Papers

The ability to detect out-of-distribution (OOD) inputs is fundamental to safe deployment of machine learning systems. Yet, current methods often rely on feature representations that are optimised solely for classification accuracy, neglecting the distinct requirements of epistemic uncertainty. We introduce GOEN (Geometry-Optimised Epistemic Network), a simple pipeline that combines multi-scale features, L2 normalisation, Mahalanobis distance, and a calibration head trained with real hard OOD examples. Through systematic ablation we uncover a counter-intuitive finding: CenterLoss, a popular regulariser for feature compactness, significantly degrades OOD detection performance, reducing average OOD AUROC from 0.9483 to 0.9366 despite improving classification accuracy. The best variant, GOEN-NoCenterLoss, achieves an average OOD AUROC of 0.9483, surpassing all baselines including deep ensembles (0.8827), KNN (0.8967), and ODIN (0.8870) on CIFAR-10 benchmarks, while maintaining competitive in-distribution accuracy. Our results challenge the prevailing assumption that better classification geometry automatically leads to better epistemic uncertainty. Instead, we show that overly tight feature clusters compress inter-class margins and distort the covariance structure needed for effective OOD detection. GOEN is efficient, training in under 20 minutes on a single GPU, and provides a practical blueprint for building AI systems that reliably recognise their own limitations.

Large language models (LLMs) are increasingly deployed in quantitative finance for stock price forecasting. This review synthesizes recent applications of LLMs in this domain, including extracting sentiment from financial news and social media, analyzing financial reports and earnings-call transcripts, tokenizing or symbolizing stock price series, and constructing multi-agent trading systems. Particular attention is paid to practical pitfalls that are often understated in the literature, such as fragility in sentiment analysis, dataset and horizon design, performance evaluation metrics, data leakage, illiquidity premia, and limits of stock price predictability. Organized from a hedge-fund perspective, the review is intended to guide both academic researchers and hedge fund managers in integrating LLMs into real-world trading pipelines and in stress-testing their robustness under realistic market frictions.

Reasoning capabilities are crucial for reliable medical visual question answering (VQA); however, existing datasets rarely include reasoning explanations. We address this by generating reasoning trajectories for six medical VQA benchmarks using the COMCTS algorithm with open-source vision-language models, with an LLM serving as the verification judge. Building on these generated datasets, we propose a two-stage training framework: supervised fine-tuning followed by Group Relative Policy Optimization (GRPO) with a novel process-based reward. While standard approaches rely solely on exact-match rewards for final answers, we introduce a trajectory-aware reward that measures the similarity between generated and ground-truth reasoning processes. Specifically, we embed reasoning steps using sentence transformers and compute the Dynamic Time Warping (DTW) distance between the resulting vector sequences. Experiments across six benchmarks demonstrate that combining the DTW-based process reward with exact-match reward consistently outperforms SFT-only training, raising mean accuracy from 0.598 to 0.689, mean BERTScore from 0.845 to 0.881, and mean ROUGE-L from 0.665 to 0.748. Our results highlight the importance of process supervision in training reasoning-capable medical VLMs. We make our code and generated reasoning datasets publicly available at https://anonymous.4open.science/r/MICCAI-R1-MED-VQA-code-B14B/

Alzheimer's Dementia (AD) is a progressive neurodegenerative disease marked by irreversible decline, making reliable modeling of its progression essential for effective patient care. Progression-aware methods such as survival analysis are therefore crucial tools for the early detection and monitoring of AD. Recent advancements in deep learning have demonstrated remarkable performance in survival tasks, but alarmingly fewer studies have been conducted in the domain of AD. Further, the studies that do exist do not consider learned bias within the model itself, which could result in unfair and unreliable predictions toward certain marginalized groups. As such, we conduct a rigorous study of fairness in AD progression analysis along with a thorough feature importance study to determine the characteristics which are most important for reliable AD predictions. Furthermore, we propose two novel fairness metrics, called Time-Dependent Concordance Impurity and Kaplan-Meier Fairness, to quantify bias with respect to sensitive attributes such as sex, race, and education in nonparametric survival models. Our study demonstrates that while deep learning powered survival models are robust tools which can aid clinicians in AD care decisions, they often exhibit considerable bias, representing important avenues for future research.

