Papers

Artificial intelligence (AI) is reshaping scientific discovery, evolving from specialized computational tools into autonomous research partners. We position \textit{\textbf{Agentic Science}} as a pivotal stage within the broader \textit{\textbf{AI for Science}} paradigm, where AI systems progress from partial assistance to full scientific agency. Enabled by large language models (LLMs), multimodal systems, and integrated research platforms, agentic AI exhibits capabilities in hypothesis generation, experimental design, execution, analysis, and iterative refinement-behaviors once regarded as uniquely human. This survey offers a \textbf{domain-oriented review} of autonomous scientific discovery across life sciences, chemistry, materials, and physics, synthesizing research progress and advances within each discipline. We unify three previously fragmented perspectives-process-oriented, autonomy-oriented, and mechanism-oriented-through \textbf{a comprehensive framework }that connects foundational capabilities, core processes, and domain-specific realizations. Building on this framework, we (i) trace the evolution of AI for Science, (ii) identify five core capabilities underpinning scientific agency, (iii) model discovery as a dynamic four-stage workflow, (iv) review applications across life sciences, chemistry, materials science, and physics, and (v) synthesize key challenges and future opportunities. This work establishes a domain-oriented synthesis of autonomous scientific discovery and positions Agentic Science as a structured paradigm for advancing AI-driven research.

This paper presents a new system for academic peer review that is more objective, efficient, and community-guided. Our system incorporates author-assisted evaluation (Author-AAE) and community-guided review (CGR) into the peer review of AI conferences. This is in contrast to existing approaches that prioritize alternative systems that only address some of these challenges. Our evaluation uses data from three major AI conferences that used our system and from a survey of reviewers. Their feedback indicates that our system’s reviews are superior to single-LLM-based reviews due to their reduced subjectivity and enhanced quality. The reviewers’ scores for our system’s reviews were significantly higher than for single-LLM-based reviews across multiple metrics: “Reproducibility and Quality” (by 0.427 ± 0.007), “Review Quality” (by 0.265 ± 0.09), and “Alignment between opinion and paper score” (by 0.503 ± 0.090). In addition, we discovered that single-LLM-based reviews are more likely to be rejected by the program committee after author major revisions (on average by 0.182 ± 0.103) and are much more likely to be rejected overall (on average by 0.300 ± 0.124), compared to our system’s reviews. These results suggest that our system performs better in reducing the arbitrary nature of the current peer review system and can serve as an inspiration for the scientific community to explore new review systems.

Training small language models is challenging due to their limited capacity to capture complex patterns and their susceptibility to overfitting. To address these issues, we investigate gradient noise injection as a regularization strategy, building on prior work while introducing a noise schedule that decays exponentially over training. Unlike existing techniques, our method explicitly controls the trade-off between exploration and stability during optimization. We compare the exponential decay schedule with linear and adaptive variants, demonstrating empirically that the exponential schedule yields superior convergence and generalization. Extensive experiments on diverse text corpora, including shakespeare\_char, enwik8, text8, and larger benchmark datasets, show consistent improvements in training dynamics, validation loss, and final performance. We report error bars and statistical significance tests to ensure robustness of the results. Detailed implementation information, including model architectures, hyperparameter settings, dataset sizes, and optimization strategies, is provided to support reproducibility, and we release our code and trained models publicly. Furthermore, we compare gradient noise injection with other regularization methods such as dropout, weight decay, and data augmentation, both in isolation and in combination, revealing complementary effects on training stability and generalization. Finally, we analyze the computational cost of gradient noise injection relative to these baselines, highlighting its practical efficiency in resource-constrained environments. Together, these contributions position gradient noise injection as a theoretically grounded, empirically validated, and computationally practical method for improving the robustness of small language models.

