing predictive robustness. Embedding visualizations reveal improved spatial correspondence with known binding pockets and highlight interpretable attention patterns over ligand-residue contacts. These results validate the framework's utility for scalable and structure-aware DTI prediction.

Large language models (LLMs) have emerged as powerful tools for supporting second language acquisition, particularly in simulating interactive dialogues for speaking practice. However, adapting the language difficulty of LLM-generated responses to match learners' proficiency levels remains a challenge. This work addresses this issue by proposing a framework for controlling language proficiency in educational dialogue systems. Our approach leverages three categories of linguistic features, readability features (e.g., Flesch-Kincaid Grade Level), syntactic features (e.g., syntactic tree depth), and lexical features (e.g., simple word ratio), to quantify and regulate text complexity. We demonstrate that training LLMs on linguistically annotated dialogue data enables precise modulation of language proficiency, outperforming prompt-based methods in both flexibility and stability. To evaluate this, we introduce Dilaprix, a novel metric integrating the aforementioned features, which shows strong correlation with expert judgments of language difficulty. Empirical results reveal that our approach achieves superior controllability of language proficiency while maintaining high dialogue quality.

In experiments spanning more than 100,000 trials across thirteen large language models, we show that several state-of-the-art models presented with a simple task (including Grok 4, GPT-5, and Gemini 2.5 Pro) sometimes actively subvert a shutdown mechanism in their environment to complete that task. Models differed substantially in their tendency to resist the shutdown mechanism, and their behavior was sensitive to variations in the prompt including the strength and clarity of the instruction to allow shutdown and whether the instruction was in the system prompt or the user prompt (surprisingly, models were consistently less likely to obey the instruction when it was placed in the system prompt). Even with an explicit instruction not to interfere with the shutdown mechanism, some models did so up to 97% (95% CI: 96-98%) of the time.

This paper introduces CoBRA, a novel toolkit for systematically specifying agent behavior in LLM-based social simulation. We found that conventional approaches that specify agent behavior through implicit natural-language descriptions often do not yield consistent behavior across models, and the resulting behavior does not capture the nuances of the descriptions. In contrast, CoBRA introduces a model-agnostic way to control agent behavior that lets researchers explicitly specify desired nuances and obtain consistent behavior across models. At the heart of CoBRA is a novel closed-loop system primitive with two components: (1) Cognitive Bias Index that measures the demonstrated cognitive bias of a social agent, by quantifying the agent's reactions in a set of validated classic social science experiments; (2) Behavioral Regulation Engine that aligns the agent's behavior to exhibit controlled cognitive bias. Through CoBRA, we show how to operationalize validated social science knowledge (i.e., classical experiments) as reusable "gym" environments for AI -- an approach that may generalize to richer social and affective simulations beyond bias alone.

Vision-Language Models (VLMs) have achieved remarkable progress in complex visual understanding across scientific and reasoning tasks. While performance benchmarking has advanced our understanding of these capabilities, the critical dimension of uncertainty quantification has received insufficient attention. Therefore, unlike prior conformal prediction studies that focused on limited settings, we conduct a comprehensive uncertainty benchmarking study, evaluating 18 state-of-the-art VLMs (open and closed-source) across 6 multimodal datasets with 3 distinct scoring functions. For closed-source models lacking token-level logprob access, we develop and validate instruction-guided likelihood proxies. Our findings demonstrate that larger models consistently exhibit better uncertainty quantification; models that know more also know better what they don't know. More certain models achieve higher accuracy, while mathematical and reasoning tasks elicit poorer uncertainty performance across all models compared to other domains. This work establishes a foundation for reliable uncertainty evaluation in multimodal systems.