Quantization has emerged as a mainstream approach for deploying Large Language Models (LLMs) on resource-constrained devices, yet compressing precision below 4-bit typically causes severe performance degradation or prohibitive retraining costs. In this paper, we propose EdgeRazor, a lightweight framework for LLMs via Mixed-Precision Quantization-Aware Distillation. It contains three modules: Structural Quantization with Mixed Precision for fine-grained control of bit-widths, Layer-Adaptive Feature Distillation that dynamically selects the most informative features for alignment, and Entropy-Aware KL Divergence for forward-reverse balance on both human-annotated and distilled datasets. Evaluations conducted on MobileLLM and Qwen families show that under weight-activation quantization, the 1.88-bit Qwen3-0.6B-EdgeRazor outperforms the state-of-the-art 2-bit baselines by 11.27 and surpasses the strongest 3-bit baselines by 4.38, while the quantized MobileLLM-350M-EdgeRazor requires a training budget 4-10$\times$ lower than the leading quantization-aware training method. In terms of efficiency, EdgeRazor achieves higher compression ratios at all bit-widths, and the 1.58-bit Qwen3-0.6B-EdgeRazor reduces storage from 1.11 GB to 0.19 GB while accelerating decoding by 15.16$\times$ over the 16-bit baseline. These results empirically validate the effectiveness and efficiency of EdgeRazor. The codes can be accessed from \href{https://github.com/zhangsq-nju/EdgeRazor}{GitHub} and \href{https://huggingface.co/collections/zhangsq-nju/edgerazor-nbit}{Huggingface}.

Understanding how large language models encode task identity from few-shot demonstrations is a central open problem in mechanistic interpretability. Prior work uses linear probing to localize task representations, reporting high classification accuracy at specific layers. We reveal a striking dissociation: probing accuracy completely fails to predict causal importance. Single-position activation intervention achieves 0% task transfer across all 28 layers of Llama-3.2-3B-despite 100% probing accuracy at those same positions. This null result is itself a key finding, demonstrating that task encoding is fundamentally distributed. Multi-position intervention-replacing activations at all demonstration output tokens simultaneously-achieves up to 96% transfer (N=50, 95% CI: [87%, 99%]) at layer 8, pinpointing for the first time the causal locus of ICL task identity. We establish the generality of these findings across four models spanning three architecture families (LLaMA, Qwen, Gemma), discovering a universal intervention window at ~30% network depth. Causal tracing uncovers an asymmetric architecture: the query position is strictly necessary (53-100% disruption) while no individual demonstration position is necessary (0% disruption)-resolving a key ambiguity in prior accounts. Crucially, transfer depends on internal representation compatibility, not surface similarity (r=-0.05 vs r=0.31), ruling out trivial explanations. These results establish the distributed template hypothesis: ICL task identity is encoded as output format templates distributed across demonstration tokens, fundamentally reshaping our understanding of how in-context learning operates.

Recently, a new paradigm named drifting model has been proposed for mapping distributions, which achieves the SOTA image generation performance over ImageNet via one-step neural functional evaluation (NFE). The basic idea is to compute a drifting term at each training iteration and then push the output of the model towards the direction of the drifting term. In this paper, we propose a lookahead drifting model. At each training iteration, we compute a set of drifting terms sequentially. Each drifting term is calculated by making use of previously computed ones as well as the positive samples and the output of the model. %One key step is to properly scale the drifting terms so that their magnitudes are in a comparable range. In principle, the drifting terms obtained at a later stage capture higher order gradient information towards the positive samples. At each training iteration, the model is optimized by pushing its output towards the direction of the (weighted) summation of the drifting terms. Experimental results on toy examples and CIFAR10 demonstrate the better performance of the new method than the baseline.

