论文检索

Azka Ikramullah, Abdul Majeed, Kyunghyun Lee, Seong Oun Hwang

Large language models (LLMs) often perform better when prompted to explain their reasoning, but it remains unclear how well such gains persist as reasoning depth increases. In this work, we propose a depth-aware evaluation framework alongside the performance results on two structured datasets: CLUTRR (kinship reasoning) and ProofWriter (logical entailment), comparing direct vs. reasoning (reasoning depth = number of inference steps required) prompts across five models. Reasoning gave small gains at shallow depths but quickly weakened and often reversed as tasks grew more complex. In ProofWriter, GPT-5 reached 90% accuracy at depth four in direct model, yet its reasoning accuracy fell below baseline after depth two. Smaller open-source models showed only unstable or negligible gains, underscoring that reasoning in LLMs remains brittle with increased depth.

Rui Jerry Huang, Anastasia Miin, Wendy Liu

Large language models (LLMs) demonstrate strong reasoning capabilities, yet the inference-time performance of existing solutions remains limited by self-biases, coordination inefficiencies, lack of robust error detection, and dependency on high-quality verifiers. To address these challenges, we propose Adaptive Coopetition (AdCo), a lightweight, multi-agent multi-round inference-time framework that enhances collective reasoning through adaptive decision-making guided by coarse verifier signals. Without relying on high-performance verifiers, AdCo achieves a 20% relative accuracy improvement on math reasoning benchmarks, with consistent performance on different sample sizes and agent configurations. This adaptive, signal-guided ‘coopetition’ framework enhances reasoning robustness by leveraging diverse model knowledge and reasoning traces, while also promoting uncertainty-driven exploration, especially when participants have comparable capabilities.

Aisha Hassan, Tushar Shinde

Efficient spam detection in resource-constrained environments remains challenging due to class imbalance, noisy text, and the computational demands of large Transformer models. We introduce a novel coreset selection framework based on a unified Entropy, Class-Balanced Uncertainty-Density Ranking (CBUDR) scheme. Our method prioritizes highly informative and uncertain samples while ensuring diversity and class balance within the selected subset. The framework flexibly supports multiple selection strategies, including Top-K, Bottom-K, and adaptive class-wise schemes, enabling robust performance even when training on as little as 5% of the dataset. Extensive experiments on benchmark datasets (UCI SMS, UTKML Twitter, LingSpam) show that our ranking scheme achieves competitive accuracy, precision, and recall while significantly reducing computational cost. These results demonstrate that carefully designed coreset strategies can surpass full-data performance in both balanced and imbalanced settings, highlighting the potential for deployment on low-power devices and mobile platforms.

Sunny Gupta, Nikita Jangid, Amit Sethi

Federated Learning (FL) often suffers from severe performance degradation when faced with non-IID data, largely due to local classifier bias. Traditional remedies such as global model regularization or layer freezing either incur high computational costs or struggle to adapt to feature shifts. In this work, we propose UniVarFL, a novel FL framework that emulates IID-like training dynamics directly at the client level, eliminating the need for global model dependency. UniVarFL leverages two complementary regularization strategies during local training: Classifier Variance Regularization, which aligns class-wise probability distributions with those expected under IID conditions, effectively mitigating local classifier bias; and Hyperspherical Uniformity Regularization, which encourages a uniform distribution of feature representations across the hypersphere, thereby enhancing the model’s ability to generalize under diverse data distributions. Extensive experiments on multiple benchmark datasets demonstrate that UniVarFL outperforms existing methods in accuracy, highlighting its potential as a highly scalable and efficient solution for real-world FL deployments, especially in resource-constrained settings.

Arnav Gupta, Albert Liu, Eliza Haines, Riyadh Alghamdi, Aniketh S. Kota, Leslie M. Collins, Boyla O. Mainsah

The current standard for training brain-computer interface (BCI) machine learning models is user-specific. There is a high interest in developing generic models that are trained on data from other users to minimize BCI calibration time; however, this is limited by noisy, non-stationary brain signals and high inter-user variability. We investigate the trade-off between training data quality and quantity on P300 BCI performance in individuals with amyotrophic lateral sclerosis (ALS) with representative traditional machine learning (stepwise linear discriminant analysis, SWLDA) and deep learning (EEGNet) models. Results show that data quality and domain alignment are more critical than dataset size: user-specific models trained on significantly less data outperformed generic models; generic models trained on ALS data outperformed models trained on non-ALS data; block-averaging of features was mostly detrimental to EEGNet but beneficial to SWLDA; and accounting for inter-stimulus interval differences between ALS and non-ALS data had minimal effect. Our findings highlight the importance of individualized model tuning for reliable P300 BCIs.