Decentralized stochastic optimization with nonsmooth objectives and only zeroth-order oracle access arises in federated learning and privacy-sensitive applications, yet existing methods suffer from high variance and dimension-dependent complexity. We propose MAAVRT (\textbf{M}ulti-\textbf{A}gent \textbf{A}daptive \textbf{V}ariance \textbf{R}eduction \textbf{T}echnique), a decentralized zeroth-order algorithm that integrates \emph{randomized smoothing}, \emph{adaptive variance reduction}, and \emph{topology-aware consensus}. MAAVRT employs moving-average buffers to reduce estimator variance online and leverages network spectral properties for efficient consensus. Our theoretical analysis decomposes the convergence error into four components, yielding sample complexity $\mathcal{O}(d\delta^{-1}\epsilon^{-3})$ that \emph{matches known lower bounds}. Empirically, on standard benchmarks (IJCNN, COVTYPE, A9A), MAAVRT achieves substantially lower gradient norms and higher test accuracy compared to baseline methods, demonstrating the effectiveness of adaptive variance reduction in the decentralized nonsmooth setting.

Decentralized stochastic optimization with nonsmooth objectives and only zeroth-order oracle access arises in federated learning and privacy-sensitive applications, yet existing methods suffer from high variance and dimension-dependent complexity. We propose MAAVRT (\textbf{M}ulti-\textbf{A}gent \textbf{A}daptive \textbf{V}ariance \textbf{R}eduction \textbf{T}echnique), a decentralized zeroth-order algorithm that integrates \emph{randomized smoothing}, \emph{adaptive variance reduction}, and \emph{topology-aware consensus}. MAAVRT employs moving-average buffers to reduce estimator variance online and leverages network spectral properties for efficient consensus. Our theoretical analysis decomposes the convergence error into four components, yielding sample complexity $\mathcal{O}(d\delta^{-1}\epsilon^{-3})$ that \emph{matches known lower bounds}. Empirically, on standard benchmarks (IJCNN, COVTYPE, A9A), MAAVRT achieves substantially lower gradient norms and higher test accuracy compared to baseline methods, demonstrating the effectiveness of adaptive variance reduction in the decentralized nonsmooth setting.

VISION TRANSFORMERS FOR SEMICONDUCTOR DEFECT DETECTION: A COMPREHENSIVE SURVEY OF AI-DRIVEN IMAGE SEGMENTATION FROM CNNS TO FOUNDATION MODELS (2015-2025)

Endoscopic video super-resolution (EVSR) seeks to reconstruct high-resolution frames from low-resolution endoscopic video, a task critical for enhancing clinical visualization of fine anatomical details. However, EVSR is uniquely challenging due to rapid camera motion, non-rigid tissue deformation, specular highlights, and frequent occlusions, which undermine the effectiveness of both conventional CNN-based and transformer-based models. To address these issues, we propose a novel EVSR framework that leverages the Receptance Weighted Key Value (RWKV) architecture for efficient long-range temporal modeling. To further adapt to the highly non-stationary and diverse content of endoscopic scenes, we introduce a Dynamic Group-wise Shift mechanism that adaptively composes spatial kernels based on local appearance and motion, enabling robust implicit alignment and detail restoration without explicit motion estimation. Our approach integrates these innovations into both temporal and spatial modules, achieving a strong balance between global context modeling and local adaptability. Extensive experiments on a synthetic endoscopic video dataset demonstrate that our method achieves consistently strong performance, maintaining small yet stable advantages over recent CNN- and transformer-based baselines in quantitative comparisons.

Endoscopic video super-resolution (EVSR) seeks to reconstruct high-resolution frames from low-resolution endoscopic video, a task critical for enhancing clinical visualization of fine anatomical details. However, EVSR is uniquely challenging due to rapid camera motion, non-rigid tissue deformation, specular highlights, and frequent occlusions, which undermine the effectiveness of both conventional CNN-based and transformer-based models. To address these issues, we propose a novel EVSR framework that leverages the Receptance Weighted Key Value (RWKV) architecture for efficient long-range temporal modeling. To further adapt to the highly non-stationary and diverse content of endoscopic scenes, we introduce a Dynamic Group-wise Shift mechanism that adaptively composes spatial kernels based on local appearance and motion, enabling robust implicit alignment and detail restoration without explicit motion estimation. Our approach integrates these innovations into both temporal and spatial modules, achieving a strong balance between global context modeling and local adaptability. Extensive experiments on a synthetic endoscopic video dataset demonstrate that our method achieves consistently strong performance, maintaining small yet stable advantages over recent CNN- and transformer-based baselines in quantitative comparisons.