Deep learning models continue to scale, with some requiring more storage than many large-scale datasets. Thus, we introduce a new paradigm: Continual Distillation (CD), where a student learns sequentially from a stream of teacher models without retaining access to earlier teachers. CD faces two challenges: teacher training data is unavailable, and teachers have varying expertise. We show that external unlabeled data enables Unseen Knowledge Transfer (UKT), allowing the student to acquire information from domains not present in the training data, while known to the teacher. We also show that sequential distillation causes Unseen Knowledge Forgetting (UKF) when transferred knowledge is lost after training on later teachers. To better trade off between UKT and UKF, we propose Self External Data Distillation (SE2D), a method that preserves logits on external data to stabilize learning across heterogeneous teachers. Experiments on multiple benchmarks show that SE2D reduces UKF and improves cross-domain generalization. The code and implementation for this work are publicly available at: https://github.com/Nicolas1203/continual_distillation.

Parameter-efficient transfer learning (PETL) has emerged as a pivotal paradigm for adapting pre-trained foundation models to downstream tasks, significantly reducing trainable parameters yet suffering from substantial memory overhead caused by gradient backpropagation during fine-tuning. While memory-efficient transfer learning (METL) circumvents this challenge by bypassing backbone gradient computation via lightweight small side networks, its stringent memory constraint severely limits learning capacity of side networks, thereby significantly compromising performance. To address these limitations, we propose a novel Mixed-Precision Interactive Side Mixture-of-Experts framework (MP-ISMoE). Specifically, we first propose a Gaussian Noise Perturbed Iterative Quantization (GNP-IQ) scheme to quantize weights into lower-bits while effectively decreasing quantization errors. By leveraging memory conserved from GNP-IQ, we subsequently employ Interactive Side Mixture-of-Experts (ISMoE) to scaling up side networks without sacrificing overall memory efficiency. Different from conventional mixture-of-experts, ISMoE learns to select optimal experts by interacting with salient features from frozen backbones, thus suppressing knowledge forgetting and boosting performance. Extensive experiments across diverse vision-language and language-only tasks demonstrate that MP-ISMoE remarkably promotes accuracy compared to state-of-the-art METL approaches, while maintaining comparable parameter and memory efficiency.

This paper focuses on a key challenge in Neural Architecture Search (NAS): integrating established architectural knowledge while exploring new designs under expensive evaluations. Large language models (LLMs) are a promising assistant for NAS because they can translate rich architectural and coding priors into executable code edits. However, in practice, seemingly local revisions often propagate into non-local behavioral and performance shifts because a single edit can inadvertently couple multiple interacting functional factors, a phenomenon we refer to as functional entanglement. To make LLM knowledge usable under such entanglement, we propose Structured Progressive Knowledge Activation (SPARK), which activates relevant priors by explicitly selecting the functional factor to modify and conditioning the edit on that factor. This factor-conditioned editing reduces entangled side effects and yields more targeted, reliable architecture modifications. On CLRS-DFS, SPARK achieves a 28.1x sample-efficient architecture evolution speedup and yields a 22.9% relative improvement in OOD accuracy. Our code is available at https://github.com/AIM-ResearchLab/SPARK.

Representation learning seeks meaningful sensory representations without supervision and can model aspects of human development. Although many neural networks empirically learn useful features, a principled account of what makes a representation "good" remains elusive. We study unsupervised categorization of transformations between pairs of inputs under algebraic constraints. Classical disentanglement favors mutually independent factors and fails when factors are coupled. Our prior Galois-theoretic approach decomposes a group via normal subgroups by learning a product of two transformations with one factor constrained to a normal subgroup, covering both commutative and non-commutative cases. That method, however, relied on auxiliary assumptions (e.g., motion and isometry restrictions) not required by decomposition theory, and ablations did not separate theory-based from auxiliary effects. We propose parameter division for a single transformation: we split its parameter into components, impose homomorphism constraints mapping the full transformation to one component, and identify the normal subgroup as the set of transformations when that component is fixed to the identity. This formulation drops the previous auxiliary assumptions and applies more broadly. We evaluate on image pairs involving rotation, translation, and scale; ablations show that group-decomposition constraints drive appropriate categorization.

