AIConfigurator: Lightning-Fast Configuration Optimization for Multi-Framework LLM Serving

Optimizing Large Language Model (LLM) inference in production systems is increasingly difficult due to dynamic workloads, stringent latency/throughput targets, and a rapidly expanding configuration space. This complexity spans not only distributed parallelism strategies (tensor/pipeline/expert) but also intricate framework-specific runtime parameters such as those concerning the enablement of CUDA graphs, available KV-cache memory fractions, and maximum token capacity, which drastically impact performance. The diversity of modern inference frameworks (e.g., TRT-LLM, vLLM, SGLang), each employing distinct kernels and execution policies, makes manual tuning both framework-specific and computationally prohibitive. We present AIConfigurator, a unified performance-modeling system that enables rapid, framework-agnostic inference configuration search without requiring GPU-based profiling. AIConfigurator combines (1) a methodology that decomposes inference into analytically modelable primitives - GEMM, attention, communication, and memory operations while capturing framework-specific scheduling dynamics; (2) a calibrated kernel-level performance database for these primitives across a wide range of hardware platforms and popular open-weights models (GPT-OSS, Qwen, DeepSeek, LLama, Mistral); and (3) an abstraction layer that automatically resolves optimal launch parameters for the target backend, seamlessly integrating into production-grade orchestration systems. Evaluation on production LLM serving workloads demonstrates that AIConfigurator identifies superior serving configurations that improve performance by up to 40% for dense models (e.g., Qwen3-32B) and 50% for MoE architectures (e.g., DeepSeek-V3), while completing searches within 30 seconds on average. Enabling the rapid exploration of vast design spaces - from cluster topology down to engine specific flags.

EternalMath: A Living Benchmark of Frontier Mathematics that Evolves with Human Discovery

Current evaluations of mathematical reasoning in large language models (LLMs) are dominated by static benchmarks, either derived from competition-style problems or curated through costly expert effort, resulting in limited coverage of research-level mathematics and rapid performance saturation. We propose a fully automated, theorem-grounded pipeline for evaluating frontier mathematical reasoning, which directly transforms recent peer-reviewed mathematical literature into executable and verifiable reasoning tasks. The pipeline identifies constructive or quantitative results, instantiates them into parameterized problem templates, and generates deterministic solutions through execution-based verification, enabling scalable, reproducible, and continuously updatable evaluation without reliance on large-scale expert authoring. By design, this approach supports temporal extensibility, intrinsic correctness checking, and domain-specific customization across mathematical subfields. Applying this pipeline yields \textbf{EternalMath}, an evolving evaluation suite derived from contemporary research papers. Experiments with state-of-the-art LLMs reveal substantial performance gaps, indicating that mathematical reasoning at the research frontier remains far from saturated and underscoring the need for evaluation methodologies that evolve in step with human mathematical discovery.

Semantic Radio Access Networks: Architecture, State-of-the-Art, and Future Directions

Radio Access Network (RAN) is a bridge between user devices and the core network in mobile communication systems, responsible for the transmission and reception of wireless signals and air interface management. In recent years, Semantic Communication (SemCom) has represented a transformative communication paradigm that prioritizes meaning-level transmission over conventional bit-level delivery, thus providing improved spectrum efficiency, anti-interference ability in complex environments, flexible resource allocation, and enhanced user experience for RAN. However, there is still a lack of comprehensive reviews on the integration of SemCom into RAN. Motivated by this, we systematically explore recent advancements in Semantic RAN (SemRAN). We begin by introducing the fundamentals of RAN and SemCom, identifying the limitations of conventional RAN, and outlining the overall architecture of SemRAN. Subsequently, we review representative techniques of SemRAN across physical layer, data link layer, network layer, and security plane. Furthermore, we envision future services and applications enabled by SemRAN, alongside its current standardization progress. Finally, we conclude by identifying critical research challenges and outlining forward-looking directions to guide subsequent investigations in this burgeoning field.