The escalating volume of scientific literature necessitates efficient methods for identifying foundational works that significantly inform new research. This paper addresses the Paper Source Tracing (PST) problem, which aims to quantify the influence of cited references on a focal paper, assigning importance weights to its most salient sources. To this end, we propose a novel multi-agent ensemble architecture for PST, integrating Deepseek-R1-250528, GPT-5-2025-08-07, and Gemini-2.5-pro. Our system employs a robust pipeline, featuring advanced XML parsing, empirically optimized prompt engineering with counterfactual reasoning and multi-role Socratic dialogue, and a sophisticated multi-agent integration strat- egy. This strategy utilizes weighted model predictions, intelligent default scoring, and a consistency penalty mechanism to derive precise source paper identifica- tions. Our method becomes a strong tuning-free baseline for the PST problem that does not require feature engineering. Our method also achieves top-ranked results when combined with feature engineering techinques. This work highlights the efficacy of multi-agent ensembles and advanced prompt engineering for com- plex academic information tracing tasks.

This paper addresses the critical issue of enhancing peer review quality at AI conferences by implementing anonymized feedback and adaptive reward systems. The growing volume of conference submissions and limited reviewer accountability result in inconsistent review quality, bias, and a lack of transparency, posing significant challenges to the integrity of AI research. Our proposed solution involves a dynamic feedback loop that anonymizes and aggregates feedback to minimize biases, coupled with an adaptive reward system to motivate reviewers while preserving the integrity of the review process. Utilizing sentiment analysis, feedback is processed to detect and mitigate potential biases, enhancing the fairness and efficacy of peer reviews. Experiments conducted using a logistic regression model on the Yelp Polarity dataset demonstrate a significant improvement in sentiment classification accuracy, from 54.1\% to 83.4\%, indicating the effectiveness of our anonymized feedback loop. However, the bias detection score of 0.0 across all runs highlights the need for further refinement in bias mitigation. Our method's scalability and adaptability across various conference settings are supported by its successful implementation in sentiment analysis tasks. Overall, this study provides a robust framework for enhancing the accountability and quality of peer reviews, with implications for future research aimed at integrating advanced bias detection and mitigation techniques.

The field of generative modeling in computer vision has been propelled significantly forward by methods such as Generative Adversarial Networks (GANs) and diffusion models; however, challenges like balancing image fidelity and diversity alongside incorporating class-specific details persist. These traditional approaches often exhibit limitations in adaptability and computational efficiency. This paper introduces Reinforced Adaptive Diffusion Networks (RAD-Nets), a novel generative framework that synergizes diffusion processes with reinforcement learning to enhance image synthesis through dynamic parameter optimization. The core innovation lies in integrating a Reinforced Learning Layer and an Adaptive Feedback Mechanism, which employ real-time feedback to iteratively refine outputs. The Multi-Objective Optimization module within RAD-Nets specifically targets the concurrent enhancement of image quality, diversity, and class fidelity, addressing the issues found in static optimization techniques. Empirical evaluations demonstrate that RAD-Nets outperform existing generative models on standard benchmarks like CIFAR-10 and CelebA, achieving superior metrics in quality and diversity without compromising fidelity. By focusing on class-conditional image synthesis, RAD-Nets also demonstrate significant improvements in class-specific feature representation, marking a substantial advancement over conventional generative modeling frameworks.

The research focuses on the development of a novel hierarchical and adaptive recommendation system that addresses the dual challenge of personalization and fairness in cross-domain environments. Traditional recommendation systems have struggled to effectively integrate diverse user interactions and adapt to rapidly evolving user preferences while maintaining fairness. The proposed solution leverages three core innovations: cross-domain collaborative filtering, meta-reinforcement learning, and fairness-aware mechanisms. By synthesizing data from multiple domains, the system constructs enriched user profiles that inform a meta-reinforcement learning framework, enhancing adaptability to user behavior changes. Additionally, fairness-aware mechanisms are incorporated to mitigate biases and ensure equitable content distribution. This integrated approach aims to resolve key challenges in recommendation systems, namely the precise prediction of preferences and the equitable treatment of diverse user groups. Empirical evaluations demonstrate that the proposed methodology not only improves recommendation accuracy but also enhances fairness metrics, thereby fostering a balanced and inclusive recommendation landscape.