Adaptive optimizers like AdamW apply uniform hyperparameters across all parameter groups, ignoring heterogeneous optimization dynamics across layers and modules. We address this limitation by proposing MetaAdamW - a new optimizer that integrates a self-attention mechanism to dynamically modulate per-group learning rates and weight decay. The modulation factors are produced by a lightweight Transformer encoder that operates on statistical features (gradient norms, momentum norms, correlations) extracted from each parameter group. To train the attention module, we introduce a meta-learning objective that combines gradient alignment, loss decrease, and generalization gap. A key novel contribution is the extension of homoscedastic uncertainty weighting (HUW) with task-specific priorities that directly scale the regularization terms - enabling domain knowledge to guide automatic loss balancing. Extensive experiments on five diverse tasks-time series forecasting (ETT), language modeling (WikiText-2), machine translation (Multi30k), image classification (CIFAR-10), and sentiment analysis (IMDB) - demonstrate that MetaAdamW consistently outperforms the standard AdamW baseline in terms of validation loss, accuracy, or perplexity. Depending on the task, MetaAdamW either reduces overall training time (by up to 17.11%) or improves performance (by up to 11.08%) while introducing only moderate overhead; in some cases, it can also mitigate issues of insufficient convergence caused by premature early stopping. Ablation studies validate the effectiveness of each component, including feature versions, grouping strategies, and the proposed priority-injected uncertainty weighting.

Achieving endogenous regime switching is crucial for the emergence of autonomous intelligence, yet remains a central challenge for existing machine learning frameworks, where such transitions are typically externally imposed. In this work, we introduce a classification that distinguishes scalar-reducible dynamics, which can be expressed as gradient flows driven by a scalar objective, from scalar-irreducible dynamics that cannot be reduced to such a form. While most existing machine learning systems operate within the scalar-reducible class, we demonstrate that scalar-irreducible dynamics naturally enable internally generated regime switching through feedback between fast dynamical variables and slow structural adaptation. Using a minimal dynamical model, we illustrate how this mechanism produces sustained endogenous regime transitions without external scheduling. Our results suggest a new dynamical paradigm for regime exploration and provide a potential route toward autonomous learning systems whose adaptive behavior is organized internally rather than externally prescribed.

This paper presents a detailed study of how graph neural networks can be used on edge intelligent meters in a microgrid to forecast photovoltaic power generation. The problem background and the adopted technologies are introduced, including ONNX and ONNX Runtime. The hardware and software specifications of the smart meter are also briefly described. Then, the paper focuses on the training and deployment of two graph machine learning models, GCN and GraphSAGE, with particular emphasis on developing and deploying a customized ONNX operator for GCN. Finally, a case study is conducted using real datasets from a village microgrid. The performance of the two models is compared on both the PC and the smart meter, exhibiting successful deployments and executions on the smart meter.

Recent work on KV cache quantization, culminating in TurboQuant, has approached the Shannon entropy limit for per-vector compression of transformer key-value caches. We observe that this limit applies to a strictly weaker problem than the one that actually matters: compressing the KV cache as a sequence. The tokens stored in a KV cache are not arbitrary floating-point data – they are samples from the exact formal language the model was trained on, and the model is by construction a near-optimal predictor of that language. We introduce sequential KV compression, a two-layer architecture that exploits this structure. The first layer, probabilistic prefix deduplication, identifies semantically equivalent shared prefixes across sessions using the trie metric d_T(s, s') = -log_2 P_M(s ^ s') from Probabilistic Language Tries (PLTs). The second layer, predictive delta coding, stores only the residual of each new KV vector from the model's own prediction of it, achieving a per-token entropy bound of H(KV_{i+1} | KV_{<=i}) <= H(token_{i+1} | token_{<=i}). We prove that at typical language model perplexity – approximately 10-20 for fluent English text – this bound is 3.3-4.3 bits on average per token position, compared to TurboQuant's 3 bits per vector component (with typical attention heads having 64-128 components). The theoretical compression ratio over TurboQuant is approximately 914,000x at the Shannon limit. Even at 1000x above the entropy floor – a deliberately pessimistic worst-case overhead, two orders of magnitude above the 2-5x typical of practical source coders – the ratio remains approximately 914x over TurboQuant, with compression improving rather than degrading as context length grows. The two layers are orthogonal and compose with existing per-vector quantization methods including TurboQuant.