Based on Data Balancing and Model Improvement for Multi-Label Sentiment Classification Performance Enhancement

Multi-label sentiment classification plays a vital role in natural language processing by detecting multiple emotions within a single text. However, existing datasets like GoEmotions often suffer from severe class imbalance, which hampers model performance, especially for underrepresented emotions. To address this, we constructed a balanced multi-label sentiment dataset by integrating the original GoEmotions data, emotion-labeled samples from Sentiment140 using a RoBERTa-base-GoEmotions model, and manually annotated texts generated by GPT-4 mini. Our data balancing strategy ensured an even distribution across 28 emotion categories. Based on this dataset, we developed an enhanced multi-label classification model that combines pre-trained FastText embeddings, convolutional layers for local feature extraction, bidirectional LSTM for contextual learning, and an attention mechanism to highlight sentiment-relevant words. A sigmoid-activated output layer enables multi-label prediction, and mixed precision training improves computational efficiency. Experimental results demonstrate significant improvements in accuracy, precision, recall, F1-score, and AUC compared to models trained on imbalanced data, highlighting the effectiveness of our approach.

Rethinking KL Regularization in RLHF: From Value Estimation to Gradient Optimization

Reinforcement Learning from Human Feedback (RLHF) leverages a Kullback-Leibler (KL) divergence loss to stabilize training and prevent overfitting. However, in methods such as GRPO, its implementation may be guided by principles from numerical value estimation-a practice that overlooks the term's functional role as an optimization loss. To analyze this issue, we establish a unified framework that connects two seemingly distinct implementation styles: using the mathematical term $k_n$ as a detached coefficient for the policy's score function ('$k_n$ in reward') or as a direct loss function through which gradients are propagated ('$k_n$ as loss'). We show that the latter can always be analyzed via an equivalent gradient coefficient in the former, unifying the two perspectives. Through this framework, we prove that the conventional '$k_1$ in reward' (like in PPO) is the principled loss for Reverse KL (RKL) regularization. We further establish a key finding: under on-policy conditions, the '$k_2$ as loss' formulation is, in fact, gradient-equivalent to '$k_1$ in reward'. This equivalence, first proven in our work, identifies both as the theoretically sound implementations of the RKL objective. In contrast, we show that the recently adopted '$k_3$ as loss' (like in GRPO) is merely a first-order, biased approximation of the principled loss. Furthermore, we argue that common off-policy implementations of '$k_n$ as loss' methods are biased due to neglected importance sampling, and we propose a principled correction. Our findings provide a comprehensive, gradient-based rationale for choosing and correctly implementing KL regularization, paving the way for more robust and effective RLHF systems.

SemSteDiff: Generative Diffusion Model-based Coverless Semantic Steganography Communication

Semantic communication (SemCom), as a novel paradigm for future communication systems, has recently attracted much attention due to its superiority in communication efficiency. However, similar to traditional communication, it also suffers from eavesdropping threats. Intelligent eavesdroppers could launch advanced semantic analysis techniques to infer secret semantic information. Therefore, some researchers have designed Semantic Steganography Communication (SemSteCom) scheme to confuse semantic eavesdroppers. However, the state-of-the-art SemSteCom schemes for image transmission rely on the pre-selected cover image, which limits the universality. To address this issue, we propose a Generative Diffusion Model-based Coverless Semantic Steganography Communication (SemSteDiff) scheme to hide secret images into generated stego images. The semantic related private and public keys enable legitimate receiver to decode secret images correctly while the eavesdropper without completely true key-pairs fail to obtain them. Simulation results demonstrate the effectiveness of the plug-and-play design in different Joint Source-Channel Coding (JSCC) frameworks. The comparison results under different eavesdroppers' threats show that, when Signal-to-Noise Ratio (SNR) = 0 dB, the peak signal-to-noise ratio (PSNR) of the legitimate receiver is 4.14 dB higher than that of the eavesdropper.