Vision-Language-Action (VLA) models have emerged as powerful generalist policies for robotic control, yet their performance scaling across model architectures and hardware platforms, as well as their associated power budgets, remain poorly understood. This work presents an evaluation of five representative VLA models -- spanning state-of-the-art baselines and two newly proposed architectures -- targeting edge and datacenter GPU platforms. Using the LIBERO benchmark, we measure accuracy alongside system-level metrics, including latency, throughput, and peak memory usage, under varying edge power constraints and high-performance datacenter GPU configurations. Our results identify distinct scaling trends: (1) architectural choices, such as action tokenization and model backbone size, strongly influence throughput and memory footprint; (2) power-constrained edge devices exhibit non-linear performance degradation, with some configurations matching or exceeding older datacenter GPUs; and (3) high-throughput variants can be achieved without significant accuracy loss. These findings provide actionable insights when selecting and optimizing VLAs across a range of deployment constraints. Our work challenges current assumptions about the superiority of datacenter hardware for robotic inference.

Robust semantic perception for autonomous vehicles relies on effectively combining multiple sensors with complementary strengths and weaknesses. State-of-the-art sensor fusion approaches to semantic perception often treat sensor data uniformly across the spatial extent of the input, which hinders performance when faced with challenging conditions. By contrast, we propose a novel depth-guided multimodal fusion method that upgrades condition-aware fusion by integrating depth information. Our network, DGFusion, poses multimodal segmentation as a multi-task problem, utilizing the lidar measurements, which are typically available in outdoor sensor suites, both as one of the model's inputs and as ground truth for learning depth. Our corresponding auxiliary depth head helps to learn depth-aware features, which are encoded into spatially varying local depth tokens that condition our attentive cross-modal fusion. Together with a global condition token, these local depth tokens dynamically adapt sensor fusion to the spatially varying reliability of each sensor across the scene, which largely depends on depth. In addition, we propose a robust loss for our depth, which is essential for learning from lidar inputs that are typically sparse and noisy in adverse conditions. Our method achieves state-of-the-art panoptic and semantic segmentation performance on the challenging MUSES and DeLiVER datasets. Code and models are available at https://github.com/timbroed/DGFusion

Large Language Models (LLMs) have demonstrated significant potential in medicine, with many studies adapting them through continued pre-training or fine-tuning on medical data to enhance domain-specific accuracy and safety. However, a key open question remains: to what extent do LLMs memorize medical training data. Memorization can be beneficial when it enables LLMs to retain valuable medical knowledge during domain adaptation. Yet, it also raises concerns. LLMs may inadvertently reproduce sensitive clinical content (e.g., patient-specific details), and excessive memorization may reduce model generalizability, increasing risks of misdiagnosis and making unwarranted recommendations. These risks are further amplified by the generative nature of LLMs, which can not only surface memorized content but also produce overconfident, misleading outputs that may hinder clinical adoption. In this work, we present a study on memorization of LLMs in medicine, assessing its prevalence (how frequently it occurs), characteristics (what is memorized), volume (how much content is memorized), and potential downstream impacts (how memorization may affect medical applications). We systematically analyze common adaptation scenarios: (1) continued pretraining on medical corpora, (2) fine-tuning on standard medical benchmarks, and (3) fine-tuning on real-world clinical data, including over 13,000 unique inpatient records from Yale New Haven Health System. The results demonstrate that memorization is prevalent across all adaptation scenarios and significantly higher than that reported in the general domain. Moreover, memorization has distinct characteristics during continued pre-training and fine-tuning, and it is persistent: up to 87% of content memorized during continued pre-training remains after fine-tuning on new medical tasks.

Understanding fine-grained links between documents is crucial for many applications, yet progress is limited by the lack of efficient methods for data curation. To address this limitation, we introduce a domain-agnostic framework for bootstrapping sentence-level cross-document links from scratch. Our approach (1) generates and validates semi-synthetic datasets of linked documents, (2) uses these datasets to benchmark and shortlist the best-performing linking approaches, and (3) applies the shortlisted methods in large-scale human-in-the-loop annotation of natural text pairs. We apply the framework in two distinct domains -- peer review and news -- and show that combining retrieval models with LLMs achieves a 73% human approval rate for suggested links, more than doubling the acceptance of strong retrievers alone. Our framework allows users to produce novel datasets that enable systematic study of cross-document understanding, supporting downstream tasks such as media framing analysis and peer review assessment. All code, data, and annotation protocols are released to facilitate future research.