Yao Gao, Feng Li, Jeroen Van Dessel, Yi Sun, Robin Willaert

What if the next generation of human-computer interaction is not a screen... but a conversation? Large Language Models (LLMs) offer a new paradigm for interacting with computers through text, but they lack shape reasoning capabilities. We introduce Textual Anatomy Encoding (TAE), a workflow that connects LLMs with 3D anatomies. TAE employs clinician-validated semantic annotations and rule-based prompts to achieve deterministic and interpretable landmark localization. The results indicate that TAE enables LLMs to move beyond textual knowledge, achieving an accurate understanding of anatomical localization. This framework opens opportunities for diagnosis, surgical planning, and scalable medical annotation, positioning LLMs as a foundation for next-generation human–computer interaction in healthcare.

Jingchao Gao, Ziqing Lu, Raghu Mudumbai, Xiaodong Wu, Jirong Yi, Myung Cho, Catherine Xu, Hui Xie, Weiyu Xu

In this paper, we study the adversarial robustness of deep neural networks (DNN) for classification against optimal classifiers. We look at the smallest magnitude of possible additive perturbations that can change a classifier's output. We provide a matrix-theoretic explanation of the adversarial fragility of DNNs for classification. In particular, our theoretical results show that the adversarial robustness of a neural network can degrade as the input dimension d increases. Analytically, we show that the adversarial robustness of neural networks can be only 1/√d of the best possible adversarial robustness of optimal classifiers. Our theories match remarkably well with empirical results. The matrix-theoretic explanation aligns with an earlier information-theoretic feature-compression-based explanation for the adversarial fragility of neural networks.

Aline Cambri Fredere, Gabriel De Oliveira Ramos, Luciano Garim Garcia, Mateus da Rocha Simionato, José Manuel Marques Teixeira de Oliveira, Ariane Santos da Silveira

Machine learning applications in reservoir modeling are hindered by the limited availability of well log data, a common challenge in the oil and gas industry. We propose VAEc-tMC, a domain-informed Conditional Variational Autoencoder that generates synthetic well-log data conditioned on rock type. Addressing a critical gap by existing generative models that rely solely on statistical reconstruction, our model embeds geological domain knowledge into the latent space, and optimizes a modified objective with an adaptive Student-t reconstruction loss and a beta-weighted KL regularizer, improving stability under heavy-tailed data. When used for data augmentation, the synthetic samples preserve inter-log dependencies and substantially enhance downstream classification, accuracy 39→63%, F1-score 36→68%, AUC 0.46→0.80 on a held-out well. Beyond the geological context, the proposed approach illustrates a generalizable strategy where domain-aware generative models with adaptive loss functions provide a robust solution for data-efficient learning in scientific domains facing data scarcity, noise, and heavy-tailed distributions.

Giovanni B. Esposito, Daniele Cesarini, Andrea Bartolini

We present a context-aware diffusion model for multivariate time series generation in dynamic and partially observed environments, with applications to data-center computing node's telemetry and beyond. The model integrates pretrained textual embeddings to represent feature semantics, enabling flexible, context-guided generation and improved adaptability to unseen or re-ordered input features. Built on a transformer architecture, it employs both time-wise and feature-wise masking to support missing data during training and inference. We show that the model is robust to permutations with respect to the feature dimension, mantaining stable performance in settings where input configurations vary. Empirical evaluations on HPC sensor data illustrate the model’s versatility across generation and imputation tasks. This work introduces a modular and generalizable framework for time series modeling in complex, high-dimensional systems which can serve as a digital-twin for data-center's compute node telemetry.