Generative Diffusion Models for Wireless Networks: Fundamental, Architecture, and State-of-the-Art

With the rapid development of Generative Artificial Intelligence (GAI) technology, Generative Diffusion Models (GDMs) have shown significant empowerment potential in the field of wireless networks due to advantages, such as noise resistance, training stability, controllability, and multimodal generation. Although there have been multiple studies focusing on GDMs for wireless networks, there is still a lack of comprehensive reviews on their technological evolution. Motivated by this, we systematically explore the application of GDMs in wireless networks. Firstly, starting from mathematical principles, we analyze technical advantages of GDMs and present six representative models. Furthermore, we propose the multi-layer wireless network architecture including sensing layer, transmission layer, application layer, and security plane. We also introduce the core mechanisms of GDM at each of the layers. Subsequently, we conduct a rigorous review on existing GDM-based schemes, with a focus on analyzing their innovative points, the role of GDMs, strengths, and weaknesses. Ultimately, we extract key challenges and provide potential solutions, with the aim of providing directional guidance for future research in this field.

NegVQA: Can Vision Language Models Understand Negation?

Negation is a fundamental linguistic phenomenon that can entirely reverse the meaning of a sentence. As vision language models (VLMs) continue to advance and are deployed in high-stakes applications, assessing their ability to comprehend negation becomes essential. To address this, we introduce NegVQA, a visual question answering (VQA) benchmark consisting of 7,379 two-choice questions covering diverse negation scenarios and image-question distributions. We construct NegVQA by leveraging large language models to generate negated versions of questions from existing VQA datasets. Evaluating 20 state-of-the-art VLMs across seven model families, we find that these models struggle significantly with negation, exhibiting a substantial performance drop compared to their responses to the original questions. Furthermore, we uncover a U-shaped scaling trend, where increasing model size initially degrades performance on NegVQA before leading to improvements. Our benchmark reveals critical gaps in VLMs' negation understanding and offers insights into future VLM development. Project page available at https://yuhui-zh15.github.io/NegVQA/.

Learning a General Model: Folding Clothing with Topological Dynamics

The high degrees of freedom and complex structure of garments present significant challenges for clothing manipulation. In this paper, we propose a general topological dynamics model to fold complex clothing. By utilizing the visible folding structure as the topological skeleton, we design a novel topological graph to represent the clothing state. This topological graph is low-dimensional and applied for complex clothing in various folding states. It indicates the constraints of clothing and enables predictions regarding clothing movement. To extract graphs from self-occlusion, we apply semantic segmentation to analyze the occlusion relationships and decompose the clothing structure. The decomposed structure is then combined with keypoint detection to generate the topological graph. To analyze the behavior of the topological graph, we employ an improved Graph Neural Network (GNN) to learn the general dynamics. The GNN model can predict the deformation of clothing and is employed to calculate the deformation Jacobi matrix for control. Experiments using jackets validate the algorithm's effectiveness to recognize and fold complex clothing with self-occlusion.

RIS-Assisted Joint Sensing and Communications via Fractionally Constrained Fractional Programming

This paper studies an uplink dual-functional sensing and communication system aided by a reconfigurable intelligent surface (RIS), whose reflection pattern is optimally configured to trade-off sensing and communication functionalities. Specifically, the Bayesian Cram\'er-Rao lower bound (BCRLB) for estimating the azimuth angle of a sensing user is minimized while ensuring the signal-to-interference-plus-noise ratio constraints for communication users. We show that this problem can be formulated as a novel fractionally constrained fractional programming (FCFP) problem. To deal with this highly nontrivial problem, we extend a quadratic transform technique, originally proposed to handle optimization problems containing ratio structures only in objectives, to the scenario where the constraints also contain ratio structures. First, we consider the case where the fading coefficient is known. Using the quadratic transform, the FCFP problem is turned into a sequence of subproblems that are convex except for the constant-modulus constraints which can be tackled using a penalty-based method. To further reduce the computational complexity, we leverage the constant-modulus conditions and propose a novel linear transform. This new transform enables the FCFP problem to be turned into a sequence of linear programming (LP) subproblems, which can be solved with linear complexity in the dimension of reflecting elements. Then, we consider the case where the fading coefficient is unknown. A modified BCRLB is used to make the problem more tractable, and the proposed quadratic transform-based algorithm is used to solve the problem. Finally, numerical results unveil nontrivial and effective reflection patterns that the RIS can be configured to generate to facilitate both functionalities.