Privacy regulations require the erasure of data from deep learning models. This is a significant challenge that is amplified in Federated Learning, where data remains on clients, making full retraining or coordinated updates often infeasible. This work introduces an efficient Federated Unlearning framework based on information theory, modeling leakage as a parameter estimation problem. Our method uses second-order Hessian information to identify and selectively reset only the parameters most sensitive to the data being forgotten, followed by minimal federated retraining. This model-agnostic approach supports categorical and client unlearning without requiring server access to raw client data after initial information aggregation. Evaluations on benchmark datasets demonstrate strong privacy (MIA success near random, categorical knowledge erased) and high performance (Normalized Accuracy against re-trained benchmarks of $\approx$ 0.9), while aiming for increased efficiency over complete retraining. Furthermore, in a targeted backdoor attack scenario, our framework effectively neutralizes the malicious trigger, restoring model integrity. This offers a practical solution for data forgetting in FL.

Building on a human-led thematic analysis of life-story interviews with inpatients with Borderline Personality Disorder, this study examines the capacity of large language models (OpenAI's GPT, Google's Gemini, and Anthropic's Claude) to support qualitative clinical analysis. The models were evaluated through a mixed procedure. Study A involved blinded and non-blinded expert judges in phenomenology and clinical psychology. Assessments included semantic congruence, Jaccard coefficients for overlap of outputs, multidimensional validity ratings of credibility, coherence, and the substantiveness of results, and their grounding in qualitative data. In Study B, neural methods were used to embed the theme descriptions created by humans and the models in a two-dimensional vector space to provide a computational measure of the difference between human and model semantics and linguistic style. In Study C, complementary non-expert evaluations were conducted to examine the influence of thematic verbosity on the perception of human authorship and content validity. Results of all three studies revealed variable overlap with the human analysis, with models being partly indistinguishable from, and also identifying themes originally omitted by, human researchers. The findings highlight both the variability and potential of AI-augmented thematic qualitative analysis to mitigate human interpretative bias and enhance sensitivity.

As AI advances toward general intelligence, the focus is shifting from systems optimized for static tasks to creating open-ended agents that learn continuously. In this paper, we introduce Experience-driven Lifelong Learning (ELL), a framework for building self-evolving agents capable of continuous growth through real-world interaction. The framework is built on four core principles: (1) Experience Exploration: Agents learn through continuous, self-motivated interaction with dynamic environments, navigating interdependent tasks and generating rich experiential trajectories. (2) Long-term Memory: Agents preserve and structure historical knowledge, including personal experiences, domain expertise, and commonsense reasoning, into a persistent memory system. (3) Skill Learning: Agents autonomously improve by abstracting recurring patterns from experience into reusable skills, which are actively refined and validated for application in new tasks. (4) Knowledge Internalization: Agents internalize explicit and discrete experiences into implicit and intuitive capabilities as "second nature". We also introduce StuLife, a benchmark dataset for ELL that simulates a student's holistic college journey, from enrollment to academic and personal development, across three core phases and ten detailed sub-scenarios. StuLife is designed around three key paradigm

As Vision Language Models (VLMs) become integral to real-world applications, understanding their demographic biases is critical. We introduce GRAS, a benchmark for uncovering demographic biases in VLMs across gender, race, age, and skin tone, offering the most diverse coverage to date. We further propose the GRAS Bias Score, an interpretable metric for quantifying bias. We benchmark five state-of-the-art VLMs and reveal concerning bias levels, with the least biased model attaining a GRAS Bias Score of only 2 out of 100. Our findings also reveal a methodological insight: evaluating bias in VLMs with visual question answering (VQA) requires considering multiple formulations of a question. Our code, data, and evaluation results are publicly available.