Predicting post-surgical seizure outcomes in pharmacoresistant epilepsy is a clinical challenge. Conventional deep-learning approaches operate on static, single-timepoint pre-operative scans, omitting longitudinal morphological changes. We propose Neuro-Oracle, a three-stage framework that: (i) distils pre-to-post-operative MRI changes into a compact 512-dimensional trajectory vector using a 3D Siamese contrastive encoder; (ii) retrieves historically similar surgical trajectories from a population archive via nearest-neighbour search; and (iii) synthesises a natural-language prognosis grounded in the retrieved evidence using a quantized Llama-3-8B reasoning agent. Evaluations are conducted on the public EPISURG dataset ($N{=}268$ longitudinally paired cases) using five-fold stratified cross-validation. Since ground-truth seizure-freedom scores are unavailable, we utilize a clinical proxy label based on the resection type. We acknowledge that the network representations may potentially learn the anatomical features of the resection cavities (i.e., temporal versus non-temporal locations) rather than true prognostic morphometry. Our current evaluation thus serves mainly as a proof-of-concept for the trajectory-aware retrieval architecture. Trajectory-based classifiers achieve AUC values between 0.834 and 0.905, compared with 0.793 for a single-timepoint ResNet-50 baseline. The Neuro-Oracle agent (M5) matches the AUC of purely discriminative trajectory classifiers (0.867) while producing structured justifications with zero observed hallucinations under our audit protocol. A Siamese Diversity Ensemble (M6) of trajectory-space classifiers attains an AUC of 0.905 without language-model overhead.

As coding agents have seen rapid capability and adoption gains, users are applying them to general tasks beyond software engineering. In this post, we investigate whether coding agents can successfully generalize to end-to-end business process automation. We identify gaps in current evaluations, and conduct a case study to evaluate a coding agent on practical business tasks in an open-core Enterprise Resource Planning system. We find that the agent reliably completes simple tasks but exhibits characteristic failures on complex tasks, suggesting that bridging domain logic and code execution is a key bottleneck to generalizability.

Large language model agents rely on a memory harness to write, organize, retrieve, and use past experience. A harness that works for one task often fails on another, because conversations, embodied planning, and expert reasoning require different storage and retrieval behavior. To address this limitation, we introduce Mstar, a method that automatically discovers task-optimized memory programs through reflective code evolution. Specifically, Mstar operationalizes a memory harness as an executable Python memory program that defines the Schema, Logic, and Instruction, and optimizes these components jointly. We evaluate Mstar on four tasks covering conversation, embodied planning, and specialized reasoning. Our results demonstrate that Mstar improves performance over existing baselines across all evaluated tasks. Furthermore, the evolved memory programs exhibit structurally distinct processing mechanisms for each evaluated domain. These findings suggest that every task benefits from its own memory harness, and that memory-program search provides a concrete way to discover it.