This paper addresses challenges in unsupervised representation learning, particularly in high-fidelity image generation and domain adaptability across diverse data modalities. Current frameworks such as GANs and VQ-VAE have shown promise but face limitations in maintaining consistent performance across variable data distributions without significant supervision. To overcome these challenges, we propose a Multi-Modal Vector Quantized Variational AutoEncoder (VQ-VAE) integrated with Self-Supervised Learning (SSL). Our innovative approach incorporates a harmonizer module within the VQ-VAE architecture, which aligns and transforms data representations across multiple modalities. By leveraging self-supervised learning techniques, the model iteratively refines its parameters, enhancing both image reconstruction quality and adaptability to new domains with minimal supervision. The proposed framework processes CIFAR-10 datasets to facilitate structured data integration, employing advanced standardization and batching techniques for optimal performance. Empirical evaluations reveal substantial improvements in image reconstruction fidelity and domain adaptability compared to standard VQ-VAE models, corroborated by metrics such as PSNR, SSIM, and FID. The seamless integration of modality-specific feature extraction and embedding generalization within our framework demonstrates the potential to advance unsupervised learning paradigms. Our contribution establishes a robust solution, optimizing the generative process, and expanding applicability in real-world scenarios characterized by unlabeled, multi-modal datasets.

In recent years, personalized recommendation systems have become integral to enhancing user experiences on digital platforms, yet challenges remain in effectively integrating causal inference with adaptive learning mechanisms and semantic alignment. Traditional systems predominantly rely on correlation-based models, often overlooking the dynamic causal relationships within user interaction data that could enhance recommendation precision and contextual relevance. This paper addresses these gaps by presenting a novel framework that synergizes causal inference using structural equation models and causal diagrams, adaptive learning algorithms via a refined hybrid multi-armed bandit strategy, and semantic content mapping with advanced natural language processing techniques such as Latent Dirichlet Allocation and BERT-based embeddings. Through this integrated approach, our method dynamically adjusts recommendations to align with user preferences and adapt to context changes. Empirical evaluation demonstrates our method's superiority in achieving higher accuracy and relevance in personalized content delivery compared to existing models. The findings underscore the potential of our framework to significantly improve recommendation cohesion and user satisfaction, marking a substantial advancement in the field of contextual personalization.

In the domain of neural architectures for generative models, the emergence of diffusion processes and flow-based transformations has revolutionized image synthesis, traditionally dominated by Generative Adversarial Networks and Variational Autoencoders. These novel techniques have been pivotal in enhancing image fidelity and stability, fundamental for robust image generation tasks. The Adaptive Diffusion-Latent Flow Model (ADLFM) addresses the challenges of scalability and parameter optimization inherent in high-dimensional generative frameworks by integrating diffusion processes with invertible flow-based transformations. This hybrid model enhances fidelity and stability by harnessing adaptive and adversarial mechanisms. ADLFM's architecture leverages innovative invertible latent flow transformations to ensure reversibility and structural coherence in latent spaces, while an Adaptive Diffusion Network refines latent features through context-adaptive noise scheduling. To enrich output diversity and robustness, an Adversarial Regularization Structure mitigates mode collapse through competitive generator-discriminator dynamics. Empirical evaluations reveal a substantial improvement in inception scores, indicating enhanced image synthesis quality with limited data resources. Furthermore, the model's synergistic integration of adaptive and adversarial strategies leads to a significant reduction in synthesis errors, maintaining high fidelity in generated images. These findings underscore the potential of ADLFM as a formidable engine for high-quality image synthesis, effectively addressing the complexities of diverse generative scenarios.