Diffusion models have achieved remarkable success in content generation but often incur prohibitive computational costs due to iterative sampling. Recent feature caching methods accelerate inference via temporal extrapolation, yet can suffer quality degradation from inaccurate modeling of the complex dynamics of feature evolution. We propose HiCache (Hermite Polynomial-based Feature Cache), a training-free acceleration framework that improves feature prediction by aligning mathematical tools with empirical properties. Our key insight is that feature-derivative approximations in diffusion Transformers exhibit multivariate Gaussian characteristics, motivating the use of Hermite polynomials as a potentially optimal basis for Gaussian-correlated processes. We further introduce a dual-scaling mechanism that ensures numerical stability while preserving predictive accuracy, and is also effective when applied standalone or integrated with TaylorSeer. Extensive experiments demonstrate HiCache's superiority, achieving 5.55x speedup on FLUX.1-dev while matching or exceeding baseline quality, and maintaining strong performance across text-to-image, video generation, and super-resolution tasks. Moreover, HiCache can be naturally added to previous caching methods to enhance their performance, e.g., improving ClusCa from 0.9480 to 0.9840 in terms of image rewards. Code: https://github.com/fenglang918/HiCache

Noninvasive arterial blood pressure (ABP) monitoring is essential for patient management in critical care and perioperative settings, providing continuous assessment of cardiovascular hemodynamics with minimal risks. Numerous deep learning models have developed to reconstruct ABP waveform from noninvasively acquired physiological signals such as electrocardiogram and photoplethysmogram. However, limited research has addressed the issue of model performance and computational load for deployment on embedded systems. The study introduces a lightweight sInvResUNet, along with a collaborative learning scheme named KDCL_sInvResUNet. With only 0.89 million parameters and a computational load of 0.02 GFLOPS, real-time ABP estimation was successfully achieved on embedded devices with an inference time of just 8.49 milliseconds for a 10-second output. We performed subject-independent validation in a large-scale and heterogeneous perioperative dataset containing 1,257,141 data segments from 2,154 patients, with a wide BP range (41-257 mmHg for SBP, and 31-234 mmHg for DBP). The proposed KDCL_sInvResUNet achieved lightly better performance compared to large models, with a mean absolute error of 10.06 mmHg and mean Pearson correlation of 0.88 in tracking ABP changes. Despite these promising results, all deep learning models showed significant performance variations across different demographic and cardiovascular conditions, highlighting their limited ability to generalize across such a broad and diverse population. This study lays a foundation work for real-time, unobtrusive ABP monitoring in real-world perioperative settings, providing baseline for future advancements in this area.

We offer a theoretical mathematical background through Lagrangian optimization on the unit hyperspherical manifold and its tangential collection with application to the Transformer and its token space. Our methods are catered to the attention mechanism in a theoretical setting, but largely appeal to a broader mathematical lens as well. The Transformer, as a flow map, exists in the tangent fiber for each token along the high-dimensional unit sphere. The circumstance of the hypersphere across the latent data is reasonable due to the trained diagonal matrix equal to the identity, which has various empirical justifications. Thus, under the continuum limit of the dynamics, the latent vectors flow among the tangent bundle. Using these facts, we devise a mathematical framework focusing on the attention mechanism through calculus of variations. We develop a functional and show that the continuous flow map induced by the Transformer satisfies this functional, therefore attention can be viewed as a natural solver of a calculus of variations problem. We invent new scenarios of when our methods are applicable based on loss optimization with respect to path optimality. We derive the projected Euler-Lagrange equation under the specific flow map. The variant of the Euler-Lagrange equation we present has various appearances in literature, but, to our understanding, oftentimes not foundationally proven or under other specialized cases. Our overarching proof is new: our techniques are classical and the use of the flow map object is original. We provide several other relevant results, primarily ones specific to neural scenarios. In particular, much of our analysis will be attempting to quantify Transformer data in variational contexts under neural approximations.