Many applications today use large language models for code generation; however, production systems have strict latency requirements that can be difficult to meet with large models. Small language models with a few billion parameters are resource efficient but may suffer from limited reasoning, hallucinations, or poor retention of longer context. Fine tuning improves task specific accuracy by embedding domain knowledge directly into model weights, reducing reliance on runtime context. We previously implemented a baseline natural language to code generation approach using a retrieval augmented generation pipeline that dynamically selected few shot examples to embed domain specific language context for a large language model. In this study, we evaluate small language models for generating domain specific language from natural language by fine tuning variants of Mistral and other models on a dataset of natural language code pairs. Our results show that the fine-tuned models achieve improved performance and latency on test datasets compared to larger models. We also demonstrate that the trained model can be further fine-tuned for customer specific scenarios without degrading general performance, helping resolve production issues. Load testing followed by production deployment confirmed optimal performance in terms of latency and quality. These findings demonstrate that task specific fine tuning with small language models provides an efficient, faster, and cost-effective alternative to large language models for domain specific language generation.

Reservoir computing (RC) is a computational framework known for its training efficiency, making it ideal for physical hardware implementations. However, realizing the complex interconnectivity of traditional reservoirs in physical systems remains a significant challenge. This paper proposes a physical RC scheme inspired by the biological vestibular system. To overcome hardware complexity, we introduce a designed uncoupled topology and demonstrate that it achieves performance comparable to fully coupled networks. We theoretically analyze the difference between these topologies by deriving a memory capacity formula for linear reservoirs, identifying specific conditions where both configurations yield equivalent memory. These analytical results are demonstrated to approximately hold for nonlinear reservoir systems. Furthermore, we systematically examine the impact of reservoir size on predictive statistics and memory capacity. Our findings suggest that uncoupled reservoir architectures offer a mathematically sound and practically feasible pathway for efficient physical reservoir computing.

Fine-tuning Large Language Models (LLMs) typically involves either full fine-tuning, which updates all model parameters, or Parameter-Efficient Fine-Tuning (PEFT), which adjusts a small subset of parameters. However, both approaches have inherent limitations: full fine-tuning is computationally expensive, while PEFT often struggles to learn new knowledge and exhibits suboptimal performance. To overcome these issues, we propose a novel hybrid fine-tuning approach that jointly updates both LLMs and PEFT modules using a combination of zeroth-order and first-order optimization methods. To analyze our new algorithm, we develop a theoretical framework centered on the concept of hybrid smoothness condition, which accounts for the heterogeneous nature of the optimization landscape in joint LLM and PEFT training. We derive a rigorous convergence analysis for the convergence of reshuffling-type SGD algorithm under multiple learning rates and demonstrate its effectiveness through extensive empirical studies across various downstream tasks and model architectures. On the practical side, our results demonstrate consistent performance improvement, making the approach a viable solution for large-scale language model fine-tuning.

Conventional field operations spend most of their energy moving the tractor body, not the implement. Yet feasibility studies for novel agricultural vehicles rarely tie mechanics, energy harvest, draft, field geometry, economics, life-cycle CO2, and uncertainty quantification together on a single reproducible code path. This paper builds such a framework and applies it to CableTract, a two-module cable-driven field robot. A stationary Main Unit (winch + motor + battery + harvester module) (MU) and a lighter Anchor module (held by helical screw piles) tension a cable across a strip while a lightweight implement carriage rolls along it. The heavy bodies stay on the headland; only the carriage enters the field. The carriage runs a 10-implement library co-designed for the cable architecture. This co-design is the paper's central analytical lever. The framework is prototype-free. It chains a catenary cable model, a drivetrain efficiency chain, a stochastic draft model fitted to the co-designed library, an hourly solar + wind + battery simulator on six sites, a polygon coverage planner on a 50-field corpus, a contact-pressure compaction model, a discounted cash-flow economics engine with battery replacement and life-cycle CO2, and a global sensitivity analysis on 20 inputs. An operating-envelope sweep and an architectural-variant comparison close the loop. The full implementation is open source. Applied to the codesigned reference, the framework yields energy, compaction advantages and potential off-grid operation.