Taqiya Ehsan, Shuren Xia, Jorge Ortiz

Embodied agents must reason causally, as correlation-based models fail under intervention and distribution shift. This challenge arises in domains like robotics and cyber-physical systems, where agents balance efficiency and comfort under uncertainty. We introduce POLICYGRID, unifying causal discovery and control by treating each action as both decision and experiment. Leveraging constraint-based search, neural causal models, and language model priors with interventional validation, POLICYGRID yields adaptive, interpretable policies. Across synthetic, real-world, and live deployments, it achieves superior causal recovery (F1 = 0.89) and 2.8× better multi-objective performance than correlation-based baselines, demonstrating safe, generalizable decision-making.

Phuong-Anh Duong, Luong-Ngoc-Anh Vu, Binh Nguyen Thi

We tackle the challenge of setting smart prices and advertising budgets for dairy products sold through three retail channels—General Trade (GT), Modern Trade (MT), and E-commerce (EC)—in markets where data is scarce. Traditional economic models can capture the complex relationships between price, trust, and advertising, but solving these models every time a manager wants to ask a question is slow and impractical. Our solution is to turn a game-theoretic market simulator into training data: we generate 10,000 market scenarios, solve for the best pricing and ad strategies, and train an AI to imitate those decisions. On unseen scenarios, the AI remains accurate (price RMSE ≤ 0.047, ad RMSE ≤ 0.031) and economically sound (Ratio of Means = 1.0010, Regret = 0.40%). To make it easy to use, we add a simple natural-language interface: users can say things like “trust is low, ad cost is high,” and the system returns price and ad suggestions along with confidence ranges. This creates a practical bridge between AI and economic rigor—delivering defensible decisions in seconds, even when data is limited.

Kacper Dobek, Daniel Jankowski, Krzysztof Krawiec

This work explores Liquid Time-Constant Networks (LTCs) and Closed-form Continuous-time Networks (CfCs) for modeling retinal ganglion cell activity in tiger salamanders across three datasets. Compared to a convolutional baseline and an LSTM, both architectures achieved lower MAE, faster convergence, smaller model sizes, and favorable query times, though with slightly lower Pearson correlation. Their efficiency and adaptability make them well suited for scenarios with limited data and frequent retraining, such as edge deployments in vision prosthetics.

Luong Doan, Tien Nguyen, Nhung Duong, Lap Nguyen, Tuan Do

We present Graph Neural ODEs (GNODEs) for modeling tumor microenvironment dynamics with mathematically guaranteed stability and conservation properties. Unlike bulk ODEs that miss spatial heterogeneity or discrete GNNs that inadequately capture continuous biological processes, GNODEs provide continuous-time evolution with explicit adjacency-aware dynamics while maintaining provable trajectory bounds. Our framework ensures: (1) existence and uniqueness of solutions under dynamic graph topology, (2) Lyapunov stability preventing unphysical states like negative cell counts, and (3) exact conservation of biological invariants through architectural constraints. Benchmarking on synthetic tumor data demonstrates that GNODE accurately captures the dynamics of the resistant cell fraction (0.282 predicted vs 0.242 true), whereas graph-free alternatives fail completely (0.000), underscoring the importance of stability-constrained local interactions for modeling emergent resistance.

Ashim Dhor, Smily Bharadwaj

Automated cancer segmentation in Whole Slide Images (WSIs) has been dominated by a paradigm of static pattern recognition, where even advanced methods leveraging Transformers, Multiple Instance Learning, or topology-aware losses remain fundamentally descriptive and correlational. To address this limitation, we reframe WSI segmentation from a descriptive task to one of causal process modeling. We introduce Topo-GraT, a novel framework featuring a Causal Growth Field (CGF) to model tumor invasion dynamics and a Causal Flow Attention (CFA) mechanism that embeds this field as an architectural prior. This causal engine is integrated within an iterative graph refinement loop that uses segmentation uncertainty to dynamically focus computational resources on the most ambiguous tissue regions. Our comprehensive experiments on multiple WSI datasets demonstrate that Topo-GraT establishes a new state-of-the-art, significantly outperforming existing methods and reducing the 95% Hausdorff Distance, a key boundary metric, by over 15%. Crucially, our framework yields the CGF as a rich, interpretable output whose structure correlates with tumor aggressiveness, positioning it as a novel biomarker for downstream prognostic tasks. By shifting the paradigm from static recognition to causal reasoning, Topo-GraT offers a more robust, efficient, and clinically insightful approach, setting a new direction for the causally-aware medical image analysis.