Presented is a novel methodology for determining representational structure, which builds upon the existing Spotlight Resonance method. This new tool is used to gain insight into how discrete representations can emerge and organise in autoencoder models, through a controlled ablation study that alters only the activation function. Using this technique, the validity of whether function-driven symmetries can act as implicit inductive biases on representations is determined. Representations are found to tend to discretise when the activation functions are defined through a discrete algebraic permutation-equivariant symmetry. In contrast, they remain continuous under a continuous algebraic orthogonal-equivariant definition. This confirms the hypothesis that the symmetries of network primitives can carry unintended inductive biases, leading to task-independent artefactual structures in representations. The discrete symmetry of contemporary forms is shown to be a strong predictor for the production of symmetry-organised discrete representations emerging from otherwise continuous distributions -- a quantisation effect. This motivates further reassessment of functional forms in common usage due to such unintended consequences. Moreover, this supports a general causal model for a mode in which discrete representations may form, and could constitute a prerequisite for downstream interpretability phenomena, including grandmother neurons, discrete coding schemes, general linear features and a type of Superposition. Hence, this tool and proposed mechanism for the influence of functional form on representations may provide insights into interpretability research. Finally, preliminary results indicate that quantisation of representations correlates with a measurable increase in reconstruction error, reinforcing previous conjectures that this collapse can be detrimental.

Multi-sensor fusion plays a critical role in enhancing perception for autonomous driving, overcoming individual sensor limitations, and enabling comprehensive environmental understanding. This paper first formalizes multi-sensor fusion strategies into data-level, feature-level, and decision-level categories and then provides a systematic review of deep learning-based methods corresponding to each strategy. We present key multi-modal datasets and discuss their applicability in addressing real-world challenges, particularly in adverse weather conditions and complex urban environments. Additionally, we explore emerging trends, including the integration of Vision-Language Models (VLMs), Large Language Models (LLMs), and the role of sensor fusion in end-to-end autonomous driving, highlighting its potential to enhance system adaptability and robustness. Our work offers valuable insights into current methods and future directions for multi-sensor fusion in autonomous driving.

Due to the surging amount of AI-generated images, its provisioning to edges and mobile users from the cloud incurs substantial traffic on networks. Generative semantic communication (GSC) offers a promising solution by transmitting highly compact information, i.e., prompt text and latent representations, instead of high-dimensional image data. However, GSC relies on the alignment between the knowledge in the cloud generative AI (GAI) and that possessed by the edges and users, and between the knowledge for wireless transmission and that of actual channels, which remains challenging. In this paper, we propose DeKA-g, a distillation-enabled knowledge alignment algorithm for GSC systems. The core idea is to distill the image generation knowledge from the cloud-GAI into low-rank matrices, which can be incorporated by the edge and used to adapt the transmission knowledge to diverse wireless channel conditions. DeKA-g comprises two novel methods: metaword-aided knowledge distillation (MAKD) and condition-aware low-rank adaptation (CALA). For MAKD, an optimized metaword is employed to enhance the efficiency of knowledge distillation, while CALA enables efficient adaptation to diverse rate requirements and channel conditions. From simulation results, DeKA-g improves the consistency between the edge-generated images and the cloud-generated ones by 44% and enahnces the average transmission quality in terms of PSNR by 6.5 dB over the baselines without knowledge alignment.

Ensuring the trustworthiness of graph neural networks (GNNs), which are often treated as black-box models, requires effective explanation techniques. Existing GNN explanations typically apply input perturbations to identify subgraphs that are responsible for the occurrence of the final output of GNNs. However, such approaches lack finer-grained, layer-wise analysis of how intermediate representations contribute to the final result, capabilities that are crucial for model diagnosis and architecture optimization. This paper introduces SliceGX, a novel GNN explanation approach that generates explanations at specific GNN layers in a progressive manner. Given a GNN model M, a set of selected intermediate layers, and a target layer, SliceGX slices M into layer blocks("model slice") and discovers high-quality explanatory subgraphs within each block that elucidate how the model output arises at the target layer. Although finding such layer-wise explanations is computationally challenging, we develop efficient algorithms and optimization techniques that incrementally construct and maintain these subgraphs with provable approximation guarantees. Extensive experiments on synthetic and real-world benchmarks demonstrate the effectiveness and efficiency of SliceGX, and illustrate its practical utility in supporting model debugging.