Hybrid approaches that combine data-driven learning with physics-based insight have shown promise for improving the reliability of industrial condition monitoring. This work develops a hybrid condition monitoring framework that integrates primary sensor measurements, lagged temporal features, and physics-informed residuals derived from nominal surrogate models. Two hybrid integration strategies are examined. The first is a feature-level fusion approach that augments the input space with residual and temporal information. The second is a model-level ensemble approach in which machine learning classifiers trained on different feature types are combined at the decision level. Both hybrid approaches of the condition monitoring framework are evaluated on a continuous stirred-tank reactor (CSTR) benchmark using several machine learning models and ensemble configurations. Both feature-level and model-level hybridization improve diagnostic accuracy relative to single-source baselines, with the best model-level ensemble achieving a 2.9% improvement over the best baseline ensemble. To assess predictive reliability, conformal prediction is applied to quantify coverage, prediction-set size, and abstention behavior. The results show that hybrid integration enhances uncertainty management, producing smaller and well-calibrated prediction sets at matched coverage levels. These findings demonstrate that lightweight physics-informed residuals, temporal augmentation, and ensemble learning can be combined effectively to improve both accuracy and decision reliability in nonlinear industrial systems.

Subsurface stratigraphy contains important spatio-temporal information about accumulation, deformation, and layer formation in polar ice sheets. In particular, variations in internal ice layer thickness provide valuable constraints for snow mass balance estimation and projections of ice sheet change. Although radar sensors can capture these layered structures as depth-resolved radargrams, convolutional neural networks applied directly to radar images are often sensitive to speckle noise and acquisition artifacts. In addition, purely data-driven methods may underuse physical knowledge, leading to unrealistic thickness estimates under spatial or temporal extrapolation. To address these challenges, we develop K-STEMIT, a novel knowledge-informed, efficient, multi-branch spatio-temporal graph neural network that combines a geometric framework for spatial learning with temporal convolution to capture temporal dynamics, and incorporates physical data synchronized from the Model Atmospheric Regional physical weather model. An adaptive feature fusion strategy is employed to dynamically combine features learned from different branches. Extensive experiments have been conducted to compare K-STEMIT against current state-of-the-art methods in both knowledge-informed and non-knowledge-informed settings, as well as other existing methods. Results show that K-STEMIT consistently achieves the highest accuracy while maintaining near-optimal efficiency. Most notably, incorporating adaptive feature fusion and physical priors reduces the root mean-squared error by 21.01% with negligible additional cost compared to its conventional multi-branch variants. Additionally, our proposed K-STEMIT achieves consistently lower per-year relative MAE, enabling reliable, continuous spatiotemporal assessment of snow accumulation variability across large spatial regions.

Much work has been done on designing fast and accurate sampling for diffusion language models (dLLMs). However, these efforts have largely focused on the tradeoff between speed and quality of individual samples; how to additionally ensure diversity across samples remains less well understood. In this work, we show that diversity can be increased by using softened, tempered versions of familiar confidence-based remasking heuristics, retaining their computational benefits and offering simple implementations. We motivate this approach by introducing an idealized formal model of fork tokens and studying the impact of remasking on the expected entropy at the forks. Empirically, the proposed tempered heuristics close the exploration gap (pass@k) between existing confidence-based and autoregressive sampling, hence outperforming both when controlling for cost (pass@NFE). We further study how the increase in diversity translates to downstream post-training and test-time compute scaling. Overall, our findings demonstrate that simple, efficient, and diverse sampling from dLLMs is possible.

Simultaneous Speech Translation (SimulST) requires balancing high translation quality with low latency. Recent work introduced REINA, a method that trains a Read/Write policy based on estimating the information gain of reading more audio. However, we find that information-based policies often lack temporal context, leading the policy to bias itself toward reading most of the audio before starting to write. We improve REINA using two distinct strategies: a supervised alignment network (REINA-SAN) and a timestep-augmented network (REINA-TAN). Our results demonstrate that while both methods significantly outperform the baseline and resolve stability issues, REINA-TAN provides a slightly superior Pareto frontier for streaming efficiency, whereas REINA-SAN offers more robustness against 'read loops'. Applied to Whisper, both methods improve the pareto frontier of streaming efficiency as measured by Normalized Streaming Efficiency (NoSE) scores up to 7.1% over existing competitive baselines.