In the evolving landscape of digital platforms, the demand for robust recommendation systems is paramount to manage the deluge of user-generated data. Graph Neural Networks (GNNs) have emerged as a potent strategy in recognizing intricate user-item interactions due to their ability to leverage structural data insights. However, existing GNN-based models often overlook trust dynamics, a critical factor in ensuring recommendation reliability and transparency. Despite recognition of trust's potential to address biases and enhance models' interpretability, its integration with sophisticated network-based techniques remains underexplored. Responding to this gap, we propose the Trust-Enhanced Graph-Based Recommendation Model (GTERM), which seamlessly incorporates trust metrics within the GNN framework. GTERM transforms raw interaction data into a trust-augmented graph, employing graph convolutional and attention mechanisms to emphasize trust-enriched interactions, thereby refining recommendation accuracy and transparency. The proposed model achieves notable improvements over baseline methods, as evidenced in diverse experimental evaluations, demonstrating its capacity to deliver more accurate, trustworthy, and interpretable recommendations. Through the integration of trust factors, GTERM fosters user acceptance and enhances system performance by resolving key challenges related to the lack of interpretability and trustworthiness in traditional GNN-based systems.

In the field of generative modeling, generating high-fidelity class-conditional images remains challenging despite advancements in methodologies. Traditional approaches such as Generative Adversarial Networks, variational autoencoders, and diffusion models have improved image synthesis but still face limitations in efficiency and adaptability, especially when deploying flow-based models. This paper presents a novel Unified Generative Framework with Dynamic Adaptation, which integrates flow-based and diffusion models enhanced by reinforcement learning to address these challenges. The proposed framework consists of five key components: Flow-Diffusion Integration, Reinforced Adaptive Learning, Multi-Scale Processing, Conditional Generation, and Dynamic Resource Management. Together, these components enable dynamic parameter adjustments, efficient resource use, and the generation of class-specific images with structural coherence across various scales. Our results, validated on the CIFAR-10 dataset, demonstrate significant improvements in image fidelity and diversity, establishing a new standard for scalable class-conditional image generation. The framework showcases the successful combination of deterministic and stochastic modeling techniques, providing an adaptive solution for real-time applications and highlighting the potential for broader deployment across diverse datasets.

This paper addresses the challenge of enhancing the equity of welfare resource distribution to tackle long-term unemployment in Germany, where traditional bureaucratic processes are often inefficient and biased. This issue is critical as it significantly affects economic productivity and social stability. The integration of AI into welfare systems presents challenges such as data quality, inherent biases, and policy integration complexity. We propose a machine learning-based framework utilizing fairness-aware algorithms and data augmentation techniques to predict and allocate resources more equitably. Our methodology involves developing a shallow Multi-Layer Perceptron (MLP) model trained on a TF-IDF vectorized dataset, alongside a simulated bureaucratic expansion as a baseline. Experimental results show that our machine learning approach, particularly in its best-performing runs, achieves higher equity, maintaining an Equity Gap Metric of 0.0, while also delivering competitive accuracy. This demonstrates the potential of AI-driven methods to outperform traditional bureaucratic approaches in fairness and efficiency, offering valuable insights for policymakers seeking to optimize resource distribution in public policy.

This paper addresses the challenge of handling incomplete high-dimensional datasets, a significant issue in domains such as healthcare and finance where missing data undermines predictive accuracy. Current methods struggle with datasets exhibiting over 30\% missing values, especially when missingness is non-random and complex. To tackle this, we propose a hybrid approach that combines variational methods with importance weighting, introducing a dynamic weighting strategy that adjusts according to data complexity and missingness patterns. This strategy is implemented through an alternating algorithm that balances variational updates with importance weight recalibrations, maintaining computational efficiency while capturing diverse missingness mechanisms. Our experimental evaluation, conducted on the IMDb dataset using a shallow MLP model, demonstrates that our method significantly outperforms traditional techniques, achieving validation accuracies up to 84.65\% with corresponding F1 scores of 0.8505. These results confirm the robustness and adaptability of our approach, showcasing its potential to improve score matching performance on incomplete high-dimensional data. Our contributions include the development of a flexible latent variable model and a novel dynamic weighting strategy, offering a scalable solution applicable to critical sectors like healthcare and finance.