Soumyajit Dev, Malay Bhattacharyya

Dismantling interdependent directed networks to obtain the largest mutually strongly connected component (MSCC) is an NP-hard problem. To address this, we propose a novel method, Disassembling Directed Interdependent Networks (DDIN), by synergizing Reinforcement Learning (RL) and Graph Neural Networks (GNN). We introduce asymmetric GNNs to capture the asymmetry of in/out-degree and multi-relational attention to model directed inter-layer dependencies, integrated with prioritized RL for efficient node selection in large action spaces. Our contributions include (i) a directed GraphSAGE encoder separating in/out aggregations for asymmetry, (ii) multi-relational attention fusing layer semantics, and (iii) sum-tree prioritized n-step Deep Q-Network (DQN) for efficient policy search. DDIN is evaluated on 5 directed multiplexes from biological, social, and economic domains, achieving 16-23% lower AUDC compared to known baseline heuristics.

Adit Desai, Sudipta Roy, Mohna Chakraborty

Vision-Language Models (VLMs) are increasingly deployed across downstream tasks, yet their training data often encode social biases that surface in outputs. Unlike humans, who interpret images through contextual and social cues, VLMs process them through statistical associations, often leading to reasoning that diverges from human reasoning. By analyzing how a VLM reasons, we can understand how inherent biases are perpetuated and can adversely affect downstream performance. To examine this gap, we systematically analyze social biases in five open-source VLMs for an occupation prediction task, on the FairFace dataset. Across 32 occupations and three different prompting styles, we elicit both predictions and reasoning. Our findings show that the biased reasoning patterns systematically underlie intersectional disparities, highlighting the need to align VLM reasoning with human values before downstream deployment.

Ankan Deria, Dwarikanath Mahapatra, Murari Mondal, Sudipta Roy

We present a language-based noise modulation module for diffusion models that improves image color generation under textual guidance. Unlike standard approaches that inject noise uniformly, our method leverages semantic cues from text to selectively control the noise injection process, preserving local details and enhancing color accuracy even when descriptions are ambiguous or incomplete. Applied to language guided image colorization, this targeted modulation leads to more faithful and visually consistent results. The proposed module is lightweight, generalizable, and can be integrated into existing diffusion pipelines, offering a simple yet effective step toward more controllable text-to-image generation.

Alex Davies, Roussel Nzoyem Ngueguin, Nirav Ajmeri, Telmo De Menezes E Silva Filho

We evaluate how well large language model embeddings represent continuous numerical values across different precisions and ranges. Using linear models and principal component analysis on models from major providers, we show that while embeddings can reconstruct numbers with high fidelity (R2 ≥ 0.95), they introduce substantial noise, with principal components explaining less than 40% of embedding variance. Performance degrades with increasing decimal precision and mixed-sign values, revealing fundamental limitations in how these models encode numerical information.

Shrutimoy Das

Dataset distillation methods learn a representative summary of the full dataset such that training on the distilled data is more efficient in terms of time and space. The current state-of-the-art methods exploit the correspondence between infinitely wide neural networks (NNs) and kernel ridge regression to design distillation methods that result in high-quality summaries of the data. In this work, we leverage the correspondence between infinitely wide networks and Gaussian Processes(GPs) for learning a distilled dataset. We investigate the feasibility of using the inducing points method for Gaussian Processes, as a data distillation method. While most of the existing dataset distillation methods are based on loss or gradient matching, our method looks at the function space approximation, facilitated by the NN-GP correspondence. Additionally, using recent theoretical results on GP regression and neural tangent kernels(NTKs), we also provide an upper bound on the size of the distilled data. We demonstrate the utility of inducing points as distilled data on a set of datasets empirically.

Pranshu Danani, Yash Bansal, Parshiv Kapoor

Deep learning is increasingly applied to intraoperative and surgical video analysis to enable real-time workflow recognition, and decision support for improved surgical precision. A key direction is modeling surgical activity as triplets of instrument, action, and target, which provide a richer representation of procedures. However, existing approaches often depend on bounding-box annotations or lack temporal context. We propose TWiST (Temporal Weakly Supervised Triplet detection), a framework that combines weakly supervised instrument localization, temporal attention for triplet prediction, and grounding of triplets with detected instruments. Our experiments show that TWiST outperforms prior weakly supervised baselines.