We study last-iterate convergence of SGD with greedy step size over smooth quadratics in the interpolation regime, a setting which captures the classical Randomized Kaczmarz algorithm as well as other popular iterative linear system solvers. For these methods, we show that the $t$-th iterate attains an $O(1/t^{3/4})$ convergence rate, addressing a question posed by Attia, Schliserman, Sherman, and Koren, who gave an $O(1/t^{1/2})$ guarantee for this setting. In the proof, we introduce the family of stochastic contraction processes, whose behavior can be described by the evolution of a certain deterministic eigenvalue equation, which we analyze via a careful discrete-to-continuous reduction.

Emergency department triage relies heavily on both quantitative vital signs and qualitative clinical notes, yet multimodal machine learning models predicting triage acuity often suffer from modality collapse by over-relying on structured tabular data. This limitation severely hinders demographic generalizability, particularly for pediatric patients where developmental variations in vital signs make unstructured clinical narratives uniquely crucial. To address this gap, we propose a late-fusion multimodal architecture that processes tabular vitals via XGBoost and unstructured clinical text via Bio_ClinicalBERT, combined through a Logistic Regression meta-classifier to predict the 5-level Emergency Severity Index. To explicitly target the external validity problem, we train our model exclusively on adult encounters from the MIMIC-IV and NHAMCS datasets and evaluate its zero-shot generalization on a traditionally overlooked pediatric cohort. Furthermore, we employ symmetric modality dropout during training to prevent the ensemble from overfitting to adult-specific clinical correlations. Our results demonstrate that the multimodal framework significantly outperforms single-modality baselines. Most notably, applying a 30-40% symmetric modality dropout rate yielded steep performance improvements in the unseen pediatric cohort, elevating the Quadratic Weighted Kappa to 0.351. These findings highlight modality dropout as a critical regularization technique for mitigating modality collapse and enhancing cross-demographic generalization in clinical AI.

Medical vision-language datasets are often limited in size and biased toward negative findings, as clinicians report abnormalities mostly but might omit some positive/neutral findings because they might be considered as irrelevant to the patient's condition. We propose a self-supervised data enrichment method that leverages semantic clustering of report sentences. Then we enrich the findings in the medical reports in the training set by adding positive/neutral observations from different clusters in a self-supervised manner. Our approach yields consistent gains in supervised fine-tuning (5.63%, 3.04%, 7.40%, 5.30%, 7.47% average gains on COMET score, Bert score, Sentence Bleu, CheXbert-F1 and RadGraph-F1 scores respectively). Ablation studies confirm that improvements stem from semantic clustering rather than random augmentation. Furthermore, we introduce a way to incorporate semantic cluster information into the reward design for GRPO training, which leads to further performance gains (2.78%, 3.14%, 12.80% average gains on COMET score, Bert score and Sentence Bleu scores respectively). We share our code at https://github.com/igulluk/SemEnrich

Accurate volume estimation of objects from visual data is a long-standing challenge in computer vision with significant applications in robotics, logistics, and smart health. Existing methods often rely on complex 3D reconstruction pipelines or struggle with the ambiguity inherent in single-view images. To address these limitations, we introduce a new method that fuses implicit 3D cues from stereo vision with explicit prior knowledge from natural language text. Our approach extracts deep features from a stereo image pair and a descriptive text prompt that contains the object's class and an approximate volume, then integrates them using a simple yet effective projection layer into a unified, multi-modal representation for regression. We conduct extensive experiments on public datasets demonstrating that our text-guided approach significantly outperforms vision-only baselines. Our findings show that leveraging even simple textual priors can effectively guide the volume estimation task, paving the way for more context-aware visual measurement systems. Code: https://gitlab.com/viper-purdue/stereo-typical-estimator.